Google does not ban AI content. It punishes missing human work in Search and YouTube

Google does not punish text because a machine wrote it. It punishes text that no human checked, signed or improved. That is the common thread in four Google rulebooks we compared after a run of changes. On 1 October 2026 Google edited two Search guides. The same day YouTube said it would cut the reach of lazy Shorts. A week later Google finished its longest spam update of the year. None of these documents bans AI content. Each one asks for a different piece of human work.

Google added three AI warnings to its Search guides on 1 October

The first change sits in the Search Central guide on generative AI content. Google now tells site owners to fact-check and review all AI-generated content by hand before they publish it. We reported the edit on 2 October. The new lines explain why. In Google’s words, “generative models don’t retrieve facts”. They predict which words are likely to come next, so their output can contain invented facts.

The review duty covers more than the article. Google lists title tags, meta descriptions, structured data and image alt text. These are the small texts that can show up in search results. The page was last updated on 5 October.

The second change came in the guide to helpful content. Google now calls fabricated author profiles a form of deception. Its examples are AI-generated headshots, made-up names and false credentials. Google says such tricks make a page untrustworthy and signal low quality. The same guide asks site owners to explain how content was made when readers would reasonably ask.

Neither edit came with a ranking change, a deadline or a penalty. Google did not say how many pages a team should sample. It did not say who should sign off. The rules describe what Google wants, but they do not say how it will check.

Search, YouTube, Merchant Center and Discover set four different bars

We put the current texts of four Google rulebooks side by side. Each one targets a different missing human step.

Google Search has the oldest rule. Since 5 March 2024 its spam policy on scaled content abuse covers mass-produced pages “whether automation, humans or a combination” made them. The tool does not matter. What matters is many pages with little value. Google said at the time it expected 40% less low-quality content in results. In April 2024 it reported a 45% drop.

YouTube changed its Partner Program rules on 15 July 2025. The old “repetitious content” rule became “inauthentic content”. Channels cannot earn money from AI videos made with generic templates that look mass-produced. Channels with AI personas posing as human experts on health, law, money or politics cannot earn either. AI used as a tool for your own idea is allowed. On 1 October YouTube added a reach cut for reposted Shorts. A voiceover that only describes the clip will not count as your own work.

Merchant Center, Google’s tool for product listings, has the strictest rule. AI-generated product images must carry a metadata tag called TrainedAlgorithmicMedia. AI-written product titles and descriptions must be sent separately and labelled as AI-generated. Here Google demands a label, not only a review.

Discover, the article feed on Android phones and in the Google app, is where Google itself plans to use AI text. Its own team is hiring engineers to recommend AI-written stories. We describe the job ads below.

Search wants added value. YouTube wants original commentary. Merchant Center wants a label. Search guides now also want a real author and a human fact-check.

Google’s spam update ran 13 days, its longest of 2026

The September 2026 spam update started on 24 September, as we reported then, and ended on 8 October. It took 13 days and 16 hours, according to the Search Status Dashboard. The three earlier spam updates of 2026 took between 19 hours and less than three days. Google did not say what this one targeted.

Early data comes from Diib, an analytics firm. It looked at 32,569 sites with Google Analytics connected. Search Engine Journal reported its numbers. The median share of visits from unpaid Google search fell from 34% to 31%. The share of sites that lost half or more of that traffic doubled, from 5% to 10%. Diib called this an early read. Its data cannot show which sites Google flagged as spam.

Google’s Discover team plans to recommend AI-generated stories

On 11 October Search Engine Journal found four Google job ads about AI content in Discover. One listing for a staff engineer in Mountain View says the team “leads by recommending AI-generated content”. The hire will build AI agents and checks for facts, freshness, coherence and tone. Another listing says Discover has more than 500 million daily users. We covered the listings on 12 October.

Google asks publishers to fact-check all AI content by hand. Its own Discover team plans automated checks for its AI stories. The job ads do not say whether those stories will be labelled.

The EU asked publishers the same question on 2 August 2026

The AI Act, the EU law on artificial intelligence, has applied its transparency rules since 2 August 2026. Article 50(4) covers AI-written text published to inform the public on matters of public interest. Such text must be labelled as AI-made. There is one exception. A human must review it or edit it, and a person or company must take editorial responsibility.

That exception mirrors Google’s new Search wording closely. A European news site that fact-checks its AI drafts and names a responsible editor meets both. A site that publishes unchecked AI news may face a demotion in Google and a breach of EU law.

Regulators are also looking at Google’s own use of content. On 9 December 2025 the European Commission opened an antitrust probe. It asks whether Google used publishers’ content in AI Overviews and AI Mode without fair pay or a way to refuse. On 5 October 2026 Poland’s competition office charged four Google companies over talks with Polish publishers. Creators depend on Google too. In a YouTube survey of 20,617 creators from all 27 EU countries, 83% said they rely on its recommendations.

Check three places on your site this week. Open your author pages and remove any stock or AI headshot without a real person behind it. Then pull a sample of AI-written meta descriptions and alt texts and read them. Last, name the editor who signs off AI drafts on news pages.