AI in marketing without making things up: our 6 rules

AI is part of almost everything we do at Atelier ATTENTION, from research to first drafts to checking websites before launch. It saves us hours every week.

It also, sometimes, makes things up. Confidently, fluently and in perfect grammar.

For a marketing agency, that is not a small problem. A made-up statistic in a client's article damages trust that took years to build. So we work with a few simple rules. Here they are, so you can use them too.

Rule 1: Context before content

The most common mistake with AI writing is asking for a blog post on an empty page. "Write an article about hotel marketing." The result is generic, because the AI knows nothing about you.

Before AI writes anything for a client, it needs to understand:

• the business: what it offers, for whom, and what makes it different;
• the audience: their questions, their words, their objections;
• the voice: how the brand sounds, with real examples;
• the facts: prices, processes, numbers that are true and approved.

This is why a generic AI blog rarely performs. The value is not in the writing. It is in the knowledge and experience behind it, which only the business has.

Rule 2: Give AI permission to say "I don't know"

AI models are trained to be helpful. Without guidance, "helpful" can mean filling a gap with something plausible.

Anthropic's own documentation on reducing made-up answers lists three basic techniques, and we use all of them:

• Allow the model to say "I don't know." Explicit permission to admit uncertainty, which the documentation says can drastically reduce false information.
• Ask for direct quotes first. When working from long documents, have it extract the exact sentences before it reviews or summarizes.
• Require a source for every claim. If it cannot find a supporting quote, it removes the claim.

There is a trade-off. These rules make output more cautious and less creative. That is fine for facts. For brainstorming taglines, we switch them off.

Rule 3: Facts come from sources, not from memory

Every number in our articles comes from a source we can name and link: a study, an official announcement, our own data. When a figure comes from a company that sells something related, we say so.

When we cannot find a source, we leave the claim out. An article with fewer numbers is better than one with a wrong number.

A real example from this month: for this content plan, every search volume we use comes from a named source, Google Ads data pulled through DataForSEO. Where Google returned no volume, we wrote "not measured" instead of estimating. That is less impressive, and it is true.

Open notebook and pen on a white desk

Rule 4: Let customers write the brief

One of the most useful AI tasks we know is also one of the simplest. Collect your customer reviews and messages, paste them in, and ask:

• How do customers describe what we do, in their own words?
• What are they sceptical about before they buy?
• What outcomes do they mention most?

You get a summary of your customers' language in minutes. Then check: are the quotes real, and in the reviews? With Rule 3 in place, they should be.

This is how better headlines, ads and landing pages start: with the words your customers already use.

Rule 5: People check every page

AI does the groundwork. People check the result. For us, every page and every client message is read by a person before it goes out.

What does the person check?

• Facts: does every number have a source?
• Fit: does it sound like the brand?
• Value: would a real customer learn something?
• Honesty: does it promise anything we cannot deliver?

Rule 6: Start with your hardest monthly task

If you are wondering where to begin with AI in your own business, skip the easy demos. Pick the task that takes your team the most time every month, and work out how AI could do the first 80% of it. Then write down the steps, so the next time is faster.

That is where the real hours are saved.

What this means for your marketing

AI makes good marketing faster. It does not make bad marketing good. The businesses that win with AI will be the ones that feed it real knowledge, hold it to real standards and keep people responsible for the result.

What to take away

• Give AI context about your business before you ask it to write.
• Allow it to say "I don't know".
• Ask for quotes and sources, and remove unsupported claims.
• Use customer reviews to find the words that sell.
• A person checks every page before it goes out.

Want content that is fast to produce and true to your business? See our SEO content writing services, or book a call.