AI-generated content and search: what Google says

Google has written about AI-generated content in several places, at different dates and for different purposes: a guidance page, a spam policy, a self-assessment for content creators and a dated announcement. They give a consistent test, which looks at what a page does for its reader and why it exists, and ignores which tool produced the words. This page reads the documents side by side, records their dates, and marks the edges of what they say.

The documents and their dates

The dates matter because older commentary quotes older wording. The ranking systems guide, for one, lists the helpful content system among retired systems. It says the system was announced in 2022 and that in March 2024 it became part of Google's core ranking systems. A source that still presents it as a separate, current system is out of date on this point. The "last updated" lines above are what each page displayed when it was read for this article, so check them again before relying on any detail.

What the documents permit

Three statements establish the baseline.

• The guidance page says generative AI can be particularly useful when researching a topic and to add structure to original content.
• The 2023 post says automation has long been used to generate helpful content, naming sports scores, weather forecasts and transcripts, and that not all use of automation, including AI generation, is spam.
• The 2023 FAQ answers the headline question directly: appropriate use of AI or automation is not against the guidelines, meaning that it is not used to generate content primarily to manipulate search rankings.

The same FAQ also says AI gives content no special gains. In its words, it is just content: if it is useful, helpful, original and satisfies aspects of E-E-A-T, it might do well, and if it does not, it might not. Nothing in that sentence promises a position.

What the spam policy prohibits

The scaled content abuse section defines the practice as many pages generated for the primary purpose of manipulating search rankings and not helping users. It says the practice typically involves large amounts of unoriginal content that gives users little or no value, "no matter how it's created". Its examples include:

• using generative AI tools or similar tools to generate many pages without adding value for users
• scraping feeds, search results or other content to generate many pages, including through automated transformations such as synonymizing or translating, where little value reaches users
• stitching together content from different pages without adding value
• creating multiple sites to hide the scale of the content
• creating many pages whose content makes little or no sense to a reader but contains search keywords

The policy ends with an instruction: if you host such content, exclude it from Search. Elsewhere the spam page explains the consequence: sites that violate the policies may rank lower or not appear in results at all, and violations are detected through automated systems and, as needed, human review that can result in a manual action.

Two conditions appear in the definition. There are many pages, and the purpose is manipulation. A tool-generated page does not meet the definition by existing, and a human-written page does not escape it by being human-written.

Practices set against the texts

The table applies the quoted passages to concrete cases. It is a reading of Google's wording by the author, and Google's own systems decide real cases.

Merchant Center is worth a sentence. The guidance page says AI-generated images in that program must carry IPTC DigitalSourceType TrainedAlgorithmicMedia metadata, and that AI-generated product titles and descriptions must be specified separately and labeled as AI-generated. That is a rule for merchants and says nothing about ordinary articles.

Who, how and why

The helpful-content page offers a structure for judging any page, AI or not.

• Who. Readers should be able to tell who created the content. Bylines should lead to information about the author. The page warns that fabricated creator profiles, such as AI-generated headshots, made-up names or false credentials, are a form of deception that makes a page untrustworthy.
• How. Where automation or AI substantially generates content, the page asks whether its use is evident, whether background on how it was used is given, and why it was seen as useful. It describes disclosures as useful where someone might think "How was this created?" and says to add them when reasonably expected.
• Why. The page calls this the most important question. Content made primarily to help people aligns with what Google's systems seek to reward. Content made primarily to attract search visits does not, and using AI to produce content mainly to manipulate rankings is stated to violate the spam policies.

The same page explains where E-E-A-T fits. Google's systems identify a mix of factors that can indicate experience, expertise, authoritativeness and trustworthiness, and of these trust is the most important, though content need not show all four. E-E-A-T is not itself a specific ranking factor, and topics that could significantly affect health, financial stability or safety, or the well-being of society, get even more weight on it. AI involvement does not change that framework in either direction.

The guidance page also points to the Search Quality Rater Guidelines, sections 4.6.5 on scaled content abuse and 4.6.6 on main content with little or no effort, originality or added value. It adds that the guidelines are not a guide to ranking first, and that raters' ratings do not directly influence ranking.

What the documents do not say

Absences are easy to misread, so the list below is limited to the pages cited in this piece.

• No claim of AI detection. The 2023 post says spam-fighting efforts continue "however spam is produced", and its FAQ says systems including SpamBrain analyze patterns and signals to identify spam content. That describes spotting spam.
• No required label for ordinary pages. The wording is "consider adding" disclosures where reasonably expected.
• No word count. The helpful-content page lists writing to a particular word count as a warning sign, and answers its own question: Google has no preferred count.
• No number for "many". The policy gives no threshold of pages.
• No ranking promise. Every statement about doing well is conditional.

A test for a planned batch

Before generating a set of pages, write answers to five questions. Each comes from a passage above.

1. Why does each page exist? The honest answer should be a reader's need. A keyword the site wants to appear for fails the "why" question.
2. What does this page contain that the other pages in the batch lack? If the only difference is a name or a place, the batch matches the pattern the doorway and scaled content passages describe.
3. Where does the information come from? Information you supply, collect or test yourself is original. Information paraphrased from the top results is rewriting, which the helpful-content questions single out.
4. Who reviewed it, and when? A named person, with a date.
5. Would you publish half the batch? If the answer is yes, publish half and see which pages earn their keep.

Why volume is the real constraint

Google gives no figure, so the arithmetic below is an editorial yardstick only. With 10 hours a week available for review, a page that needs 30 minutes of checking allows 20 pages a week, a page that needs 60 minutes allows 10, 90 minutes allows 6 whole pages, and 120 minutes allows 5. The 4,000-page template above would take 2,000 hours at 30 minutes each, which is 200 weeks of that 10-hour allowance. A site generating more pages than it can review has decided, in effect, not to review them. The method for reviewing is in reviewing AI-written copy before it goes public, and the limits of site generation are in what an AI website builder can and cannot do.

If generated pages already exist

Start from the instruction in the policy, which is to exclude such content from Search. In practice, make an inventory of generated URLs and sort each into one of three groups: a page you would stand behind (keep, and review it), a page that could earn its place with real information (improve), and a page that exists only to fill a keyword (remove, or mark noindex). Google's block indexing documentation says a noindex rule only works if the page is not blocked by robots.txt and the crawler can reach it, so do not block those URLs in robots.txt at the same stage; sitemaps, robots.txt and Search Console covers the difference. Google's documents give no timetable for recovery, and none is implied here.

Checklist

• Read the dated documents themselves, and note the "last updated" line you read.
• Ask the three questions in order: who made this, how, and why.
• Count pages against the review hours you actually have.
• Keep generated pages out of Search unless a person has reviewed them and they add something.
• Treat the absence of a rule, a number or a detector in the documents as an absence, and make no claim in its place.