AI Writes Your Website Text? What Google Says to Check by Hand

Service descriptions, product cards, blog posts: many business owners ask ChatGPT or another AI to write them and publish within minutes. It is fast and cheap. But on 1 October Google updated its guidance page on AI content, and it now contains a sentence that everyone who runs a website should know: everything an AI generated must be checked by hand before it is published.

Below we go through what is actually written there, what you can check yourself and where money gets lost in practice. We read Google’s guidance on 7 October 2026.

What the guidance says

The page is called “Google Search’s guidance on using generative AI content on your website”. The main points:

• Using AI to write text is allowed. There is no ban on the mere fact that “AI wrote it”.
• Google says that generative models do not look facts up: they “predict a likely sequence of words based on their training data”. That is why the text can contain inaccuracies, known as “hallucinations”, and why it is “critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing”.
• You need to check not only the page text but also what shows up in search: the title (<title> element), the meta description, structured data and image alt text.
• Using AI to “stamp out” many pages that add no value for people may violate Google’s spam policy on scaled content abuse.
• Google recommends telling readers how a piece of content was created when it helps them understand the context.

Is this a new requirement? Sources disagree here, so we will give both sides. Many present it as a new rule: “Google now says so directly”. According to industry media, the December 2025 version of the guidance did not have this sentence. On the other hand, the same media report that Google announced no new penalty and no new ranking signal, and in its changelog explained the edit as bringing the document in line with presentations at its developer events. The rules on mass-produced pages with no value existed before. So this is not a “new punishment” but a written-down quality requirement. For an owner the conclusion is the same: responsibility for accuracy on your website is yours.

1. Facts: what you are responsible for

AI writes confidently and smoothly even when it is wrong. A customer will read on your website not a “model error” but your prices, your deadlines and your warranty terms.

What to check yourself, with the page open next to your price list and contract:

• prices, discounts, “from” and “up to”;
• product specifications and composition, sizes, power, volume;
• delivery, return and warranty terms and timeframes;
• address, opening hours, phone numbers;
• names of certificates, licences and partner brands: do you really have them.

You do not need any tools for this, just careful reading. Practitioners also advise not to hand AI the things where a mistake is expensive: terms of promotions, warranties and credit (legal risks), as well as “news” about things the model may not have known about. This is the view of practitioners, not a Google rule, but that is exactly where a mistake costs the most.

2. What shows up in search: title, description and photo captions

This is easy to generate in bulk and never read. In plain words:

• Page title (title) is the headline link in Google results;
• Page description (meta description) is the text under it;
• Structured data is hidden markup in the code that tells search what is on the page, for example a product, a price or reviews;
• Alt text (alt) is a caption for an image that screen readers read out and search can see.

How to check. Type site:your-domain.com into Google and look at how your pages appear in the results. Does the title make sense? Is the description repeated across dozens of pages? Does it promise something the page does not have?

How long a title should be. Practitioners disagree here: some advise staying within 60–70 characters, others 55–70, and for the description they name 120–150. Google gives no such number. Its documentation says there is no limit on the length of a <title> element, but it is truncated in results, typically to fit the device width. The same goes for the meta description. Google may also show its own title instead of yours if yours is missing, outdated, does not match the content or is the same on many pages. So the practitioners’ guideline is useful, but it is no guarantee.

Photo captions. According to Google’s documentation, alt text should be useful and information-rich and describe what is in the picture, and stuffing it with keywords may cause the site to be seen as spam. “Dalmatian puppy playing fetch” is better than “dog buy dog price”.

3. The main trap: lots of pages with no value

The temptation is clear: in one evening AI can produce a hundred “service + city” pages. But in its spam policies Google defines scaled content abuse as many pages generated primarily to manipulate search rankings rather than to help users. Among its examples it names generating many pages with AI tools without adding value.

Practitioners suggest how to tell a useful page from a stamped-out one. A page for a particular city makes sense only if it has something of its own: an address, opening hours, directions, real facts about the location. A list of cities in the text, or the same paragraphs with only the city name swapped, is a risk. Another trap practitioners write about: mass-rewriting old descriptions “for uniqueness”. That can also look like stamping out pages.

Test it yourself with one question: will a person find something on this page that is not on the neighbouring one? If not, the page is not needed.

4. For online stores: a separate rule for Merchant Center

Google’s guidance has a separate point for online stores. Product titles and descriptions generated by AI must be submitted separately and labelled as AI-generated. In Merchant Center help, the attributes for this are structured_title and structured_description with the value trained_algorithmic_media. If the text was created without AI, the usual title and description are used. For AI-generated images, the guidance requires a metadata label, TrainedAlgorithmicMedia.

In practice this is a question for whoever uploads your products: do they know which descriptions were written by AI, and are they labelled? We explain how Merchant Center works together with advertising in the Merchant Center section. For other requirements for product photos, see our article on small photos.

AI detectors and the “percentage of edits”: what practitioners say and what Google says

Owners are often advised to “run the text through an AI detector” or “rewrite 60–80% by hand”. Practitioners really do this: some name that share of edits, others measure the uniqueness of raw AI text and recommend 80–85%. They themselves say these are guidelines, not a standard.

Google’s guidance mentions neither detectors nor percentages of edits. Its yardstick is different: is the text accurate, good quality and useful, and are pages without value not being stamped out. Here is how we put our position. A detector can be used to spot a “machine” style, but it cannot be the gate to publication: the gate is a page read and checked against the facts by a person.

From our practice

We use AI for drafts ourselves. Our rule is simple: every figure, date and name is checked against the primary source before publication, not after a complaint. We check translations against Google’s help in the language of the text. We also wrote this article while checking Google’s guidance on 7 October 2026. This is our experience, not statistics: we do not know how many websites have suffered from unchecked AI texts, and Google does not publish such figures.

Summary: five questions for whoever runs your website

1. Which texts on our website were written by AI?
2. Who checks prices, deadlines and terms before publication, and how?
3. Were titles, page descriptions and photo captions checked, not just the main text?
4. Do we have dozens of similar pages made “for search”?
5. If we are a store: are AI-written product descriptions labelled in Merchant Center?

“I don’t know” is also an answer: it means a check should be scheduled. If your website is not properly connected to analytics either, see the checklist on what should be connected.

Want a free audit of your Google Ads account? Send us a request — we will look at your account and tell you what is worth fixing.