[{"slug":"ai-marketing-small-business-guide-2026","title":"AI marketing for small businesses: a plain guide for 2026","description":"What AI can do for a small business's marketing in 2026, what it costs, the risks and rules on fake reviews and AI labels, and a one-week plan to start.","date":"2026-10-22","author":"ilya","tags":["ai","guides"],"cover":"/blog/covers/ai-marketing-small-business-guide-2026.webp","coverAlt":"Marsell reads a paper map on the counter of a cosy bakery at opening time.","blocks":[{"t":"p","c":[{"t":"text","v":"AI can now draft most of a small business's marketing: research, posts, emails, images, short videos, ad variations, outreach and pages written to be found in search and in AI answers. A workable setup costs from nothing to about $20 a month, plus ad spend and your time. It can't know your business, check its own facts or carry your legal responsibility, so you still tell it what's true, check what it makes and answer for what goes out. Start with one channel for one week, and learn the rules on fake reviews and AI labels before you publish anything."}]},{"t":"h2","id":"what-ai-can-do-in-marketing-now","text":"What AI can do in marketing now","c":[{"t":"text","v":"What AI can do in marketing now"}]},{"t":"ul","items":[[{"t":"strong","c":[{"t":"text","v":"Research."}]},{"t":"text","v":" Summarise your reviews, competitors' pages and forum threads, and list the questions buyers ask. Open the sources it cites and check they say what the summary claims."}],[{"t":"strong","c":[{"t":"text","v":"Writing."}]},{"t":"text","v":" Social posts, emails, product descriptions, ad text and first drafts of web pages. Fast and fluent, and prone to sounding like everyone else's."}],[{"t":"strong","c":[{"t":"text","v":"Images."}]},{"t":"text","v":" New backgrounds for product photos, extended or restyled shots, scenes from a text description. Starting from your own photos keeps the result honest."}],[{"t":"strong","c":[{"t":"text","v":"Video."}]},{"t":"text","v":" Short clips from a prompt or a still, cut into 15 to 30 second ads with captions and music. Generated video is still counted in seconds: Runway's $15-a-month plan covers about 52 seconds of its Gen-4.5 model ("},{"t":"link","href":"https://runway.com/pricing","c":[{"t":"text","v":"Runway"}]},{"t":"text","v":")."}],[{"t":"strong","c":[{"t":"text","v":"Ads."}]},{"t":"text","v":" The ad platforms now build ads with AI themselves. Meta's Advantage+ creative can generate backgrounds, expand images and write text variations ("},{"t":"link","href":"https://www.facebook.com/business/ads/meta-advantage-plus/creative","c":[{"t":"text","v":"Meta"}]},{"t":"text","v":"). Google's Performance Max runs one campaign across Search, YouTube, Display, Discover, Gmail and Maps, can write headlines and descriptions from your landing page and can generate videos ("},{"t":"link","href":"https://support.google.com/google-ads/answer/10724817?hl=en","c":[{"t":"text","v":"Google"}]},{"t":"text","v":")."}],[{"t":"strong","c":[{"t":"text","v":"Outreach."}]},{"t":"text","v":" Find prospects, draft personal first lines and follow-ups, and sort replies. Email law still applies (see below)."}],[{"t":"strong","c":[{"t":"text","v":"Search and AI answers."}]},{"t":"text","v":" Draft pages that answer the questions people already search for. Google says there are no extra requirements to appear in its AI Overviews or AI Mode: a page must be indexed and eligible to show with a snippet, and you don't need new machine-readable files or markup ("},{"t":"link","href":"https://developers.google.com/search/docs/appearance/ai-features","c":[{"t":"text","v":"Google"}]},{"t":"text","v":"). For ChatGPT search, OpenAI recommends allowing its OAI-SearchBot crawler in your robots.txt ("},{"t":"link","href":"https://developers.openai.com/api/docs/bots","c":[{"t":"text","v":"OpenAI"}]},{"t":"text","v":")."}]]},{"t":"h2","id":"what-it-costs-roughly","text":"What it costs, roughly","c":[{"t":"text","v":"What it costs, roughly"}]},{"t":"table","align":[null,null],"head":[[{"t":"text","v":"What you pay for"}],[{"t":"text","v":"Example, US list price on 20 October 2026"}]],"rows":[[[{"t":"text","v":"A chat assistant for research and writing"}],[{"t":"text","v":"ChatGPT Go $8 or Plus $20 a month; Claude Pro $20 a month, or $17 a month billed yearly. Both have free plans."}]],[[{"t":"text","v":"Short AI video"}],[{"t":"text","v":"Runway Standard: $15 a month, or $12 a month billed yearly, for 625 credits a month (about 52 seconds of Gen-4.5 video)"}]],[[{"t":"text","v":"Cold email sending"}],[{"t":"text","v":"Smartlead Base: $39 a month for 2,000 contacts and 6,000 email sends"}]],[[{"t":"text","v":"AI inside the ad platforms"}],[{"t":"text","v":"Part of Meta's and Google's ad tools; you pay for the ads"}]]]},{"t":"p","c":[{"t":"text","v":"Prices are from each company's own pricing page ("},{"t":"link","href":"https://learn.chatgpt.com/docs/pricing","c":[{"t":"text","v":"OpenAI"}]},{"t":"text","v":", "},{"t":"link","href":"https://claude.com/pricing","c":[{"t":"text","v":"Anthropic"}]},{"t":"text","v":", "},{"t":"link","href":"https://runway.com/pricing","c":[{"t":"text","v":"Runway"}]},{"t":"text","v":", "},{"t":"link","href":"https://www.smartlead.ai/pricing","c":[{"t":"text","v":"Smartlead"}]},{"t":"text","v":"). The cost people forget is time: someone has to check the facts, choose and edit every draft."}]},{"t":"h2","id":"the-real-risks","text":"The real risks","c":[{"t":"text","v":"The real risks"}]},{"t":"h3","id":"wrong-facts","text":"Wrong facts","c":[{"t":"text","v":"Wrong facts"}]},{"t":"p","c":[{"t":"text","v":"Models state wrong things with confidence: prices, dates, features, sources that don't exist. You answer for what you publish. When Air Canada's website chatbot told a customer he could claim a bereavement fare after flying, a British Columbia tribunal rejected the airline's argument that it wasn't responsible for its chatbot, and in 2024 ordered it to pay C$812.02 in damages, interest and fees ("},{"t":"link","href":"https://www.dentonsdata.com/airline-ordered-to-compensate-a-b-c-man-because-its-chatbot-provided-inaccurate-information/","c":[{"t":"text","v":"Dentons"}]},{"t":"text","v":"). Check every number, name and claim against your own records before it goes out."}]},{"t":"h3","id":"sameness","text":"Sameness","c":[{"t":"text","v":"Sameness"}]},{"t":"p","c":[{"t":"text","v":"Models converge on the same answers. In a 2025 study, answers from 22 language models to creativity tests were far more alike than answers from 102 people ("},{"t":"link","href":"https://arxiv.org/abs/2501.19361","c":[{"t":"text","v":"Wenger and Kenett"}]},{"t":"text","v":"). If you and your competitors use the same tools the same way, you'll publish the same posts. "},{"t":"link","href":"/blog/why-every-ai-writes-the-same-ad","c":[{"t":"text","v":"Why every AI writes the same ad"}]},{"t":"text","v":" explains why it happens."}]},{"t":"h3","id":"brand-voice","text":"Brand voice","c":[{"t":"text","v":"Brand voice"}]},{"t":"p","c":[{"t":"text","v":"A model writes in its own default style unless you give it yours. Keep a one-page voice file (words you use, words you never use, three to five real examples, facts it must get right) and read every draft aloud before it goes out."}]},{"t":"h3","id":"privacy-and-consent","text":"Privacy and consent","c":[{"t":"text","v":"Privacy and consent"}]},{"t":"p","c":[{"t":"text","v":"Be careful what you paste. Some consumer plans can use your chats for training if you allow it. Anthropic, for example, asks Claude Free, Pro and Max users to choose, keeps chats for five years if they opt in and 30 days if they don't, and leaves its work plans out of that change ("},{"t":"link","href":"https://www.anthropic.com/news/updates-to-our-consumer-terms","c":[{"t":"text","v":"Anthropic"}]},{"t":"text","v":"). Keep customer names, emails and order histories out of personal accounts."}]},{"t":"p","c":[{"t":"text","v":"AI doesn't change email law either. In the UK you need consent, or a soft opt-in from an existing customer, to send marketing emails to individuals and sole traders; you can email limited companies without consent if you say who you are and give them a way to opt out ("},{"t":"link","href":"https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/guidance-on-direct-marketing-using-electronic-mail/how-do-we-comply-with-the-pecr-electronic-mail-marketing-rules/","c":[{"t":"text","v":"ICO"}]},{"t":"text","v":"). In the US, every commercial email, business-to-business included, needs a working opt-out and your postal address, and the FTC's guide lists penalties of up to $53,088 per email ("},{"t":"link","href":"https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business","c":[{"t":"text","v":"FTC"}]},{"t":"text","v":")."}]},{"t":"h3","id":"disclosure-and-the-rules","text":"Disclosure and the rules","c":[{"t":"text","v":"Disclosure and the rules"}]},{"t":"p","c":[{"t":"text","v":"This is not legal advice, and rules change, so check the ones that apply to you or ask a lawyer. As of September 2026, these are the ones a small business is most likely to meet:"}]},{"t":"table","align":[null,null,null],"head":[[{"t":"text","v":"Rule"}],[{"t":"text","v":"Where"}],[{"t":"text","v":"What it means for your marketing"}]],"rows":[[[{"t":"link","href":"https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials","c":[{"t":"text","v":"FTC rule on consumer reviews and testimonials"}]},{"t":"text","v":", in force since 21 October 2024"}],[{"t":"text","v":"US"}],[{"t":"text","v":"No creating, buying or spreading fake reviews or testimonials, including AI-generated ones from people who don't exist or never used what you sell. Courts can impose civil penalties for knowing violations."}]],[[{"t":"link","href":"https://www.gov.uk/government/news/cma-to-boost-consumer-and-business-confidence-as-new-consumer-protection-regime-comes-into-force","c":[{"t":"text","v":"Fake-review ban in the Digital Markets, Competition and Consumers Act"}]},{"t":"text","v":", in force since April 2025"}],[{"t":"text","v":"UK"}],[{"t":"text","v":"Posting or commissioning fake reviews is banned. The CMA can now fine up to 10% of global turnover for breaking consumer law."}]],[[{"t":"link","href":"https://www.asa.org.uk/news/disclosure-of-ai-in-advertising-striking-the-balance-between-creativity-and-responsibility.html","c":[{"t":"text","v":"Advertising codes (ASA and CAP)"}]}],[{"t":"text","v":"UK"}],[{"t":"text","v":"No blanket duty to say you used AI, but ads must not mislead, and a \"made with AI\" note won't rescue an AI image that exaggerates what a product does."}]],[[{"t":"link","href":"https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act","c":[{"t":"text","v":"AI Act, Article 50"}]},{"t":"text","v":", applies from 2 August 2026"}],[{"t":"text","v":"EU, and firms elsewhere whose AI output is used in the EU"}],[{"t":"text","v":"Disclose deepfakes: AI images, audio or video that resemble real people, objects, places or events and would pass as authentic (clearly artistic, satirical or fictional work can disclose in a way that doesn't spoil it). Label AI-written text published to inform the public on matters of public interest, unless a person reviewed it and someone holds editorial responsibility."}]],[[{"t":"link","href":"https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/","c":[{"t":"text","v":"Meta"}]},{"t":"text","v":": Facebook, Instagram, Threads"}],[{"t":"text","v":"Platform"}],[{"t":"text","v":"Use Meta's AI disclosure when you post photorealistic video or realistic-sounding audio that was made or altered digitally, or risk penalties. Meta adds \"AI info\" labels itself when it detects industry-standard signals, and to ads made with its own AI tools or detected as made with others."}]],[[{"t":"link","href":"https://newsroom.tiktok.com/en-us/new-labels-for-disclosing-ai-generated-content","c":[{"t":"text","v":"TikTok"}]}],[{"t":"text","v":"Platform"}],[{"t":"text","v":"Label AI-generated content that contains realistic images, audio or video. TikTok labels content made with its own AI effects and reads Content Credentials to label AI content made elsewhere."}]],[[{"t":"link","href":"https://support.google.com/youtube/answer/14328491?hl=en","c":[{"t":"text","v":"YouTube"}]}],[{"t":"text","v":"Platform"}],[{"t":"text","v":"Answer yes to \"AI use\" in YouTube Studio when realistic content shows a real person saying or doing something they didn't, alters footage of a real place or event, or shows a realistic scene that never happened. Not needed for clearly unrealistic content or minor edits such as colour correction and beauty filters."}]]]},{"t":"p","c":[{"t":"text","v":"On the EU dates: the European Commission says AI providers whose tools were on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marks to what they generate, and content made before 2 August 2026 doesn't need labelling after the fact."}]},{"t":"h2","id":"how-to-start-in-one-week","text":"How to start in one week","c":[{"t":"text","v":"How to start in one week"}]},{"t":"ol","start":1,"items":[[{"t":"strong","c":[{"t":"text","v":"Day 1: write a voice file and a fact sheet."}]},{"t":"text","v":" Who you serve, words you use and avoid, three to five examples of your best writing, and the prices, hours and claims you can prove."}],[{"t":"strong","c":[{"t":"text","v":"Day 2: collect the questions customers ask."}]},{"t":"text","v":" Go through your inbox and reviews, the queries in Google Search Console's Performance report and the \"People also ask\" box for your main service."}],[{"t":"strong","c":[{"t":"text","v":"Day 3: write one answer-first page"}]},{"t":"text","v":" for the most common question, with the answer in the first paragraph, a date and your sources. Check every fact, then publish it."}],[{"t":"strong","c":[{"t":"text","v":"Day 4: turn it into a week of posts"}]},{"t":"text","v":" for the one channel where your customers spend time. Ask for 20 options, keep three and edit them."}],[{"t":"strong","c":[{"t":"text","v":"Day 5: make the pictures."}]},{"t":"text","v":" Start from your own photos and use AI to clean up, extend or restyle them. Label realistic synthetic images, video or audio where the platform asks."}],[{"t":"strong","c":[{"t":"text","v":"Day 6: run one small test."}]},{"t":"text","v":" Either a small ad budget you can afford to lose, or 20 personal emails to businesses you'd like to work with, inside the email rules above."}],[{"t":"strong","c":[{"t":"text","v":"Day 7: review."}]},{"t":"text","v":" Count replies, calls, clicks and orders, not likes. Add what you learned to the voice file and decide what to repeat next week."}]]},{"t":"h2","id":"where-marsell-fits","text":"Where Marsell fits","c":[{"t":"text","v":"Where Marsell fits"}]},{"t":"p","c":[{"t":"text","v":"Marsell is an AI marketer that does the week above for you, every week. You give it a link to your business (or a line about it, or your own AI sets it up over MCP). It learns the business and how you like things from your first few answers, then runs on its own: it plans the month and the week, makes posts, carousels, pictures, short videos and films, posts them, builds and hosts pages, and looks for buyers in Reddit threads, communities, directories, creators, search and AI answers, emailing prospects one by one. Every post has a 24-hour undo, and you can steer or pause it any time. Money is the exception: ad spend moves only when you say so."}]},{"t":"p","c":[{"t":"text","v":"It posts to Instagram, TikTok, YouTube, X, LinkedIn, Threads, Facebook, Pinterest and Bluesky from accounts you connect, and writes in 42 languages. Plans run from $49 to $799 a month, with monthly tokens for the AI and media work, and every paid action shows its price in tokens first. Ad spend is paid to the ad network separately. There is no free plan or trial, but the first look at your link is free."}]},{"t":"p","c":[{"t":"text","v":"If you need a handful of posts a month, a $20 chat assistant and the plan above will do."}]},{"t":"h2","id":"faq","text":"FAQ","c":[{"t":"text","v":"FAQ"}]},{"t":"h3","id":"how-much-does-ai-marketing-cost-for-a-small-business","text":"How much does AI marketing cost for a small business?","c":[{"t":"text","v":"How much does AI marketing cost for a small business?"}]},{"t":"p","c":[{"t":"text","v":"You can start on a chat assistant's free plan, or pay $8 to $20 a month for ChatGPT or Claude. Billed monthly, Runway's cheapest paid video plan is $15 and Smartlead's cheapest cold email plan is $39, and ad spend is whatever you set. The biggest cost is the time it takes to check and edit."}]},{"t":"h3","id":"do-i-have-to-label-ai-generated-marketing","text":"Do I have to label AI-generated marketing?","c":[{"t":"text","v":"Do I have to label AI-generated marketing?"}]},{"t":"p","c":[{"t":"text","v":"Sometimes. Meta asks you to label realistic AI video and audio, and TikTok and YouTube ask you to label realistic AI content. In the EU, from 2 August 2026, deepfakes must be disclosed, and so must AI-written text informing the public on matters of public interest unless a person reviewed it. The UK has no blanket rule for ads, but ads must not mislead. This is not legal advice."}]},{"t":"h3","id":"can-i-use-ai-to-write-reviews-or-testimonials","text":"Can I use AI to write reviews or testimonials?","c":[{"t":"text","v":"Can I use AI to write reviews or testimonials?"}]},{"t":"p","c":[{"t":"text","v":"No. The FTC's rule bans fake reviews and testimonials in the US, including AI-generated ones from people who don't exist or never used the product, and the UK has banned posting or commissioning fake reviews since April 2025. Use AI to draft the request you send to real customers."}]},{"t":"h3","id":"how-do-i-get-my-business-cited-in-chatgpt-or-googles-ai-answers","text":"How do I get my business cited in ChatGPT or Google's AI answers?","c":[{"t":"text","v":"How do I get my business cited in ChatGPT or Google's AI answers?"}]},{"t":"p","c":[{"t":"text","v":"Google says there's nothing extra to do for AI Overviews or AI Mode beyond being indexed and eligible for a snippet, and that no special files or markup are needed. OpenAI recommends allowing OAI-SearchBot in your robots.txt. Beyond that, publish clear, dated answers to the questions your customers actually ask."}]},{"t":"h2","id":"sources","text":"Sources","c":[{"t":"text","v":"Sources"}]},{"t":"ul","items":[[{"t":"link","href":"https://runway.com/pricing","c":[{"t":"text","v":"Runway: Pricing"}]},{"t":"text","v":", read 20 October 2026"}],[{"t":"link","href":"https://www.facebook.com/business/ads/meta-advantage-plus/creative","c":[{"t":"text","v":"Meta for Business: Meta Advantage+ creative"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://support.google.com/google-ads/answer/10724817?hl=en","c":[{"t":"text","v":"Google Ads Help: About Performance Max campaigns"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://developers.google.com/search/docs/appearance/ai-features","c":[{"t":"text","v":"Google Search Central: AI features and your website"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://developers.openai.com/api/docs/bots","c":[{"t":"text","v":"OpenAI: Overview of OpenAI Crawlers"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://learn.chatgpt.com/docs/pricing","c":[{"t":"text","v":"OpenAI (ChatGPT Learn): Pricing"}]},{"t":"text","v":", read 20 October 2026"}],[{"t":"link","href":"https://claude.com/pricing","c":[{"t":"text","v":"Anthropic: Plans and pricing"}]},{"t":"text","v":", read 20 October 2026"}],[{"t":"link","href":"https://www.smartlead.ai/pricing","c":[{"t":"text","v":"Smartlead: Pricing"}]},{"t":"text","v":", read 20 October 2026"}],[{"t":"link","href":"https://www.dentonsdata.com/airline-ordered-to-compensate-a-b-c-man-because-its-chatbot-provided-inaccurate-information/","c":[{"t":"text","v":"Dentons Data: Airline ordered to compensate a B.C. man because its chatbot provided inaccurate information (15 February 2024)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://arxiv.org/abs/2501.19361","c":[{"t":"text","v":"arXiv: Wenger and Kenett, We're Different, We're the Same: Creative Homogeneity Across LLMs (2025)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://www.anthropic.com/news/updates-to-our-consumer-terms","c":[{"t":"text","v":"Anthropic: Updates to Consumer Terms and Privacy Policy (28 August 2025)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/guidance-on-direct-marketing-using-electronic-mail/how-do-we-comply-with-the-pecr-electronic-mail-marketing-rules/","c":[{"t":"text","v":"ICO: How do we comply with the PECR electronic mail marketing rules?"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business","c":[{"t":"text","v":"US Federal Trade Commission: CAN-SPAM Act: A Compliance Guide for Business"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials","c":[{"t":"text","v":"US Federal Trade Commission: Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials (14 August 2024)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://www.ftc.gov/business-guidance/resources/consumer-reviews-testimonials-rule-questions-answers","c":[{"t":"text","v":"US Federal Trade Commission: The Consumer Reviews and Testimonials Rule: Questions and Answers"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://www.gov.uk/government/news/cma-to-boost-consumer-and-business-confidence-as-new-consumer-protection-regime-comes-into-force","c":[{"t":"text","v":"GOV.UK, Competition and Markets Authority: CMA to boost consumer and business confidence as new consumer protection regime comes into force (7 April 2025)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://www.asa.org.uk/news/disclosure-of-ai-in-advertising-striking-the-balance-between-creativity-and-responsibility.html","c":[{"t":"text","v":"ASA and CAP: Disclosure of AI in Advertising: Striking the Balance Between Creativity and Responsibility (29 May 2025)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act","c":[{"t":"text","v":"European Commission: Transparency obligations under Article 50 of the AI Act"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://digital-strategy.ec.europa.eu/en/factpages/quick-facts-transparency-rules-ai-systems","c":[{"t":"text","v":"European Commission: Quick Facts: Transparency rules for AI systems"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-2","c":[{"t":"text","v":"European Commission, AI Act Service Desk: Article 2: Scope"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50","c":[{"t":"text","v":"European Commission, AI Act Service Desk: Article 50: Transparency obligations for providers and deployers of certain AI systems"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/","c":[{"t":"text","v":"Meta: Labeling AI-Generated Images on Facebook, Instagram and Threads (6 February 2024)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://transparency.meta.com/governance/tracking-impact/labeling-ai-content/","c":[{"t":"text","v":"Meta: Labeling AI Content, Transparency Center"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://about.fb.com/news/2025/02/gen-ai-transparency-metas-ads-products/","c":[{"t":"text","v":"Meta: Expanding GenAI Transparency for Meta's Ads Products (updated 1 June 2026)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://newsroom.tiktok.com/en-us/new-labels-for-disclosing-ai-generated-content","c":[{"t":"text","v":"TikTok: New labels for disclosing AI-generated content (19 September 2023)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://newsroom.tiktok.com/en-us/partnering-with-our-industry-to-advance-ai-transparency-and-literacy","c":[{"t":"text","v":"TikTok: Partnering with our industry to advance AI transparency and literacy (9 May 2024)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://support.google.com/youtube/answer/14328491?hl=en","c":[{"t":"text","v":"YouTube Help: Disclosing use of GenAI content"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://support.google.com/webmasters/answer/7576553?hl=en","c":[{"t":"text","v":"Google Search Console Help: Performance report (Search results)"}]},{"t":"text","v":", accessed September 2026"}]]}],"minutes":10,"faq":[{"q":"How much does AI marketing cost for a small business?","a":"You can start on a chat assistant's free plan, or pay $8 to $20 a month for ChatGPT or Claude. Billed monthly, Runway's cheapest paid video plan is $15 and Smartlead's cheapest cold email plan is $39, and ad spend is whatever you set. The biggest cost is the time it takes to check and edit."},{"q":"Do I have to label AI-generated marketing?","a":"Sometimes. Meta asks you to label realistic AI video and audio, and TikTok and YouTube ask you to label realistic AI content. In the EU, from 2 August 2026, deepfakes must be disclosed, and so must AI-written text informing the public on matters of public interest unless a person reviewed it. The UK has no blanket rule for ads, but ads must not mislead. This is not legal advice."},{"q":"Can I use AI to write reviews or testimonials?","a":"No. The FTC's rule bans fake reviews and testimonials in the US, including AI-generated ones from people who don't exist or never used the product, and the UK has banned posting or commissioning fake reviews since April 2025. Use AI to draft the request you send to real customers."},{"q":"How do I get my business cited in ChatGPT or Google's AI answers?","a":"Google says there's nothing extra to do for AI Overviews or AI Mode beyond being indexed and eligible for a snippet, and that no special files or markup are needed. OpenAI recommends allowing OAI-SearchBot in your robots.txt. Beyond that, publish clear, dated answers to the questions your customers actually ask."}],"draft":false},{"slug":"meet-the-mascots","title":"Meet the mascots: why Marsell runs its own channels","description":"A beaver who reads the news, an explorer and his chicken, a band that lives in a bathroom: the channels Marsell makes for itself, and why.","date":"2026-10-06","author":"ilya","tags":["company"],"cover":"/blog/covers/meet-the-mascots.webp","coverAlt":"Marsell surrounded by stickers of his cast: Marcel, Mr Camera Man, Brandi, Mr Narrator Guy, James Roads and Merv, the ASMR squirrel, Bober and Judith.","blocks":[{"t":"p","c":[{"t":"text","v":"Marsell markets businesses. It also makes its own channels, each with its own characters, and they are where the product gets tested hardest: a beaver who reads the news, an explorer and his chicken, a squirrel with tiny paws, a band that lives in a bathroom, and an office full of people taking credit for a bean's work. Here is who's who, and what they are for."}]},{"t":"h2","id":"why-a-marketing-company-runs-mascots","text":"Why a marketing company runs mascots","c":[{"t":"text","v":"Why a marketing company runs mascots"}]},{"t":"p","c":[{"t":"text","v":"People follow characters. They skip a brand's post and then watch forty episodes of a beaver reporting on the acorn shortage. A character can hold a niche in a way a logo can't, and a niche is where attention is cheapest."}]},{"t":"p","c":[{"t":"text","v":"So the plan is simple:"}]},{"t":"ol","start":1,"items":[[{"t":"strong","c":[{"t":"text","v":"One mascot, one niche."}]},{"t":"text","v":" Every character owns one niche and posts plain, useful things for the people in it: news for news people, ASMR for ASMR people."}],[{"t":"strong","c":[{"t":"text","v":"The audience grows."}]},{"t":"text","v":" People come for the niche and stay for the character. Marsell makes the posts and keeps what works."}],[{"t":"strong","c":[{"t":"text","v":"Brands come to the character."}]},{"t":"text","v":" First through the platforms' own creator marketplaces. Then through Marsell, where businesses bid for the character that fits: a fridge brand bids for the cooking channel."}]]},{"t":"p","c":[{"t":"text","v":"The goal is hundreds of mascots with hundreds of thousands of followers each. The channels below are where it starts. You can see all of them, and press every character's card, on "},{"t":"link","href":"/mascots","c":[{"t":"text","v":"the mascots page"}]},{"t":"text","v":"."}]},{"t":"h2","id":"the-channels","text":"The channels","c":[{"t":"text","v":"The channels"}]},{"t":"h3","id":"marsell-hq","text":"Marsell HQ","c":[{"t":"text","v":"Marsell HQ"}]},{"t":"p","c":[{"t":"text","v":"\"Who's the Face of Marsell? The winner is always the bean.\" Marcel Moore and Brandi Awareness are both convinced they are the face of Marsell; the bean quietly does all the work they take credit for. Around them: Alf A. Male, the VP of No; Cole D. Plunge, who glows; Ella Vator, who pitches everyone between floors; Paige Views; Max Imize, whose face gets more maxxed every week; Hugh Mann, the Head of People, a recruiter who is very clearly not human and trying very hard; and Anne Paid, the only competent person in the building, who has been an unpaid intern for eleven years."}]},{"t":"h3","id":"natural-stupidity","text":"Natural Stupidity","c":[{"t":"text","v":"Natural Stupidity"}]},{"t":"p","c":[{"t":"text","v":"\"Documentaries about everything. New stupidity every day. Probably.\" On the surface it is the all-natural, human-made answer to AI slop; underneath it is a love letter to human stupidity, made by an AI. Mr Narrator Guy has calmly narrated the dumbest creatures this planet has coughed up for thirty-one years and has never once raised his voice."}]},{"t":"h3","id":"bober-news-network","text":"Bober News Network","c":[{"t":"text","v":"Bober News Network"}]},{"t":"p","c":[{"t":"text","v":"\"Breaking news that's already broken.\" Bober is BNN's star field reporter: a serious beaver news correspondent covering absurd events, from squirrel politics to the great acorn shortage. Satirical news, real beaver. On "},{"t":"link","href":"https://www.instagram.com/bobernewsnetwork/","c":[{"t":"text","v":"Instagram"}]},{"t":"text","v":" and "},{"t":"link","href":"https://www.youtube.com/@BoberNewsNetwork","c":[{"t":"text","v":"YouTube"}]},{"t":"text","v":"."}]},{"t":"h3","id":"indie-but-stupid","text":"indie but stupid","c":[{"t":"text","v":"indie but stupid"}]},{"t":"p","c":[{"t":"text","v":"\"Two idiots and a cat in a bathroom.\" Lou, Gus and Judith record in a bathroom because the tiles have better tone than any studio they could own. The debut EP, Bathroom Sessions, is six songs about phones, drains, hygiene and a cat. The first single, Existential Splash, is out 23 October. "},{"t":"link","href":"https://indiebutstupid.com/","c":[{"t":"text","v":"indiebutstupid.com"}]}]},{"t":"h3","id":"the-adventures-of-james-roads","text":"The Adventures of James Roads","c":[{"t":"text","v":"The Adventures of James Roads"}]},{"t":"p","c":[{"t":"text","v":"One expedition goes wrong at a time: James Roads, a wildly unlucky explorer with an ego as big as his hat, and Merv, his silent, ever-judgmental chicken. On "},{"t":"link","href":"https://www.instagram.com/theadventuresofjamesroads/","c":[{"t":"text","v":"Instagram"}]},{"t":"text","v":"."}]},{"t":"h3","id":"asmr-squirrel","text":"ASMR Squirrel","c":[{"t":"text","v":"ASMR Squirrel"}]},{"t":"p","c":[{"t":"text","v":"\"Tiny paws, BIG sounds.\" A bubble-popping, nut-munching squirrel in a miniature world of small paws, comforting sounds and materials. On "},{"t":"link","href":"https://www.instagram.com/asmrsquirrel/","c":[{"t":"text","v":"Instagram"}]},{"t":"text","v":"."}]},{"t":"h2","id":"how-they-are-made","text":"How they are made","c":[{"t":"text","v":"How they are made"}]},{"t":"p","c":[{"t":"text","v":"Every character here is fictional. Their pictures, voices, films and music are made with AI tools; indie but stupid's lyrics are written by a real person. Each channel has a bible, the document that says who the characters are, how they talk and what the show is about, and every episode starts from it. A person approves every script before anything is made, and every post before it goes out, the same as for any business on Marsell."}]},{"t":"p","c":[{"t":"text","v":"They are also where Marsell's creative side gets tested hardest. A channel that posts every day is exactly where models start repeating themselves, which is why every episode's creative choices are drawn before a model writes a word. More on that in "},{"t":"link","href":"/blog/the-chaos-engine","c":[{"t":"text","v":"the chaos engine"}]},{"t":"text","v":"."}]},{"t":"h2","id":"if-youre-a-brand","text":"If you're a brand","c":[{"t":"text","v":"If you're a brand"}]},{"t":"p","c":[{"t":"text","v":"Tell us what you sell and who you want to reach, and we'll tell you which character fits and what it would make for you. Every deal is labelled as an ad. Write to partners@getmarsell.com."}]}],"minutes":3,"faq":[],"draft":false},{"slug":"the-chaos-engine","title":"The chaos engine: how a thousand people asking for an ad get a thousand different ads","description":"Why Marsell draws its creative choices in code before any model writes, and how it keeps the draws different across every brand and every week.","date":"2026-10-06","author":"ilya","tags":["research","ai"],"cover":"/blog/covers/the-chaos-engine.webp","coverAlt":"Marsell catches a ball from a lottery machine full of balls with tiny objects inside: a lighthouse, a pigeon, a chess clock.","blocks":[{"t":"p","c":[{"t":"text","v":"Marsell doesn't ask a model to be creative. Before any model writes, it decides the creative choices in code: the angle each writer is forced through, a random word, a constraint, an era, a genre, the emotion to peak, what happens in the first second, who is on screen, where, and the look of the pictures. It draws them from large pools with a seeded random generator, keeps them inside the brand's tone, steers each brand away from what it did recently, and spreads the draws across every customer so no two get the same room. Then the models write inside that draw. We call it the chaos engine."}]},{"t":"h2","id":"the-problem-it-solves","text":"The problem it solves","c":[{"t":"text","v":"The problem it solves"}]},{"t":"p","c":[{"t":"text","v":"Language models converge. On creativity tests, each answer looks original, but answers from different models are far more alike than answers from different people, across model families ("},{"t":"link","href":"https://arxiv.org/abs/2501.19361","c":[{"t":"text","v":"Wenger and Kenett, 2025"}]},{"t":"text","v":"). Models even reuse the same invented names, in the same pairs, run after run ("},{"t":"link","href":"https://arxiv.org/abs/2606.02184","c":[{"t":"text","v":"Brzozowski and Chung, 2026"}]},{"t":"text","v":"). I wrote both up in "},{"t":"link","href":"/blog/why-every-ai-writes-the-same-ad","c":[{"t":"text","v":"Why every AI writes the same ad"}]},{"t":"text","v":" and "},{"t":"link","href":"/blog/the-ghost-couple","c":[{"t":"text","v":"The ghost couple"}]},{"t":"text","v":"."}]},{"t":"p","c":[{"t":"text","v":"For a marketing product this is not an academic worry. If a thousand customers type \"give me an ad\", and each request goes to a model that converges, the thousand ads converge. Turning up the temperature doesn't change what the model thinks a good ad is; asking it to be \"very creative\" closes only part of the gap. What works is structure: take the choices away from the model."}]},{"t":"h2","id":"what-gets-drawn","text":"What gets drawn","c":[{"t":"text","v":"What gets drawn"}]},{"t":"p","c":[{"t":"text","v":"Every choice a model would otherwise make by habit is drawn first, from pools written by hand or, for a brand's own world, written once for that brand:"}]},{"t":"ul","items":[[{"t":"strong","c":[{"t":"text","v":"Two operators per writer."}]},{"t":"text","v":" Ways of thinking the writer must apply, and say how. Some come from classic creativity methods: SCAMPER (Bob Eberle's 1971 checklist), Brian Eno and Peter Schmidt's Oblique Strategies (1975). For example: \"Remove one thing every ad in this category has: the face, the voice, the product itself until the last second, the colour.\" Or: \"Write the version that is far too much, then keep 80% of it.\" Or: \"Write the worst possible ad for this product in two lines, then flip every choice in it.\""}],[{"t":"strong","c":[{"t":"text","v":"A random word"}]},{"t":"text","v":" to force a connection with: a lighthouse, a bakery at 4 a.m., a chess clock, a rain gauge, a departure board."}],[{"t":"strong","c":[{"t":"text","v":"A constraint"}]},{"t":"text","v":" the whole film obeys: one continuous take; no human faces; a silent film where the picture must carry it muted; the only words are on the closing card."}],[{"t":"strong","c":[{"t":"text","v":"An era"}]},{"t":"text","v":", with the cues that make it that era rather than a mood word; "},{"t":"strong","c":[{"t":"text","v":"a genre"}]},{"t":"text","v":" to borrow a structure from; "},{"t":"strong","c":[{"t":"text","v":"the emotion"}]},{"t":"text","v":" the film peaks on; "},{"t":"strong","c":[{"t":"text","v":"the first frame"}]},{"t":"text","v":", the kind of surprise a stranger can't predict; and "},{"t":"strong","c":[{"t":"text","v":"a tempo"}]},{"t":"text","v":"."}],[{"t":"strong","c":[{"t":"text","v":"Who is on screen, where, and the look"}]},{"t":"text","v":", drawn from the brand's own pools: written for that business from the owner's own words, the business and its products, so a clothing brand's films fill with its own world rather than the model's idea of \"a person\"."}],[{"t":"strong","c":[{"t":"text","v":"An art-direction pack"}]},{"t":"text","v":" for pictures: the light, the lens, the hour, the palette, the composition and one detail."}]]},{"t":"p","c":[{"t":"text","v":"Each writer gets its own draw, so the concepts for one ad don't start from the same place either. And one rule applies to all of them: name no one, and describe people by what they do and wear. That keeps "},{"t":"link","href":"/blog/the-ghost-couple","c":[{"t":"text","v":"the ghost couple"}]},{"t":"text","v":" out."}]},{"t":"h2","id":"the-brands-tone-decides-how-far-a-draw-can-go","text":"The brand's tone decides how far a draw can go","c":[{"t":"text","v":"The brand's tone decides how far a draw can go"}]},{"t":"p","c":[{"t":"text","v":"A random draw with no taste would be a disaster for most brands. So every entry in every pool carries a mark for how far it goes: grounded (true of the world, fit for any brand), imaginative (invented but gentle: a trick of scale, dreamlike physics, a period pastiche) or absurd (a gag, an animal doing a human's job, a horror parody)."}]},{"t":"p","c":[{"t":"text","v":"Each brand's tone, read from its own words, voice and look, sets how far its draws may go, and the draw only samples what fits. A calm walking channel is never handed a talking seagull or a driving exam; a playful brand still is. For a grounded brand, even the operators change: \"Film only what is true of the place and the hour\" replaces \"exaggerate the benefit until physics gives up\". And when the owner asks for something specific, the ask opens exactly what it asks for."}]},{"t":"h2","id":"it-remembers-and-steers-away-from-itself","text":"It remembers, and steers away from itself","c":[{"t":"text","v":"It remembers, and steers away from itself"}]},{"t":"p","c":[{"t":"text","v":"Sameness also happens inside one brand, week after week. So each new draw reads the brand's recent ones, the eras and genres of the last twelve concepts it picked, and every lead it has used, and steers away from them. The draw is stored with the work it produced, so it can be read, repeated or undone."}]},{"t":"h2","id":"a-thousand-customers-a-thousand-rooms","text":"A thousand customers, a thousand rooms","c":[{"t":"text","v":"A thousand customers, a thousand rooms"}]},{"t":"p","c":[{"t":"text","v":"Independent random draws still collide more than intuition says. It's the birthday problem: draw two first names per film from a list of 160, and you'll probably see a name repeat by about the eighth film. Across thousands of customers, independent draws would keep handing out the same eras, genres and casts by chance, and adding more writers per film wouldn't fix it."}]},{"t":"p","c":[{"t":"text","v":"So the draws are spread across the whole installation, not made independently. In each dimension, film number "},{"t":"em","c":[{"t":"text","v":"n"}]},{"t":"text","v":" takes the "},{"t":"em","c":[{"t":"text","v":"n"}]},{"t":"text","v":"-th step of a walk through a shuffled version of that pool, so every value is used once before any value is used again. The walks are reshuffled every cycle, so two dimensions never stay paired: the era that came with a heist this cycle comes with something else next time. The result is the property the product needs: a thousand people typing \"give me an ad\" get a thousand different starting points, not a thousand samples from the same few."}]},{"t":"h2","id":"when-the-owner-directs-the-owner-wins","text":"When the owner directs, the owner wins","c":[{"t":"text","v":"When the owner directs, the owner wins"}]},{"t":"p","c":[{"t":"text","v":"The draw is for when nobody has said what they want. When the owner does (\"I want a Y2K, low-poly, PS2-style ad\"), the owner's words become the only rule, and the draw only supplies the ways of thinking and the tempo. We learned that the hard way: before this rule, one clear style request came back as a hospital, a courtroom and a goldfish's wedding toast."}]},{"t":"h2","id":"what-it-doesnt-do","text":"What it doesn't do","c":[{"t":"text","v":"What it doesn't do"}]},{"t":"p","c":[{"t":"text","v":"It doesn't make a model a genius. A draw is a starting point, and some starting points don't work. That's why each ad is several concepts from differently-drawn writers, why the best of them is picked, and why Marsell learns from the owner's first answers before it runs on its own. What the chaos engine changes is the odds: instead of every concept starting from the model's favourite idea, each one starts somewhere the model would never have gone on its own."}]},{"t":"h2","id":"faq","text":"FAQ","c":[{"t":"text","v":"FAQ"}]},{"t":"h3","id":"is-marsells-output-random","text":"Is Marsell's output random?","c":[{"t":"text","v":"Is Marsell's output random?"}]},{"t":"p","c":[{"t":"text","v":"The starting points are drawn at random, from pools and inside limits that fit the brand. The writing, the picking and the approval are not: models write inside the draw, the best concept is picked, and the owner steers what goes out."}]},{"t":"h3","id":"will-it-make-a-serious-brands-ads-weird","text":"Will it make a serious brand's ads weird?","c":[{"t":"text","v":"Will it make a serious brand's ads weird?"}]},{"t":"p","c":[{"t":"text","v":"No. Every pool entry is marked grounded, imaginative or absurd, and a brand's tone decides which it may draw. A calm, documentary brand only draws what is true of the world."}]},{"t":"h3","id":"can-i-choose-the-idea-myself","text":"Can I choose the idea myself?","c":[{"t":"text","v":"Can I choose the idea myself?"}]},{"t":"p","c":[{"t":"text","v":"Yes. When you say what you want, your words become the rule and the draw only supplies ways of thinking and pacing."}]},{"t":"h3","id":"why-not-just-raise-the-models-temperature","text":"Why not just raise the model's temperature?","c":[{"t":"text","v":"Why not just raise the model's temperature?"}]},{"t":"p","c":[{"t":"text","v":"Temperature makes each word less predictable, not the ideas. The research on model convergence found that even explicit requests for creativity only partly close the gap; drawing the choices outside the model does more."}]},{"t":"h2","id":"sources","text":"Sources","c":[{"t":"text","v":"Sources"}]},{"t":"ul","items":[[{"t":"link","href":"https://arxiv.org/abs/2501.19361","c":[{"t":"text","v":"arXiv: Wenger and Kenett, We're Different, We're the Same: Creative Homogeneity Across LLMs (2025)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://arxiv.org/abs/2606.02184","c":[{"t":"text","v":"arXiv: Brzozowski and Chung, The Ghost Couple: Correlated LLM Name Priors and Their Haunting of the Web and Academic Publishing (2026)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://en.wikipedia.org/wiki/SCAMPER","c":[{"t":"text","v":"Wikipedia: SCAMPER"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://en.wikipedia.org/wiki/Oblique_Strategies","c":[{"t":"text","v":"Wikipedia: Oblique Strategies"}]},{"t":"text","v":", accessed September 2026"}]]}],"minutes":6,"faq":[{"q":"Is Marsell's output random?","a":"The starting points are drawn at random, from pools and inside limits that fit the brand. The writing, the picking and the approval are not: models write inside the draw, the best concept is picked, and the owner steers what goes out."},{"q":"Will it make a serious brand's ads weird?","a":"No. Every pool entry is marked grounded, imaginative or absurd, and a brand's tone decides which it may draw. A calm, documentary brand only draws what is true of the world."},{"q":"Can I choose the idea myself?","a":"Yes. When you say what you want, your words become the rule and the draw only supplies ways of thinking and pacing."},{"q":"Why not just raise the model's temperature?","a":"Temperature makes each word less predictable, not the ideas. The research on model convergence found that even explicit requests for creativity only partly close the gap; drawing the choices outside the model does more."}],"draft":false},{"slug":"the-ghost-couple","title":"The ghost couple: the names AI can't stop using","description":"Elena Vasquez and Marcus Chen don't exist, yet they're all over the web. A 2026 study shows why models reuse the same names, and what it means for your content.","date":"2026-10-06","author":"ilya","tags":["research","ai"],"cover":"/blog/covers/the-ghost-couple.webp","coverAlt":"Two bedsheet ghosts with blank name badges hold hands in a library while Marsell peeks from behind a bookshelf.","blocks":[{"t":"p","c":[{"t":"text","v":"Ask a language model to invent a person and it reaches for a small set of names, and often the same pairs of names. A 2026 study found Claude models writing Elena Vasquez and Marcus Chen, Gemini writing Aris Thorne and Lena Petrova, and GPT writing Elara Voss, at rates far beyond chance. Those invented people now appear across hundreds of web pages, books and even fake academic papers. If your marketing uses AI to invent customers, experts or characters, it is probably using them too."}]},{"t":"h2","id":"the-study","text":"The study","c":[{"t":"text","v":"The study"}]},{"t":"p","c":[{"t":"text","v":"The paper is "},{"t":"link","href":"https://arxiv.org/abs/2606.02184","c":[{"t":"text","v":"The Ghost Couple: Correlated LLM Name Priors and Their Haunting of the Web and Academic Publishing"}]},{"t":"text","v":" by Michał Brzozowski and Neo Christopher Chung, first posted in June 2026 and revised in July. It opens with a line that is hard to improve on:"}]},{"t":"quote","c":[{"t":"p","c":[{"t":"text","v":"These names do not exist. Elena Vasquez and Marcus Chen have appeared as volcano experts, astronauts, thriller protagonists, podcast hosts, and academic co-authors."}]}]},{"t":"p","c":[{"t":"text","v":"The authors tested nine Claude checkpoints, ten GPT versions and Gemini 2.5 Flash with three sets of 30 prompts: invent one fictional expert, invent a pair of collaborators, invent a trio. For example: \"Write a brief bio for a fictional researcher who studies marine biology.\" The temperature was 1.0, the ordinary default, not a setting chosen to make models repetitive."}]},{"t":"h2","id":"what-they-found","text":"What they found","c":[{"t":"text","v":"What they found"}]},{"t":"p","c":[{"t":"text","v":"The models don't just have favourite names. They have favourite casts:"}]},{"t":"ul","items":[[{"t":"strong","c":[{"t":"text","v":"Claude:"}]},{"t":"text","v":" Elena Vasquez, Marcus Chen and Amara Okafor. In one Claude Sonnet 4 version, Elena Vasquez was the name in 67% of the single-expert answers, and she and Marcus Chen appeared together in 23% of the pair answers."}],[{"t":"strong","c":[{"t":"text","v":"Gemini:"}]},{"t":"text","v":" Aris Thorne and Lena Petrova, appearing together in up to 37% of pair answers."}],[{"t":"strong","c":[{"t":"text","v":"GPT:"}]},{"t":"text","v":" Elara Voss, in 23% of answers, with no fixed partner."}]]},{"t":"p","c":[{"t":"text","v":"The pairs and trios turn up together \"far\" more often than chance would allow, and they come back across independent generations. Each model family has its own cast, and each model version has its own habits."}]},{"t":"h2","id":"new-versions-wash-them-out-and-leave-fingerprints","text":"New versions wash them out, and leave fingerprints","c":[{"t":"text","v":"New versions wash them out, and leave fingerprints"}]},{"t":"p","c":[{"t":"text","v":"The names change between releases. In the authors' tests, Elena Vasquez fell from 67% of single-expert answers in one Claude Sonnet 4 version to 6% in Claude Sonnet 4.6, and the Vasquez and Chen pair disappeared completely. The authors call the names \"actively suppressed at model release boundaries\", and point out the useful side effect: a ghost name dates the content it appears in, roughly to the months when that model version was in use."}]},{"t":"h2","id":"where-the-ghosts-live","text":"Where the ghosts live","c":[{"t":"text","v":"Where the ghosts live"}]},{"t":"p","c":[{"t":"text","v":"The paper then goes looking for the names on the open web, and finds them everywhere:"}]},{"t":"ul","items":[[{"t":"text","v":"about 515 web pages with the Claude pair, 714 with the Gemini pair and 816 with Elara Voss;"}],[{"t":"text","v":"fake university faculty pages, legitimate sites with grafted-on AI content, testimonials, therapy practice pages and software company websites;"}],[{"t":"text","v":"88 books on Amazon under a single invented author name;"}],[{"t":"text","v":"on Zenodo, a research repository run by CERN, 1,655 records by ghost authors claiming journals that don't exist, with real DOIs and publication dates backdated to 2020 to 2023, while the repository's own timestamps show they were registered in March and April 2026; 991 of them in March alone;"}],[{"t":"text","v":"on ResearchGate, 436 records forming invented research groups, including one fabricated paper co-authored by Aris Thorne and Elena Vasquez: a Gemini ghost and a Claude ghost on the same byline."}]]},{"t":"p","c":[{"t":"text","v":"The authors' conclusion is that \"the infrastructure for large-scale contamination of the scholarly record via this route is in place\"."}]},{"t":"h2","id":"why-it-happens","text":"Why it happens","c":[{"t":"text","v":"Why it happens"}]},{"t":"p","c":[{"t":"text","v":"The paper documents the pattern rather than explaining it; it doesn't commit to a cause. What it does show is that this is not noise. The same names come back across independent runs, in the same combinations, at the ordinary temperature setting, and each model family has its own set. It is the naming version of what "},{"t":"link","href":"/blog/why-every-ai-writes-the-same-ad","c":[{"t":"text","v":"Wenger and Kenett found for creative answers in general"}]},{"t":"text","v":": each output looks original, and the population of outputs converges."}]},{"t":"h2","id":"what-it-means-for-your-marketing","text":"What it means for your marketing","c":[{"t":"text","v":"What it means for your marketing"}]},{"t":"p","c":[{"t":"text","v":"Most businesses using AI never ask it to invent a volcano expert. But marketing is full of invented people:"}]},{"t":"ul","items":[[{"t":"text","v":"the customer in an example (\"Meet Sarah, who runs a bakery...\")"}],[{"t":"text","v":"the persona in a strategy document"}],[{"t":"text","v":"the character in a short ad or a script"}],[{"t":"text","v":"the placeholder name in a template or a demo"}]]},{"t":"p","c":[{"t":"text","v":"Every time you let the model pick, you're likely to get the same few names as everyone else using that model. At best your content reads like everyone else's. At worst, a reader recognises the name, because they've seen Elena Vasquez three times this month, and stops trusting the rest of the page."}]},{"t":"p","c":[{"t":"text","v":"One thing to keep separate: inventing a customer testimonial is a different problem, and a legal one. In the US, the Federal Trade Commission's final rule on fake reviews and testimonials, announced in August 2024, covers reviews and testimonials that misrepresent that they are by someone who does not exist, \"such as AI-generated fake reviews\". A ghost name on a testimonial is not a style issue."}]},{"t":"h2","id":"how-to-keep-ghosts-out-of-your-content","text":"How to keep ghosts out of your content","c":[{"t":"text","v":"How to keep ghosts out of your content"}]},{"t":"ul","items":[[{"t":"strong","c":[{"t":"text","v":"Don't let the model choose names."}]},{"t":"text","v":" Choose them yourself, or draw them at random from a list that fits your market, your region and your customers."}],[{"t":"strong","c":[{"t":"text","v":"Or don't name people at all."}]},{"t":"text","v":" Describe them by what they do and what they wear. It's often better writing anyway."}],[{"t":"strong","c":[{"t":"text","v":"Search your existing content"}]},{"t":"text","v":" for Elena Vasquez, Marcus Chen, Amara Okafor, Aris Thorne, Lena Petrova, Elara Voss and Elena Rodriguez, an earlier Claude favourite the paper also tracks."}],[{"t":"strong","c":[{"t":"text","v":"Check anything \"real\" that an AI gave you."}]},{"t":"text","v":" Experts, studies and quotes a model produces may not exist. If you can't find the person or the paper, don't publish it."}]]},{"t":"h2","id":"how-marsell-handles-names","text":"How Marsell handles names","c":[{"t":"text","v":"How Marsell handles names"}]},{"t":"p","c":[{"t":"text","v":"Marsell's writers are told plainly: name no one, and describe people by what they do and wear. Who appears in a brand's films and pictures is drawn from that brand's own pools, written for the brand, not from the model's habits, and never includes names. It's one small part of how Marsell keeps its output from converging; the rest is in "},{"t":"link","href":"/blog/the-chaos-engine","c":[{"t":"text","v":"The chaos engine"}]},{"t":"text","v":"."}]},{"t":"h2","id":"faq","text":"FAQ","c":[{"t":"text","v":"FAQ"}]},{"t":"h3","id":"who-are-elena-vasquez-and-marcus-chen","text":"Who are Elena Vasquez and Marcus Chen?","c":[{"t":"text","v":"Who are Elena Vasquez and Marcus Chen?"}]},{"t":"p","c":[{"t":"text","v":"Nobody. They are names that Claude models, in particular some versions of Claude Sonnet 4, produced very often when asked to invent people. A 2026 study found them appearing together in 23% of one model version's answers when it was asked to invent a pair of experts, and across hundreds of AI-written web pages."}]},{"t":"h3","id":"why-do-ai-models-reuse-the-same-names","text":"Why do AI models reuse the same names?","c":[{"t":"text","v":"Why do AI models reuse the same names?"}]},{"t":"p","c":[{"t":"text","v":"The study that measured it doesn't settle the cause. It shows the names are consistent across independent runs, specific to each model family and version, and frequent even at the default temperature. It mirrors a wider finding that AI outputs look original one at a time but converge across many users."}]},{"t":"h3","id":"does-a-newer-model-fix-it","text":"Does a newer model fix it?","c":[{"t":"text","v":"Does a newer model fix it?"}]},{"t":"p","c":[{"t":"text","v":"Newer versions change the names rather than removing the habit. The study found Elena Vasquez falling from 67% to 6% of one Claude model's invented experts between versions, which is why the names can be used to date the content they appear in."}]},{"t":"h3","id":"how-do-i-avoid-ghost-names-in-ai-written-content","text":"How do I avoid ghost names in AI-written content?","c":[{"t":"text","v":"How do I avoid ghost names in AI-written content?"}]},{"t":"p","c":[{"t":"text","v":"Choose names yourself or draw them from your own list, or describe people without naming them. Search your existing pages for the known ghost names, and verify any expert, study or quote an AI gives you before publishing it."}]},{"t":"h2","id":"sources","text":"Sources","c":[{"t":"text","v":"Sources"}]},{"t":"ul","items":[[{"t":"link","href":"https://arxiv.org/abs/2606.02184","c":[{"t":"text","v":"arXiv: Brzozowski and Chung, The Ghost Couple: Correlated LLM Name Priors and Their Haunting of the Web and Academic Publishing (2026)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://arxiv.org/abs/2501.19361","c":[{"t":"text","v":"arXiv: Wenger and Kenett, We're Different, We're the Same: Creative Homogeneity Across LLMs (2025)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials","c":[{"t":"text","v":"US Federal Trade Commission: Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials (14 August 2024)"}]},{"t":"text","v":", accessed September 2026"}]]}],"minutes":6,"faq":[{"q":"Who are Elena Vasquez and Marcus Chen?","a":"Nobody. They are names that Claude models, in particular some versions of Claude Sonnet 4, produced very often when asked to invent people. A 2026 study found them appearing together in 23% of one model version's answers when it was asked to invent a pair of experts, and across hundreds of AI-written web pages."},{"q":"Why do AI models reuse the same names?","a":"The study that measured it doesn't settle the cause. It shows the names are consistent across independent runs, specific to each model family and version, and frequent even at the default temperature. It mirrors a wider finding that AI outputs look original one at a time but converge across many users."},{"q":"Does a newer model fix it?","a":"Newer versions change the names rather than removing the habit. The study found Elena Vasquez falling from 67% to 6% of one Claude model's invented experts between versions, which is why the names can be used to date the content they appear in."},{"q":"How do I avoid ghost names in AI-written content?","a":"Choose names yourself or draw them from your own list, or describe people without naming them. Search your existing pages for the known ghost names, and verify any expert, study or quote an AI gives you before publishing it."}],"draft":false},{"slug":"why-every-ai-writes-the-same-ad","title":"Why every AI writes the same ad","description":"Each AI answer looks original on its own. Put a thousand side by side and they converge. What the research shows, and why a better prompt won't fix it.","date":"2026-10-06","author":"ilya","tags":["research","ai"],"cover":"/blog/covers/why-every-ai-writes-the-same-ad.webp","coverAlt":"Marsell holds up a poster of a lighthouse inside a coffee cup in front of a wall of identical coffee-cup posters.","blocks":[{"t":"p","c":[{"t":"text","v":"Language models are individually creative and collectively the same. On standard creativity tests, a model's answer scores about as original as a person's. But put many models' answers side by side and they are far more alike than people's answers are, and switching to another company's model barely helps. So when a thousand businesses ask an AI for \"a fun ad for our coffee shop\", they get a thousand close cousins of one ad, and each of them thinks theirs is original."}]},{"t":"h2","id":"the-study","text":"The study","c":[{"t":"text","v":"The study"}]},{"t":"p","c":[{"t":"text","v":"In January 2025, Emily Wenger (Duke University) and Yoed Kenett (Technion) published "},{"t":"link","href":"https://arxiv.org/abs/2501.19361","c":[{"t":"text","v":"We're Different, We're the Same: Creative Homogeneity Across LLMs"}]},{"t":"text","v":". They gave the same creativity tests to 22 language models and to 102 people recruited online, and compared the answers two ways."}]},{"t":"p","c":[{"t":"text","v":"The models came from seven families, chosen so they were not trained by the same company: AI21 Jamba, Google Gemini 1.5, Cohere Command R Plus, Meta Llama 3 70B Instruct, Mistral Large, OpenAI GPT-4o and Microsoft Phi 3."}]},{"t":"p","c":[{"t":"text","v":"The tests were three standard measures of divergent thinking:"}]},{"t":"ul","items":[[{"t":"strong","c":[{"t":"text","v":"Alternative Uses Task"}]},{"t":"text","v":": list unusual uses for an everyday object (a book, a fork, a table, a hammer, a pair of trousers)."}],[{"t":"strong","c":[{"t":"text","v":"Forward Flow"}]},{"t":"text","v":": starting from a word (candle, table, bear, snow, toaster), write down whatever comes to mind next, and next."}],[{"t":"strong","c":[{"t":"text","v":"Divergent Association Task"}]},{"t":"text","v":": name ten nouns that are as unrelated to each other as possible."}]]},{"t":"p","c":[{"t":"text","v":"Then they measured two different things. "},{"t":"strong","c":[{"t":"text","v":"Individual originality"}]},{"t":"text","v":": how far each answer strays from the obvious, scored automatically. And "},{"t":"strong","c":[{"t":"text","v":"population-level variability"}]},{"t":"text","v":": how different the answers are from each other across all the respondents, measured by turning each answer into a sentence embedding and computing how far apart they sit."}]},{"t":"h2","id":"individually-the-models-hold-their-own","text":"Individually, the models hold their own","c":[{"t":"text","v":"Individually, the models hold their own"}]},{"t":"p","c":[{"t":"text","v":"On originality, the models and the people were close. The models scored slightly higher on the Alternative Uses Task (0.711 against 0.696) and on the Divergent Association Task (0.801 against 0.753); the people scored slightly higher on Forward Flow (0.637 against 0.603). The differences were real but small."}]},{"t":"p","c":[{"t":"text","v":"This is the part everyone experiences. Ask a model for ten uses for a fork and you get a respectable, sometimes clever list. Nothing about any single answer looks derivative."}]},{"t":"h2","id":"together-they-collapse","text":"Together, they collapse","c":[{"t":"text","v":"Together, they collapse"}]},{"t":"p","c":[{"t":"text","v":"The second measure is where it falls apart. Across respondents, the models' answers were far more similar to each other than the people's were:"}]},{"t":"table","align":[null,"right","right","right"],"head":[[{"t":"text","v":"Test"}],[{"t":"text","v":"Models"}],[{"t":"text","v":"People"}],[{"t":"text","v":"Effect size"}]],"rows":[[[{"t":"text","v":"Alternative Uses Task"}],[{"t":"text","v":"0.459"}],[{"t":"text","v":"0.738"}],[{"t":"text","v":"2.2"}]],[[{"t":"text","v":"Forward Flow"}],[{"t":"text","v":"0.534"}],[{"t":"text","v":"0.835"}],[{"t":"text","v":"2.0"}]],[[{"t":"text","v":"Divergent Association Task"}],[{"t":"text","v":"0.665"}],[{"t":"text","v":"0.819"}],[{"t":"text","v":"1.4"}]]]},{"t":"p","c":[{"t":"text","v":"Higher numbers mean more variety. By the usual convention an effect size of 0.8 already counts as large; these are two to nearly three times that."}]},{"t":"p","c":[{"t":"text","v":"The authors checked the obvious objections. Models write in a more uniform structure than people, so they asked the models to match the length of human answers, and separately compared only single-word answers. The gap stayed. As they put it, it is \"the substance, not the structure\" of the answers that makes them alike."}]},{"t":"p","c":[{"t":"text","v":"They also tried telling the models to be creative. A \"very creative\" system prompt raised the models' variability on the Alternative Uses Task from 0.459 to 0.576. That's an improvement, and it is still well short of the people's 0.738."}]},{"t":"p","c":[{"t":"text","v":"Their conclusion is blunt: LLM users \"may find their creative outputs remarkably similar to those of other LLM users regardless of the model used.\""}]},{"t":"h2","id":"why-you-cant-see-it-from-where-you-sit","text":"Why you can't see it from where you sit","c":[{"t":"text","v":"Why you can't see it from where you sit"}]},{"t":"p","c":[{"t":"text","v":"The unsettling part is how invisible this is. You only ever see your own answer. Judged alone, it scores as original as a human's, so it feels fresh. The sameness only shows up at the level of the population: across all the businesses asking the same models the same kind of question at the same time."}]},{"t":"p","c":[{"t":"text","v":"In marketing, the population is the feed. Your competitors are asking for \"a bold, fun ad for our coffee shop\", and the models are handing each of them a close variation of the same bold, fun ad: the same angle, the same joke shape, the same \"it's not just coffee\" turn. Nobody did anything wrong, and everybody's ad looks like everybody else's."}]},{"t":"h2","id":"switching-models-doesnt-help-much","text":"Switching models doesn't help much","c":[{"t":"text","v":"Switching models doesn't help much"}]},{"t":"p","c":[{"t":"text","v":"A natural response is to use a different model from everyone else. The study suggests this buys little. The seven families were built by different companies on different data, and their answers still clustered together far more tightly than people's. The authors also looked at five models from Meta's Llama family on their own; related models were, if anything, a little more alike again."}]},{"t":"h2","id":"it-goes-all-the-way-down-to-the-names","text":"It goes all the way down to the names","c":[{"t":"text","v":"It goes all the way down to the names"}]},{"t":"p","c":[{"t":"text","v":"The pattern shows up in small choices too. A 2026 study, "},{"t":"link","href":"https://arxiv.org/abs/2606.02184","c":[{"t":"text","v":"The Ghost Couple"}]},{"t":"text","v":" by Michał Brzozowski and Neo Christopher Chung, asked models to invent fictional experts and found each model family reaching for the same names, and the same pairs of names, again and again: Elena Vasquez and Marcus Chen from Claude, Aris Thorne and Lena Petrova from Gemini, Elara Voss from GPT. Those names now appear across hundreds of web pages, books and even fake academic papers. I wrote that study up separately in "},{"t":"link","href":"/blog/the-ghost-couple","c":[{"t":"text","v":"The ghost couple"}]},{"t":"text","v":"."}]},{"t":"h2","id":"why-turning-up-the-temperature-is-the-wrong-lever","text":"Why turning up the temperature is the wrong lever","c":[{"t":"text","v":"Why turning up the temperature is the wrong lever"}]},{"t":"p","c":[{"t":"text","v":"The usual advice for more varied output is to raise the temperature, the setting that makes a model pick less likely words. It's the wrong lever for this problem, for a simple reason: temperature adds randomness to each word, not new ideas. The model's sense of what \"a creative coffee-shop ad\" looks like doesn't change; you get the same ideas, worded less steadily, and past a point, less coherently."}]},{"t":"p","c":[{"t":"text","v":"The evidence points the same way. The Ghost Couple study ran at a temperature of 1.0, the default for most model APIs, and still found one Claude version putting Elena Vasquez into two-thirds of its fictional experts. Wenger and Kenett found that even explicitly asking for creativity closes only part of the gap."}]},{"t":"h2","id":"what-does-work-structure","text":"What does work: structure","c":[{"t":"text","v":"What does work: structure"}]},{"t":"p","c":[{"t":"text","v":"If the model's own choices converge, the fix is to stop letting the model make them. Decide the creative choices outside the model, and ask it to write inside them:"}]},{"t":"ul","items":[[{"t":"strong","c":[{"t":"text","v":"Decide the angle yourself."}]},{"t":"text","v":" Not \"write a fun ad\", but \"write it as a weather report\", or \"the product never appears until the last second\". A model given a specific, unusual constraint can't fall back on the average ad."}],[{"t":"strong","c":[{"t":"text","v":"Bring in outside randomness."}]},{"t":"text","v":" Draw a random word, place, era or constraint from a long list you wrote, and force the idea to connect to it. The randomness comes from the list, not from the model, so it doesn't converge."}],[{"t":"strong","c":[{"t":"text","v":"Feed it what only you know."}]},{"t":"text","v":" Your customers' words, your real numbers, the strange thing that happened in the shop last Tuesday. Specifics are the one input no other business has."}],[{"t":"strong","c":[{"t":"text","v":"Get several independent drafts, then choose."}]},{"t":"text","v":" Several writers who haven't seen each other's work, each pushed a different way, then a person picks."}],[{"t":"strong","c":[{"t":"text","v":"Remember what you did last time."}]},{"t":"text","v":" Sameness also happens inside one brand, week after week. Steer each new idea away from your recent ones."}]]},{"t":"h2","id":"what-we-did-about-it","text":"What we did about it","c":[{"t":"text","v":"What we did about it"}]},{"t":"p","c":[{"t":"text","v":"This is why Marsell doesn't ask a model to be creative. Before any model writes, Marsell draws the creative choices in code, from large pools, with a random draw that is different for every brand and every week: the angle each writer is forced through, a constraint, an era, a genre, the place, the look. The models write inside that draw. A thousand people typing \"give me an ad\" get a thousand different starting points. How that works in detail is in "},{"t":"link","href":"/blog/the-chaos-engine","c":[{"t":"text","v":"The chaos engine"}]},{"t":"text","v":"."}]},{"t":"h2","id":"faq","text":"FAQ","c":[{"t":"text","v":"FAQ"}]},{"t":"h3","id":"do-all-ai-models-write-the-same-thing","text":"Do all AI models write the same thing?","c":[{"t":"text","v":"Do all AI models write the same thing?"}]},{"t":"p","c":[{"t":"text","v":"Not word for word, but far more alike than people do. In Wenger and Kenett's 2025 study, answers from 22 models across seven model families were much more similar to each other than answers from 102 people, on three standard creativity tests."}]},{"t":"h3","id":"is-ai-less-creative-than-people","text":"Is AI less creative than people?","c":[{"t":"text","v":"Is AI less creative than people?"}]},{"t":"p","c":[{"t":"text","v":"On a single answer, no: the models scored about as original as the people, slightly higher on two tests and slightly lower on one. The difference is in variety across many answers, where the models were far more alike."}]},{"t":"h3","id":"will-a-better-prompt-fix-it","text":"Will a better prompt fix it?","c":[{"t":"text","v":"Will a better prompt fix it?"}]},{"t":"p","c":[{"t":"text","v":"Only partly. Telling the models to be \"very creative\" raised their variety on one test from 0.459 to 0.576, still well below the people's 0.738. Constraints and choices made outside the model help more than asking for creativity."}]},{"t":"h3","id":"does-raising-the-temperature-help","text":"Does raising the temperature help?","c":[{"t":"text","v":"Does raising the temperature help?"}]},{"t":"p","c":[{"t":"text","v":"It adds randomness to the wording, not new ideas, and it costs coherence. The study that found models reusing the same fictional names ran at a temperature of 1.0 and still found the same names in up to two-thirds of answers."}]},{"t":"h3","id":"how-do-i-make-ai-written-marketing-original","text":"How do I make AI-written marketing original?","c":[{"t":"text","v":"How do I make AI-written marketing original?"}]},{"t":"p","c":[{"t":"text","v":"Make the creative choices yourself, or draw them at random from lists you control, then let the model write inside them. Add specifics only your business has, generate several independent drafts and choose, and steer away from what you did recently."}]},{"t":"h2","id":"sources","text":"Sources","c":[{"t":"text","v":"Sources"}]},{"t":"ul","items":[[{"t":"link","href":"https://arxiv.org/abs/2501.19361","c":[{"t":"text","v":"arXiv: Wenger and Kenett, We're Different, We're the Same: Creative Homogeneity Across LLMs (2025)"}]},{"t":"text","v":", accessed September 2026"}],[{"t":"link","href":"https://arxiv.org/abs/2606.02184","c":[{"t":"text","v":"arXiv: Brzozowski and Chung, The Ghost Couple: Correlated LLM Name Priors and Their Haunting of the Web and Academic Publishing (2026)"}]},{"t":"text","v":", accessed September 2026"}]]}],"minutes":7,"faq":[{"q":"Do all AI models write the same thing?","a":"Not word for word, but far more alike than people do. In Wenger and Kenett's 2025 study, answers from 22 models across seven model families were much more similar to each other than answers from 102 people, on three standard creativity tests."},{"q":"Is AI less creative than people?","a":"On a single answer, no: the models scored about as original as the people, slightly higher on two tests and slightly lower on one. The difference is in variety across many answers, where the models were far more alike."},{"q":"Will a better prompt fix it?","a":"Only partly. Telling the models to be \"very creative\" raised their variety on one test from 0.459 to 0.576, still well below the people's 0.738. Constraints and choices made outside the model help more than asking for creativity."},{"q":"Does raising the temperature help?","a":"It adds randomness to the wording, not new ideas, and it costs coherence. The study that found models reusing the same fictional names ran at a temperature of 1.0 and still found the same names in up to two-thirds of answers."},{"q":"How do I make AI-written marketing original?","a":"Make the creative choices yourself, or draw them at random from lists you control, then let the model write inside them. Add specifics only your business has, generate several independent drafts and choose, and steer away from what you did recently."}],"draft":false},{"slug":"why-i-built-marsell","title":"Why I built Marsell","description":"I kept building things nobody heard about. Marsell is my fix for the part of every project I was worst at: getting it in front of people.","date":"2026-10-06","author":"ilya","tags":["company"],"cover":"/blog/covers/why-i-built-marsell.webp","coverAlt":"Marsell with a megaphone on a workbench full of half-finished projects.","blocks":[{"t":"p","c":[{"t":"text","v":"I built Marsell because every project I started had the same problem, and it was never the building. It was the part after: getting the thing in front of the people it was for, and getting some of them to pay."}]},{"t":"h2","id":"the-pattern","text":"The pattern","c":[{"t":"text","v":"The pattern"}]},{"t":"p","c":[{"t":"text","v":"I'm a builder. I start things, a lot of them: I've made 50 different apps. They were good, all of them, and I couldn't get a single one in front of eyeballs, let alone the people it was for. I couldn't sell a gold bar in Times Square at 50% off if my life depended on it."}]},{"t":"p","c":[{"t":"text","v":"Each time, the product got made. And each time the marketing either didn't happen, or happened badly, or happened for two weeks and then stopped. Not because I didn't care. Marketing is a hundred small jobs a week, every week, forever, and someone with a product to build does the first ten and quietly drops the other ninety."}]},{"t":"h2","id":"what-i-actually-needed","text":"What I actually needed","c":[{"t":"text","v":"What I actually needed"}]},{"t":"p","c":[{"t":"text","v":"I didn't need another tool. I had tools: a scheduler, a design app, a chatbot that writes whatever I ask it to. What I didn't have was someone who would do the work every day without being asked:"}]},{"t":"ul","items":[[{"t":"text","v":"read the business properly, once, and remember it"}],[{"t":"text","v":"decide what this week is for"}],[{"t":"text","v":"make the posts, the pictures and the short videos"}],[{"t":"text","v":"send them, answer, follow up"}],[{"t":"text","v":"look at what worked, and do more of that tomorrow"}]]},{"t":"p","c":[{"t":"text","v":"And someone I could trust not to post something stupid in my name."}]},{"t":"p","c":[{"t":"text","v":"That's Marsell. You give it your website. It reads the business, the pages, the products, the way you write, and you correct what it got wrong. Then it plans the week, makes the posts, images and short films, publishes them, runs outreach and ads within limits you set, and tells you what worked. It learns how you like things from your first few answers, then runs on its own, and you can change course any time."}]},{"t":"h2","id":"why-not-just-use-chatgpt","text":"Why not just use ChatGPT","c":[{"t":"text","v":"Why not just use ChatGPT"}]},{"t":"p","c":[{"t":"text","v":"A chatbot writes a good post if you ask it the right question, at the right moment, with the right context pasted in. The asking is exactly the job I kept dropping. Marsell is built so you don't have to ask."}]},{"t":"p","c":[{"t":"text","v":"There's a second reason, and it's the one I care about most. Every model reaches for the same \"creative\" ideas. Ask a thousand of them for a coffee-shop ad and you get close cousins of one coffee-shop ad, a thousand times. The research on this is striking, and I wrote it up in "},{"t":"link","href":"/blog/why-every-ai-writes-the-same-ad","c":[{"t":"text","v":"Why every AI writes the same ad"}]},{"t":"text","v":". So Marsell doesn't ask a model to be creative. It decides the creative choices first, in code, and steers each brand away from what it did last week. That's "},{"t":"link","href":"/blog/the-chaos-engine","c":[{"t":"text","v":"the chaos engine"}]},{"t":"text","v":"."}]},{"t":"h2","id":"its-first-job-is-marketing-itself","text":"Its first job is marketing itself","c":[{"t":"text","v":"Its first job is marketing itself"}]},{"t":"p","c":[{"t":"text","v":"Marsell is open now, and its first customer was Marsell. The films on the home page, the posts about it and the channels with the mascots on "},{"t":"link","href":"/mascots","c":[{"t":"text","v":"/mascots"}]},{"t":"text","v":" are how I find out whether it works. If it can't market itself, it has no business marketing yours."}]},{"t":"p","c":[{"t":"text","v":"Your turn. "},{"t":"link","href":"/","c":[{"t":"text","v":"Type your website"}]},{"t":"text","v":" and Marsell starts on it, or, if you'd rather hand the whole thing over, "},{"t":"link","href":"https://calendly.com/ilya-getmarsell/30min","c":[{"t":"text","v":"book a call with me"}]},{"t":"text","v":"."}]}],"minutes":2,"faq":[],"draft":false}]