AI ad mistakes and how we avoid them

We use AI every day: for ad copy variations, translation into several languages, first drafts of scripts and faster analysis of results. It saves a lot of time. We’ve also seen where AI makes costly mistakes when nobody checks the output.

Literal translation

This is the most common mistake: AI translates the ad correctly in grammar terms, but it reads like a translation. The call to action is translated word for word and sounds odd in the market’s language. Expressions that feel natural in English simply make no sense in another language.

So our translations go through three steps: a base translation, a rewrite the way a local would say it, and then combining the best of both. Product and brand names are never translated, and key terms for each language are kept in a separate list so they stay consistent across all ads.

The wrong tone

In some languages the difference between the informal and formal “you” decides whether an ad sounds friendly or pushy. AI often mixes them in one text. We set the tone for each market and product in advance and check that it holds everywhere. Customer reviews we quote always sound like a real person, in informal language.

Claims the platform won’t accept

AI writes persuasively and easily crosses the line: it promises results the product can’t guarantee or makes health claims. For supplements, cosmetics and intimate care, that means rejected ads and warnings on the account. Every AI-generated text goes through a person who knows the category’s rules.

Look-alike visuals

AI images often look alike: the same lighting, the same smiling people, the same “perfect” setting. If all your ads look like that, Meta treats them as nearly identical and they reach the same group of people. People also recognize them as generated and trust them less. We use AI visuals for quick tests of ideas, but winning directions are often produced with real photos or video.

How we work with AI

• AI makes more variations, and a person decides which ones go live.
• Every ad is approved by a person before launch.
• Guidelines (brief, tone, banned claims, terms per language) are written down once and reused every time, instead of being explained from scratch.
• Finished videos with real people are never altered with AI. If a fix is needed, we only cut, without changing what was said.