Still loading. On a slow connection this can take a minute. Reload

Sign in

Marsell markets and sells for you. On autopilot.

Type your website. Marsell learns your business and how you like things, plans the month, makes the posts, films and pages, and finds the people who will buy. Then it runs on its own.

No website? Any link to your work is fine, or sign up with just your email.

Sign-up takes a minute. Your marketing, off your plate.

The Marsell blog

10 marketing jobs to hand to AI in 2026, and 3 to keep

Ten marketing jobs AI does well in 2026, how to hand each one over and what to check before it goes out, plus three jobs to keep for yourself.

By Ilya Arbabi, 24 October 2026

Hand AI the marketing jobs that are repetitive, easy to check and cheap to redo: research, first drafts, repurposing, image variations, captions and alt text, reporting, finding the questions customers ask, outreach research, ad variations and scheduling. Keep three for yourself: positioning, relationships and the final call on taste. The test for any job is whether you can check the result faster than you could do it. If you can, hand it over. If checking it takes the same judgement as doing it, keep it.

What the evidence says

Many small businesses already use AI for marketing. In Constant Contact's survey of 2,500 small businesses in the US, UK, Canada and Australia, published in September 2025, 48% used AI in their marketing, most often to write email and social copy. Across all firms the numbers are lower: the US Census Bureau found that 17% to 20% of US businesses used AI in any business function between December 2025 and May 2026, and the ONS found that 28% of UK businesses with fewer than ten employees used at least one AI technology in June 2026.

The gains are real and uneven. In an MIT experiment published in Science, 453 professionals, marketers among them, took 40% less time on writing tasks with ChatGPT, and their work was rated 18% better. In a Boston Consulting Group study with 758 of its consultants, GPT-4 lifted performance on a creative product task by 40% but cut it by 23% on a business problem-solving task, and the ideas of those using it were 41% less varied. A study of 5,179 customer support agents found the same shape: 14% more issues resolved per hour on average, 34% for novices and little change for the most experienced.

AI is strong at some jobs and confidently wrong at others that look similar. The list below sorts marketing work by which side it falls on.

10 jobs to hand over

1. Research on competitors and customers

AI reads faster than you do. Give it your closest competitors' websites and public reviews, plus your own customer emails, and ask for a table: who each competitor targets, what it promises, what it charges and what its customers complain about.

How to hand it over: name the sources, ask for a link beside every claim, and ask it to mark anything it inferred rather than read. Strip customers' personal details before you paste their emails in.

Check before it goes out: open the links. Prices and features change, and a model fills gaps with plausible guesses. The BCG problem-solving result is the warning here: treat the table as research and decide what it means yourself.

2. First drafts

Blog posts, product pages, emails and video scripts. This is where the MIT result applies: you edit a draft instead of facing a blank page.

How to hand it over: a short brief with the reader, the one point, the facts to use and a sample of your own writing. List what to avoid, such as the phrases your competitors use and claims you cannot prove.

Check before it goes out: every fact, number and promise; whether it sounds like you; and whether it adds something only you know. Google's guidance says generative AI can help you research a topic and structure original content, but using it to generate many pages that add nothing for readers may breach its spam policy on scaled content abuse.

3. Repurposing one piece into many

One good piece of thinking can become a LinkedIn post, an email, three short captions and a 30-second video script. AI reshapes well because the ideas are already yours.

How to hand it over: start from your best material (a talk, a long post, a customer story), name each destination and give its rules: length, tone, whether links work there.

Check before it goes out: that the meaning has not drifted and quotes are still accurate, that each version suits its platform rather than being the same text cut shorter, and that one phrase is not repeated across all six.

4. Image variations

New backgrounds for a product shot, seasonal versions of an ad, the same scene in square, vertical and wide formats.

How to hand it over: reference images, your colours and fonts, and a list of what must not change: the product's shape, the logo, the text on the label.

Check before it goes out: that the product looks like the real one, since a buyer who receives something different will say so; that there is no garbled text; and the rules where it will appear. TikTok requires creators to label realistic AI-generated content. Meta requires a disclosure for photorealistic video or realistic audio that was digitally created or altered, and may penalise accounts that skip it. Google Merchant Center requires AI-generated product images to carry IPTC metadata that marks them as made by AI.

5. Captions and alt text

Alt text for 200 product photos is dull, necessary work, exactly the kind to hand over. Social captions work the same way: first versions in bulk, edited by you.

How to hand it over: send each image with its context: the page it sits on, where it links and what the reader needs from it.

Check before it goes out: that the alt text does the image's job rather than describing pixels. The W3C's alt text decision tree says a purely decorative image gets empty alt text, an image inside a link or button should say where it goes or what it does, and an image of text should contain that text. Google lists alt text among the metadata it expects to be accurate when it is generated automatically.

6. Reporting

AI is good at turning exports into a readable summary: what went up, what went down, and which post or email did the work.

How to hand it over: the raw exports from your analytics, ads and email tools, the date range and the one question you care about, such as which channel brought customers rather than clicks.

Check before it goes out: that the totals match the dashboards they came from, and treat every "because" with suspicion. A model will offer a confident cause for any change, and two weeks of data rarely proves one. Ask for one next step you can test, then decide yourself whether it is worth doing.

7. Finding the questions customers ask

Customers tell you what they want to know in searches, forum threads and support emails. AI can sort thousands of them into a short list worth answering.

How to hand it over: export the queries from your Google Search Console Performance report, which shows the searches that found you with their clicks, impressions, click-through rate and position. Add People Also Ask questions, Reddit and Quora threads and your support inbox, then ask for clusters ranked by how often each comes up.

Check before it goes out: that the questions use your customers' words, that each has enough demand to deserve a page, and that you can answer it honestly and specifically.

8. Outreach research and first lines

The slow part of cold outreach is research: one true, specific thing about each prospect. AI can read a prospect's site and recent posts and suggest an opening line that shows you looked.

How to hand it over: a list of prospects with their websites, what you sell and why it might matter to them, and a request for one observation per prospect with the link it came from.

Check before it goes out: that the observation is true, recent and not creepy. Then the law. In the US, CAN-SPAM covers business-to-business email too: accurate headers and subject lines, a clear sign that the message is an ad, your postal address and an opt-out honoured within 10 business days, with penalties of up to $53,088 per email. In the UK, the ICO says you can email companies without consent, but sole traders and some partnerships count as individuals, who generally need to have consented, and every message must say who you are and how to opt out.

9. Ad variations to test

Testing needs variety, and variety is cheap for a model. A Google responsive search ad, for example, takes up to 15 headlines of 30 characters and 4 descriptions of 90, and Google assembles the combinations itself.

How to hand it over: one offer, one audience, the proof you can back up, and a request for variations that differ in angle (price, speed, outcome, risk) rather than in wording.

Check before it goes out: that every claim can be substantiated, the character limits are respected and the variants really differ. Google also advises most advertisers against pinning headlines to fixed positions, because pinning cuts the combinations it can try.

10. Scheduling

Once the week's posts exist, choosing slots, handling time zones and queuing them is clerical work.

How to hand it over: set the rules first, such as no more than one post a day per channel and no posts on the days you're closed, then let the tool place posts and fill gaps within them.

Check before it goes out: that links and tracking tags work, that nothing clashes with a launch or a holiday, and that you can pause everything at once. When there is bad news in your industry or your town, a cheerful scheduled post is the mistake automation makes most publicly.

3 jobs to keep

Positioning and strategy

Positioning is a decision about who you are for and who you are not for. AI can list options and argue against yours, but it draws on what most businesses already say, and in the BCG study the ideas of people using GPT-4 were 41% less varied. A position anyone could take is not a position. Use AI to pressure-test your strategy, then choose it yourself.

Relationships and community

Customers, partners, journalists and your local community respond to a person. Answer complaints yourself, go to the events, send the thank-you notes. And never let a tool speak for your customers: the FTC's 2024 rule bans fake reviews and testimonials, including AI-generated ones attributed to people who do not exist.

The final call on taste

Taste is what makes your marketing look like yours: knowing when a line is almost right, when an image is on-brand but off-message, and when to say nothing. It is also where the responsibility sits. The FTC's CAN-SPAM guide says you "can't contract away your legal responsibility" when someone else sends email for you, and the US Copyright Office says prompts alone do not make you the author of what a model produces, though your creative arrangement and changes can be protected. Whatever writes it, what goes out under your name is yours: set the rules it works within, and look at what it did.

Where Marsell fits

Marsell, an AI marketer, is built around that split. You give it a link to your business. 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, pictures, short videos and films, posts them, finds the questions buyers ask in Reddit threads, search and AI answers, emails prospects one by one and reports what worked. Every post has a 24-hour undo, and you can steer or pause it any time. Ad spend moves only when you say so.

FAQ

Which marketing tasks should I not hand to AI?

Positioning, relationships and the final call on taste. They need judgement about your market and they carry your reputation. AI can help you think them through, but it should not decide them.

Does AI-written content hurt SEO?

It depends on the pages, not the tool. Google's guidance says generative AI can be useful for research and for structuring original content, and asks for accuracy, quality and relevance. Using it to generate many pages without adding value for readers may breach its spam policy on scaled content abuse.

Do I have to label AI-generated images and videos?

It depends on the platform. TikTok requires labels on realistic AI-generated content, Meta requires a disclosure for photorealistic video and realistic audio, and Google Merchant Center requires metadata on AI-generated product images. Check each platform's current rules before you post.

How many small businesses use AI for marketing?

In Constant Contact's 2025 survey of 2,500 small businesses, 48% used AI in their marketing, and LocaliQ's report for 2025 put the figure at nearly 60%. Across all US businesses, the Census Bureau found that 17% to 20% used AI in any function between December 2025 and May 2026.

How much time does AI save on marketing writing?

In the MIT experiment published in Science, professionals took 40% less time on writing tasks with ChatGPT, and their work was rated 18% better. What you save depends on how much checking the job needs.

Sources