The Limits of AI Content Production: What Do Google’s Policies Actually Say?

Last updated: 10 September 2026Rules, fees and platform policies change. This guide is reviewed and kept current.

There are two extreme claims circulating about how producing content with AI affects search visibility: “Google penalises AI content” and “scale with AI and you win”. Both are wrong — and Google’s own documentation says so with surprising clarity.

What Google prohibits is not using AI. The behaviour it prohibits has a name, and the definition is verbatim: scaled content abuse is generating many pages whose primary purpose is manipulating search rankings rather than helping users. Among the examples listed in the policy is “using generative AI tools or other similar tools to generate many pages without adding value for users”.

Notice the distinction: what gets penalised is not the tool but the combination of scale plus absence of value. That distinction draws a fairly clear line around where an e-commerce site can use AI and where it should not. Below I set out what the policies say word for word, what is and is not required on AI search surfaces, and the EU obligation that came into force in August 2026.

Is labelling mandatory? Not for Search; yes for product data

This is the most frequently misreported topic. There are two different rules in two different Google products.

Labelling is not mandatory on the Google Search side. The language in Google’s guidance on generative AI content is entirely advisory: sharing information about how a piece of content was created can help give your readers more context, and consider adding information about how the content was created in a way that suits your audience. The self-assessment questions in the helpful content guidance sit in the same frame: is the use of automation, including AI generation, made obvious to visitors through disclosures or other means?

So Google has no requirement to label AI content; it has a documented recommendation. Not labelling is not a policy violation — it is a gap on the quality and trust (E-E-A-T) side. Getting that distinction right matters: the sentence “Google wants an AI label” is wrong everywhere it is published.

On the Merchant Center side, labelling is mandatory. Google Merchant Center’s AI content policy sets an explicit technical requirement:

• Images: all images created using generative AI must include metadata indicating the image was AI-generated, using the IPTC DigitalSourceType tag. Accepted values: TrainedAlgorithmicMedia, CompositeSynthetic, AlgorithmicMedia. Also: do not remove embedded metadata tags such as IPTC DigitalSourceType from images created with generative AI tools.
• AI-generated titles: the structured_title field, with the digital_source_type (= trained_algorithmic_media) and content sub-attributes.
• AI-generated descriptions: structured_description, with the same two sub-attributes.

So if you generate product descriptions with AI and feed them to Merchant Center, you have a compliance obligation here. This is a line item many e-commerce sites are unaware of. (The policy page does not state an effective date, so I am not giving one.)

Reviews: there is no separate “AI policy”, but it is already a violation

I could not find a named Google policy on “AI-generated reviews”. But the existing rules already exclude it. Two sentences in the review snippet documentation are definitive enough:

• Ratings must come directly from users.
• If the entity being reviewed controls the reviews about itself, the relevant pages are not eligible for the star review feature.

Reviews “not based on a genuine experience of the product or service” and reviews given in exchange for a benefit where the incentive is not clearly and conspicuously disclosed are also prohibited.

The conclusion: generating reviews with AI does not violate a “new rule”; it violates a very old one. The consequence is not new either — removal from the star review feature and the risk of a manual action. In short, this is a plain boundary that is not even within the scope of the AI debate.

Site reputation abuse: tightened in 2024

This policy matters for e-commerce sites publishing other people’s content on their own site. The official definition: site reputation abuse is the tactic of publishing third-party content on a site largely because of the established ranking signals it has earned from the host site’s first-party content.

And an exemption: having third-party content is not a violation in itself; for it to be a violation, the third-party content has to be published largely because of the host site’s established ranking signals.

However, on 19 November 2024 the policy was tightened: using third-party content to take advantage of a site’s ranking signals counts as a violation regardless of first-party involvement. So the defence “we work on it too, we have editorial control” no longer holds. If you host guest posts, sponsored content or coupon/discount pages on your site, you need to be sure the real reason for it is not your site’s authority.

AI Overviews and AI Mode: the things you do not need to do

This is where the most money and time gets spent — and Google’s documentation says most of what is being spent is unnecessary. From the AI features page: there are no additional requirements to appear in AI Overviews or AI Mode, and no other special optimisations are needed. And: there is no special schema.org structured data you need to add.

Google’s generative AI optimisation guidance, updated in July 2026, goes further. From its “misconceptions” section:

• llms.txt: you do not need to create new machine-readable files, AI text files, markup or Markdown to appear in Google Search. Doing so neither harms nor helps your site’s visibility or ranking in Google Search, because Google Search ignores them.
• Chunking: there is no requirement to break your content into tiny pieces for AI to understand it better. There is no such thing as an ideal page length.
• Writing differently for AI: you do not need to write in a special format just for generative AI search.
• Chasing artificial “mentions”: hunting for artificial mentions across the web is not as useful as it appears.
• And most importantly — this is a spam violation: creating separate content for every possible variation people might search for (for example focusing on other questions people ask, or fan-out queries), doing so primarily to manipulate rankings in Google Search or generative AI responses, violates Google’s scaled content abuse spam policy.

That last item puts most of the services sold as “let us optimise for AI search” directly into the spam category. Google’s own framing is this: as far as Google Search is concerned, optimising for generative AI search is optimising for the search experience — that is, it is still SEO.

So what is actually required?

There are two technical conditions, and the second is easily missed:

1. The page must be indexed and eligible to be shown with a snippet, and must meet Search’s technical requirements.
2. In the guidance’s words: in addition to Search’s technical requirements, for a site to be eligible to appear in generative AI features it must be opted into Search generative AI features in Search Console.

And a warning: a page meeting all requirements, best practices and policies does not mean Google will crawl, index or serve its content.

The official position on structured data is balanced: structured data is not required for generative AI search, but it is a good idea to keep using it as part of your general SEO strategy, because it helps you qualify for rich results in Google Search.

What is said specifically for e-commerce

The relevant section of the guidance gives clear direction: where appropriate, generative AI responses can include product listings, product information and information about local businesses. Using products such as Merchant Center (for example Merchant Center feeds) and Google Business Profile can help your products and services appear both in AI responses and in other Google Search results.

So the concrete lever for AI visibility in e-commerce is not a writing technique: it is a properly fed Merchant Center feed and correct product structured data. On product page markup the distinction is this: merchant listing for pages where the customer can buy directly, product snippet for pages where they cannot (editorial or review pages). And the two are not alternatives: providing a Merchant Center feed alongside structured data on web pages maximises your eligibility for experiences.

“Non-commodity content” — a new and useful concept

Google now uses the term non-commodity content, and the example it gives is useful to anyone writing an e-commerce blog. Commodity content: “7 Tips for First-Time Home Buyers”. Non-commodity content: “Why We Skipped the Inspection and How We Saved: A Look at the Sewer Line”.

And the sentence it adds underneath is a standard to measure against: do not recycle what others on the internet have already said, or what a generative AI model could easily produce.

That draws the line around AI content production exactly. What AI produces easily is, by definition, commodity content. What you have and AI cannot produce: the return-reason breakdown from your own order data, the outcome of the negotiation you had with your own carrier, the conversations you had with the 40 customers who sent your product back. AI cannot write those — but if you tell it, it can edit them. That is the right division of labour.

Google-Extended: what it does and does not do

There is a widespread confusion here too. Google’s official definition: Google-Extended is a standalone product token that web publishers can use to manage whether the content Google crawls from their sites is used to train future generations of Gemini models and for grounding in Gemini Apps. And: Google-Extended does not affect a site’s inclusion in Google Search, nor is it used as a ranking signal in Google Search.

The practical consequence: blocking Google-Extended does not affect AI Overviews or AI Mode, because those two surfaces are fed from the Search index. The levers there are the snippet controls — nosnippet, max-snippet, data-nosnippet — or noindex. But using them has a price: suppressing the snippet also suppresses ordinary Search features. So a setting like “I do not want to appear in AI Overviews but I do want to appear in normal search” is not something Google currently offers.

A technical note: Google-Extended is managed only through robots.txt; it does not exist as a robots meta tag.

On the standards side there is formal work at the IETF on AI crawler preferences — the AIPREF working group. The vocabulary draft is at revision 07 as of August 2026 and is not yet an RFC. The draft defines two usage categories, train-ai and search, each expressible as allowed / disallowed / unknown. So it is too early to say “the standard says X” in this area; it is a heading to revisit in a few months.

From August 2026: the EU transparency obligation

This is the newest and least-known part of the picture. The Article 50 transparency rules of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689) have been in force since 2 August 2026. From the European Commission’s announcement of 31 July 2026:

From 2 August 2026 … new transparency rules will start to apply, requiring certain AI systems to tell users that they are interacting with AI and that content has been generated or modified by AI.

From the same announcement: chatbots and other interactive AI systems will have to tell users they are dealing with AI rather than a human. Deepfakes will have to be labelled. AI-generated or modified content will also carry machine-readable marks so it can be detected more easily.

The Commission also published guidance on the Article 50 transparency obligations on 20 July 2026 and announced that more than 180 organisations had signed the Code of Practice on Transparency of AI-Generated Content.

A misunderstanding worth correcting: the AI Omnibus that entered into force in 2026 (27 July 2026) extended the transition periods for high-risk systems (to 2 December 2027 and 2 August 2028) and added a ninth prohibited practice — but it did not postpone the transparency obligation.

Does this affect a Turkish e-commerce site? The answer depends on the extraterritorial scope in Article 2 of the Regulation, and I have not verified that from the primary text — so I am not writing a definitive answer. The practical approach: if you run a chatbot serving customers in the EU, ask your legal adviser whether you are in scope. But while you wait, the thing to do is cheap and correct anyway: have your chatbot say in its first message that it is an AI assistant. That is the right behaviour under any regime and it does no harm to customer satisfaction — people set their expectations appropriately when they know they are talking to a machine.

A practical workflow: where should you use AI?

All of the policies above point to the same place. Use AI where you can verify its output and where you are not creating scale.

• Use it for: turning findings from your own data into prose. Editing a process you have described yourself. Reformatting an existing piece for different channels. Summarising reviews and return reasons (summarising, not generating — the kind of work AI is most reliable at). Generating product descriptions from structured input — provided you do the Merchant Center labelling.
• Do not use it for: generating pages for search query variations (an explicit spam violation). Generating reviews. Publishing a number, rate or regulatory claim whose source you have not verified. Writing an experience narrative about a product you have never touched.
• Measure: Google’s recommendation is explicit — use the generative AI performance report in Search Console to measure how your content performs in generative AI features in Google Search and Discover. And a warning: no third-party tool has access to our internal ranking or AI systems. So what the tools selling an “AI visibility score” measure is not Google’s data.

A note on timing

Google’s recent update calendar is relevant to when this was published: the August 2026 spam update started on 18 August and completed around 21 August. If you are seeing traffic fluctuation in this period, it may be that change rather than anything you did on your own site. Before deleting content in a panic or rewriting wholesale, wait at least two weeks and look at the query-level change in Search Console.

Setting the expectation correctly

All of these policies reward the same thing: content nobody but you could write. Google calls it “non-commodity content”, the helpful content guidance calls it “people-first”, and the spam policy describes the opposite of it.

AI does not change that equation; it simply reduced the cost of producing commodity content to zero. The value of something whose cost is zero also approaches zero — so the way to compete with content AI produces easily is to invest in what AI cannot produce: writing up your own data, your own operational experience and your own mistakes.

And a final note of honesty: scaling content production with AI can work in the short term. Google’s spam updates run periodically and with a lag. But that does not mean there is no risk — it means the bill arrives later. What matters to a business is not traffic rising for one quarter but still being there in three years.

Related reading

• Making content citable in AI search: getting cited by AI search.
• Building automation on human approval: AI automation in small e-commerce.
• Tool selection, cost model and the limits of KVKK: no-code automation setup.
• Page performance on the technical side: WooCommerce speed optimisation.

Sources: Google Search Central spam policies (last updated 15 May 2026), helpful content guidance (10 December 2025), generative AI content guidance (10 December 2025), AI features page (10 December 2025), generative AI optimisation guidance (10 July 2026), review snippet documentation (24 July 2026), crawler documentation (14 July 2026); Google Merchant Center AI content policy; European Commission AI Act pages (3 August 2026) and the press release of 31 July 2026; IETF AIPREF working group drafts. All as of August 2026. This post is not legal advice.

Sources and official links

• Google Search Central — Spam Policies — The official definition of scaled content abuse
• Google Search Central — Creating Helpful, Reliable, People-First Content — People-first content and the E-E-A-T self-assessment questions
• Google Search Central — AI Features and Your Website — Visibility in AI features and the controls available to publishers
• Google Search Central — Intro to Structured Data — How structured data works and how to use JSON-LD
• Google Search Central — Supported rich result types — The rich result types Google currently supports

Rules, fee schedules and platform policies change. The figures in this post reflect the position at the time of writing; confirm the current position from the official sources above before you act.

About the author

Toros Panos

I work on e-commerce and marketplace operations. On this site I publish practical guides on Trendyol, Hepsiburada and Amazon operations, unit economics, micro-export and no-code automation — each built on official platform documentation and current regulation.

Consulting . All articles . LinkedIn