AI for marketing: practical uses, prompts and limits

What is AI actually useful for in marketing?

AI is useful for the parts of marketing work that turn known material into new shapes: a brief into a first draft, a long article into social posts, a pile of customer comments into themes, one headline into ten alternatives. In each case you supply the substance and the tool does the rearranging, which is why the output is only as good as what you give it.

It is much weaker at the parts that depend on knowing your market. It does not know which promise your customers actually believe, what your sales team keeps hearing, or why last quarter's campaign underperformed. If you ask it to supply those things, it will produce something plausible and generic. Treat that as a prompt for your own thinking, not as an answer.

What should you give an AI tool before asking for marketing copy?

Give it what you would give a freelance copywriter on their first day: the audience, the single action you want, the facts it may use, the facts it may not use, the channel limits and an example of your voice. A one-line request such as 'write a launch email for our new subscription' forces the tool to guess every one of those, and the guesses are where invented claims come from.

The most important line in a marketing prompt is often the restriction: 'Use only the facts below. Do not add awards, customer numbers, rankings, testimonials or deadlines.' Without it, many tools fill gaps with the kind of proof that marketing copy usually contains, whether or not it is true for you.

What does a reviewed AI marketing draft look like?

Take a fictional roastery, Copperleaf Coffee, launching a monthly subscription. The supplied facts are: two 250g bags a month, roasted within the week before dispatch, free UK delivery, pause or cancel online at any time, launching on 3 November. The audience is people who have already bought a single bag from the website.

A weak prompt ('write an exciting email about our new coffee subscription') produced a draft that opened with 'Join thousands of coffee lovers who rate us the best roaster in the North', promised 'the freshest coffee you have ever tasted' and closed with 'Only 50 subscriptions available, sign up before midnight'. None of that was in the brief. There is no customer count, no ranking, no supply limit and no deadline.

The reviewed prompt pasted the facts, named the audience as existing single-bag buyers, set a 120-word limit and added the restriction line. The draft that came back was plainer: it reminded the reader of their last order, explained the two-bag delivery and the roasting window, and made pausing and cancelling easy to find. The marketer then made two edits of their own. They replaced 'freshly roasted' with the specific roasting window, because the specific version is checkable, and they cut a sentence implying the subscription was cheaper than buying single bags, because nobody had confirmed the price comparison.

The useful lesson is not that the second draft was perfect. It is that the person reviewing it could point to the line in the brief behind every sentence, and could explain each of their edits.

How do you keep a consistent brand voice with AI?

Show the tool your voice instead of describing it. Adjectives such as 'friendly but professional' mean different things to every reader and to every model. Two or three short passages you are happy with, plus a short list of words you never use, give it something concrete to match.

Then ask it to check rather than only to write. A useful second prompt is: 'Compare this draft with the examples above. List any sentence that sounds unlike them and say why.' You still decide, but the tool can surface drift you might read past. If your team uses Notion AI, Canva AI or a similar built-in feature, keep the same examples in a shared document so everyone works from one reference, and confirm with whoever administers the workspace what the feature is allowed to read.

Which AI tools do marketers use, and does the choice matter?

Most marketing teams use a mix of general assistants, such as ChatGPT, Claude, Gemini or Copilot, and AI features built into tools they already pay for, such as design, writing, email and analytics platforms. The general assistants are flexible for drafting and thinking; the built-in features are convenient because they sit where the work already lives.

For most day-to-day copy work, the choice of tool matters less than the habit of supplying a proper brief and checking the result. What does matter is the account type and its settings. A personal account, a business workspace and an API connection can have different terms on retention and use of your content. Check the terms of the specific plan your company has approved before pasting in unreleased campaign material or customer data.

What are the rules on AI-generated ads, reviews and claims?

Advertising rules apply to the message, not to how it was made. The UK Committee of Advertising Practice said in May 2025 that its existing rules apply regardless of how content is generated, edited or targeted, that there is no blanket UK legal requirement to disclose AI use in ads, and that a disclosure is very unlikely to fix a fundamentally misleading message. Its suggested test is whether the audience is likely to be misled if AI use is not disclosed.

In the United States, the Federal Trade Commission announced a final rule in August 2024 banning fake consumer reviews and testimonials, and it specifically covers reviews that claim to come from someone who does not exist, such as AI-generated fake reviews. Asking a tool to 'write five customer reviews' for a product page is therefore not a harmless shortcut.

This is general information, not legal advice. Rules differ by country and sector, and regulated products such as financial services, health and alcohol carry extra requirements. If a campaign is consequential, involve whoever normally signs off claims.

How should a marketing team check AI output before it goes out?

Check the draft against the brief line by line, and treat anything you cannot trace as unsupported until someone confirms it. The checks below cover most of what goes wrong with AI-assisted marketing copy.

Where is AI not the right answer in marketing?

AI is a poor substitute for talking to customers. It can summarise what customers said, but it cannot tell you what they would have said if you had asked a better question. It is also the wrong place to settle positioning, pricing or a response to a complaint that is becoming public, because those depend on judgement, context and accountability that sit with people.

It is also not worth using when the job is small and you already know what to write. A two-line reply to a partner rarely improves by going through a prompt, a draft and a review. Use it where the rearranging is genuine work, and write directly where it is not.

How can a marketing team practise this?

Pick one recurring, low-risk task, such as a newsletter item built from an approved blog post, and have everyone do it with the same fictional brief. Compare the drafts, then compare the edits. The edits are where the skill shows: who spotted the invented proof, who tightened the call to action, who kept the voice.

Tosheo.ai is being built to run this kind of practice as a planned marketing path: a short diagnostic, 10 to 15 minute lessons with worked examples from marketing work, and hands-on tasks graded against a visible rubric. Managers are planned to see skill progress and completion, not the chats themselves. The product is pre-build, so treat this as intended scope rather than a description of a live service.

Sources and scope

Committee of Advertising Practice (ASA), Disclosure of AI in advertising: striking the balance between creativity and responsibility, 29 May 2025 ↗

US Federal Trade Commission, Final rule banning fake reviews and testimonials, 14 August 2024 ↗