What happened
In early October 2026 Google updated two Search Central pages that describe what it considers quality content:
- the guidance on generative AI content, updated on 1 October;
- the page on helpful, people-first content, whose latest version is dated 5 October.
Google's changelog gives the reason: the documentation was brought in line with the presentations the company gives at its developer events. So this is neither a reaction to a specific algorithm update nor a new policy from scratch.
Rather than retell other people's retellings, we compared the current versions of both pages with Wayback Machine snapshots from late September 2026.
What is genuinely new
The AI content page gained one paragraph. It explains that generative models don't retrieve facts but predict a likely sequence of words, so they can be wrong. Hence the requirement: before publishing, all AI-generated content must be manually fact-checked for accuracy. The review extends to title elements, meta descriptions, structured data and image alt text — everything that can appear in search results.
The page mentioned metadata before. What is new is the direct statement about manually reviewing all AI content.
The helpful content page gained a whole section. Google defines a page's main content and names four attributes search quality raters assess it on:
| Attribute | What it means in practice |
|---|---|
| Effort | How much human work went in. Generating large amounts of text without manual oversight is named as an example of minimal effort. Citing sources does not replace effort. |
| Originality | Whether the page offers information or a perspective not already available elsewhere. |
| Talent or skill | Whether there is enough skill for the page to do its job: clear writing, a working calculator, well-made video. Not every topic needs an expert. |
| Accuracy | For informational pages, factual accuracy. For topics affecting money, health or safety, high accuracy consistent with expert consensus. |
The same update adds a paragraph on authorship. Fabricated author profiles — AI-generated headshots, made-up names, false credentials — are explicitly called deception, and Google says any deception makes a page untrustworthy to both users and its automated systems.
What Google did not say
Several retellings after the update contradict the source.
"Google has banned AI text." No. The first sentence of the AI content page says generative AI is useful for researching a topic and structuring original material. The problem named is generating many pages without value for users. That scaled-content rule applied before October too.
"AI content must now be labelled." No. Google suggests considering a disclosure where readers might ask how the content was made. It is a recommendation, and it was on the page before October. Ordinary search pages have no mandatory labelling; the hard requirements are in Merchant Center (see below).
"Rater scores lower rankings." No. Google states separately that raters do not directly influence ranking: their ratings help check whether the algorithms work well. The rater criteria are useful as a mirror for self-assessment, not as a list of penalties.
Online shops: here the rules are hard
If you send products to Google Merchant Center, recommendations give way to specific requirements for AI content:
- AI images must carry IPTC
DigitalSourceTypemetadata with the valueTrainedAlgorithmicMedia. Embedded tags from generators must not be removed. - AI product titles and descriptions must be submitted in the separate
structured_titleandstructured_descriptionattributes, labelled as created with generative AI.
These requirements did not appear in October, and the AI content page itself points to them. If a neural network writes your feed descriptions and they sit in the ordinary description field, the feed formally does not comply.
A checklist for site owners
- Who checks the facts. Name the person who checks figures, dates and names against the primary source before publishing. Without them, an AI text does not go out.
- Check what appears in search, not just the text. Titles, descriptions, markup and alt text get the same review as the body.
- Something nobody else has. Your own calculation, test, photos, customer data, a calculator. If the page could be assembled from the top five search results, it fails the originality test.
- Real authors or none. The name and photo of a real person who wrote or reviewed the text, or an honest "editorial team". A fabricated expert with an AI headshot is the worst option.
- A pace your review can sustain. Publish as much as you can actually proofread. A hundred unreviewed pages a week is exactly what Google calls scaled content.
- For shops — the feed. Check that AI descriptions and AI product images are labelled the way Merchant Center requires.
Our position
The October update doesn't change the rules of the game, but it removes ambiguity. Before, one could treat manual review as a nice-to-have. Now Google states plainly that it is critical and lists what to check. For a small business that is mostly good news: a competitor flooding its site with unreviewed generated text is taking more risk than one who publishes less often but checks.
The position does not hold for sites where text plays no part in search: internal knowledge bases, password-protected pages, sites whose traffic comes only from ads. Google Search rules simply don't apply there, though factual accuracy remains your risk with customers.
Checked on 6 October 2026 on Google Search Central and Merchant Center Help; archived versions from the Wayback Machine, snapshots of 27 and 28 September 2026.