ChatGPT or Gemini can write the text of an ad and draw the picture for it in a minute. The temptation to launch it straight away is understandable: it is fast and free. But an ad is a public promise made in your business’s name, and both the platform and the buyer judge it with no discount for “the AI did it”. Below are six questions to ask yourself before launch. Each check can be done by you, without a marketer.
We checked the rules of Google, Merchant Center and Meta against their help pages on 27 October 2026. Where there is no primary source, we say so plainly.
1. Have the facts been checked separately from the style?
Smooth text hides mistakes. You read the ad, like how it sounds, and do not notice that the AI changed the volume, the ingredients or the warranty period, or invented a discount that does not exist. We wrote about how AI makes up details in “How AI Hallucinates: What It Can Make Up About Your Business”.
What to do. Read in two passes. First: facts only. Price, size, composition, warranty, address, opening hours: check every figure against your price list and your website. Second: only then ask whether it sounds good. The order matters: if you start with style, you will read the facts “by eye”.
2. Does the ad promise too much?
AI loves strong words: “guaranteed”, “the best”, “results in a day”. For an advertiser this is risky, especially in medicine, finance and weight loss.
What Google Ads policy says (the Ukrainian help page on its rule about unreliable claims; in English the policy is titled “Unreliable claims”). Prohibited are inaccurate claims or promises of improbable results, even if theoretically possible. Weight-loss claims that are unrealistically fast within a set period or with minimal effort are named separately, as are implausible promises of large profits with minimal investment, effort and risk. Guarantees are treated more carefully: if you guarantee specific results, clearly define your refund rules. So the word “guaranteed” is not banned in itself, but a refund condition has to stand behind it.
On consequences: for violating this policy Google says it sends a warning at least 7 days before suspending an account. For certain serious violations in healthcare (for example, unauthorised online pharmacies) the help page speaks of immediate suspension without warning. The policy pages for medicines and for financial products are separate, with their own restrictions and permission requirements, so read them if that is what you advertise.
Meta is similar in spirit. In its Advertising Standards, in the section on personal health and appearance (checked through a search; the page itself did not open during our check), it says that ads for diets and weight loss must not create a negative self-image, and a before-and-after image is given as a prohibited example. For credit cards, loans and insurance Meta requires, among other things, that they are shown only to people aged 18 and over and, where needed, that you hold a licence in the targeted country.
What to do. Cross out every “we guarantee” that you would not write into a contract. The same goes for “the best”, “No. 1”, “results in a day”: can you prove it? If not, replace it with a fact: “in business since 2015”, “we reply within 15 minutes”. We found no separate ban on the word “best” in Google’s help, so we do not claim it is banned. We suggest cutting it for another reason: buyers no longer believe it.
3. Whose picture is it?
A generated photo can look oddly like someone else’s product or carry another company’s logo or lettering in the frame. You may not notice it, but a competitor or the platform will.
What to do. Upload the picture to Google Lens (image search) and look at the results. They should not show other brands or products that look “like twins” of yours. Practitioners recommend the same step before publishing an AI picture, but we found no official Google or Meta rule that requires this check: it is our advice, not a platform requirement. Look at the details too: hands, text on packaging, numbers. Generators often break exactly those.
A separate note on Merchant Center. Google’s help page on AI-generated content (updated 1 October 2026) says that for online shops, Merchant Center has policies on AI-generated content: AI-generated images must contain the IPTC DigitalSourceType metadata with the value TrainedAlgorithmicMedia, and AI-generated product data such as titles and descriptions must be specified separately and labelled as AI-generated. If you run a product feed, check this in the technical set-up (or ask whoever maintains it).
4. Is it the same on the website?
The price, promotion and availability in the ad must match what a person will see on the website. If the AI “corrected” the price in the text and the page shows another, the ad faces two problems. First: a buyer who came for the price from the ad cannot find it and leaves. Second: for product ads Google compares the price in your feed with the price on the page and disapproves the product.
In Merchant Center this problem is called “Mismatched product price”. Google’s help suggests these steps: in the products section open the tab of products that need action, find the product, edit it, enter the price as it is on your website and save. After fixing, you request a review: according to the help page it can take up to 12 hours. One more detail from the help page: the price should be in the page’s HTML, not appear only through JavaScript, otherwise Google may not see it. We wrote about a similar case in “Price in Shopping Ads: Why People Click a Competitor”.
What to do. Open the ad and the product page side by side and compare line by line: price, old price, promotion, end date, availability, delivery, currency.
5. What do you count as success?
AI will easily make 20 ad variants. That is an example, not a figure from research: the point is that there will be many variants and you will have to choose between them. Many people go wrong here: the winner is declared to be the variant with the most likes or clicks. But a like does not pay for the goods, and a click costs money even if the person left at once.
Practitioners argue about what to use as the basis. Some look at the click-through rate (CTR), because it shows within a few days. Others hold that only leads and calls should be counted, because a catchy text can bring the wrong people. We found no official figure in Google’s help on how many impressions or clicks are needed to declare a variant the winner. Our position from practice: the winner is decided by a lead or a call, and clicks only help understand why.
What to do. Before launch, write down how much you are ready to pay for one lead. After a week, compare the variants by exactly that number, not by how nice the ad looks. If there are few leads and the difference is random, do not rush to name a winner.
6. Who read it before launch?
This is a question about a person, not about the AI. Google Search Central’s page “Using generative AI content” (updated 1 October 2026) says verbatim: “generative AI outputs may contain inaccuracies (also known as hallucinations)” and “It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing.” It adds that this review also applies to metadata such as title elements.
An important caveat: this page is about website content in Google Search, not about advertising. It says nothing about ads. We carry its logic over to ads by analogy, because the reason is the same: AI makes mistakes with confidence. We covered website texts in more detail in “AI Writes Your Website Text? What Google Says to Check by Hand”.
What to do. The second pair of eyes rule: an AI-made ad is read by someone who did not write the prompt. They do not know what you meant, so they see only what is actually written. Give them a checklist of these six questions.
From our practice
Most often we see not a “terrible” mistake but a small one: an old price in the text, “free delivery” that no longer exists, a promotion that ended two weeks ago. The AI inserts all of this confidently and without hesitation. So our order is: the AI gives a draft, a person checks the facts, then a second person reads it, and only then launch. This adds a check, but it is cheaper than paying for an ad with a mistake in it.
Summary
AI saves time on text and pictures but does not take responsibility away from the person who launches the ad. Check: facts separately from style; promises you are ready to keep; other brands in the picture; a match with the website; what you count as success; and who read it before you.
If the ad is already running and you are not sure all is well, we will look at it from the side of the rules and the results.
Want a free audit of your Google Ads account? Send us a request — we will look at your account and tell you where your advertising wastes money.