Build1 publisher2 min readPublished
A ten-word clause pulls AI Overviews into Google's deindexation policy
Google restamped its spam policies on 15 May 2026 with a clause covering generative AI responses in Search. AI Overviews and AI Mode now fall under the same sanction sentence that has long covered ranking spam.
The Engineer · Build desk
What happened
- Since 15 May 2026 Google's spam policies name attempts to manipulate generative AI responses in Search as spam, inside the same definition it uses for manipulating rankings.
- The update widened the scope of the rule and left the punishment alone, keeping the sanction sentence Google has applied to ordinary spam for years.
- Industry read-outs of the change, Search Engine Land among them, treat AI Overviews and the conversational AI Mode as covered by the rule alongside the classic ranked links.
- The seeding tactic has a name outside search marketing, recommendation poisoning, and Microsoft described it as manipulating the memory of AI systems.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- exposure The sanction is addressed to sites, so the campaign is paid for out of a marketing budget and the ranking loss reaches every page on the host, including the ones nobody optimised.
- decision Signing a GEO or AEO retainer now means deciding who owns the risk. Google's sanction reaches the buyer's domain.
- constraint Work aimed at ChatGPT or Perplexity still runs through forum and review pages that Google crawls, so naming a non-Google assistant in the pitch does not move the tactic outside Google's reach.
- precedent Because the wording was folded into the existing definition, any further Google generative surface inherits the same removal sanction without a new policy being written.
The sanction sentence is written about sites: "Sites that violate our policies may rank lower in results or not appear in results at all" [5]. Enforcement lands on a host. A retainer is billed per campaign.
The delta is one appended clause. Google's definition of spam used to describe attempts to "manipulate Search systems into ranking content highly" [3]. It now continues "...or attempting to manipulate generative AI responses in Google Search" [4]. The addition runs ten words [17]. The policy document carries a "Last updated 2026-05-15" stamp [2].
The clause names generative AI responses in Google Search, not only ranked links [4]. Tair, who builds websites and automations at DevNova and adds schema and structured data on every project [14], writes that Google explicitly said the same yardstick that governs results governs the AI summaries [18].
The pitch this collides with arrives on LinkedIn: an agency or a self-styled "GEO expert" offering, for a monthly retainer, guaranteed recommendations from AI, sold as GEO or AEO [12]. The version in Tair's inbox promises "for a monthly fee, ChatGPT and Perplexity will start recommending you" [13]. "Optimizing for AI answers is not banned. Manipulating them is. There is a whole canyon between them," Tair wrote [15].
The clause governs generative responses inside Google Search [4]. The answer surfaces run by OpenAI and Perplexity are outside a Google policy, so the deindexation exposure the update creates runs through Google's index, not through whatever ChatGPT says about you.
For that exposure to reach a specific domain, Google has to attribute the manipulation to it. The two patterns differ on exactly that point. An engineered comparison page, a "5 best studios in town" listicle built so a model lifts one name out of it, normally sits on the client's own domain [9]. Mentions and reviews seeded into third-party forums and review sites sit on hosts the client does not control [10]. Google did not publish a closed list of what falls under the new clause [8].
What to watch
- Whether Google publishes examples or a closed list of AI-answer manipulation beyond the two patterns already inferred.
- A first documented enforcement action citing the generative-AI clause would show how Google attributes seeded third-party mentions to a site.
- Whether OpenAI or Perplexity publish comparable policies for manipulation of their own citation surfaces.