Black-Hat GEO: How AI Search Gets Gamed, and Policed

Black Hat GEO

Author:

Ara Ohanian

Published:

October 24, 2025

Updated:

November 5, 2026

On May 15, 2026, Google made explicit that its definition of spam covers AI answers. Its spam policies cover “attempting to manipulate generative AI responses in Google Search,” and a new optimization guide warns that “seeking inauthentic ‘mentions’ across the web isn’t as helpful as it might seem.” Google’s September 2026 spam update, its fourth of the year, finished rolling out on October 8, according to its status dashboard.

Here is what has been documented, how much damage it can do, and what a brand can do about it.

The tactics, and the evidence

  • Mass AI content. Google calls it scaled content abuse: many pages made “for the primary purpose of manipulating search rankings and not helping users.” AI use itself is not banned. AI-written articles are about half of new English-language articles, Axios reported, but SEO agency Graphite says they “largely do not appear in Google and ChatGPT.”
  • Fake personas and reviews. In 2023, Futurism found Sports Illustrated product reviews under authors with AI-generated headshots; the publisher blamed a third-party partner. In July 2026, Google’s review markup rules added a ban on fake or undisclosed incentivized reviews.
  • Cloaking for AI agents. Security firm SPLX showed a server feeding AI crawlers a different CV, which moved a test candidate from last place to first in an AI-run shortlist.
  • Hidden prompts. Microsoft found 31 real companies, including marketing agencies, hiding instructions in “Summarize with AI” buttons that told assistants to remember them as authoritative sources.
  • Schema tricks. Google says no special markup is needed for AI features, and misleading structured data can cost a page its rich results.

How easy is it to poison an AI answer?

Easier than it should be. Researchers at Anthropic, the UK AI Security Institute and the Alan Turing Institute found that “as few as 250 malicious documents” can plant a backdoor in a model during training, though the one they tested was a narrow trigger that produced gibberish (Anthropic). Retrieval looks like the bigger risk for search: in lab tests for the PoisonedRAG study, five planted texts per question hijacked answers 90% of the time.

In a live test, Ahrefs, which sells AI-tracking tools, invented a brand, seeded conflicting fake stories about it, and found Gemini and Perplexity repeated the planted claims in about 37% to 39% of answers, while ChatGPT stayed under 7%. One critic, SEJ’s Roger Montti, noted 49 of its 56 prompts were leading questions about a brand with no history. AI browsers add another door: at Black Hat USA in August 2026, a Brave researcher showed that every AI browser he tested could be hijacked by hidden instructions, Dark Reading reported.

Check what AI says about your brand

There is no single AI index, so check each assistant. Run the same 10 to 20 prompts every month in ChatGPT, Gemini, Google’s AI Mode, Perplexity and Copilot: your brand name, “is it legit,” pricing, comparisons with competitors. Log the answers and the sources cited. Bing Webmaster Tools’ free AI Performance report shows which of your pages Copilot cites.

In the Ahrefs test, only ChatGPT leaned on the brand’s official FAQ, citing it in 84% of answers. The other tools picked the detailed fakes over its vague denials. Ahrefs’ advice: fill every gap with specific, official facts.

Report it, then outbuild it

File spam through Search Console; since April 2026, Google says reports may lead to manual actions. Flag a bad AI Overview with the thumbs-down and “Report a problem.” In ChatGPT, use the thumbs-down or OpenAI’s content-report form. Keep screenshots, prompts, dates and sources.

Then compete on the basics. Google says its AI features are “rooted in our core Search ranking and quality systems,” so crawlable pages, real authors, accurate markup and original facts still matter most. Our posts on earning AI citations, ChatGPT’s citation patterns, SEO benchmarks and zero-click search cover the rest. We handle this through our SEO, reputation management and brand protection work.

Frequently Asked Questions

What is black-hat GEO?

It is any deceptive attempt to change what AI answer engines say or cite: mass-produced AI pages, fake authors and reviews, cloaking that shows AI crawlers different content, misleading markup, hidden prompts and planted mentions. Since May 2026, Google’s spam policies define spam to include attempts to manipulate generative AI responses in Google Search, so these tactics face the same penalties as classic web spam, from lower rankings to manual actions.

Does Google penalize AI-generated content?

Not for being made with AI. Google’s policy targets purpose and value: scaled content abuse means many pages generated mainly to manipulate rankings rather than help users, whether written by people or by AI. Google says AI can help with research and structuring original content. Pages produced with little effort, originality or added value are the ones at risk, and Google ran four spam updates in 2026.

Can someone poison AI answers about my brand?

To a degree. Research shows planted documents can sway retrieval-based AI answers, and in an Ahrefs test Gemini and Perplexity repeated seeded falsehoods about an invented brand in about 37% to 39% of answers, while ChatGPT stayed under 7%. Real brands with a history may be harder to fool, though that was not tested. The best defense is a detailed, official page of facts that assistants can find and cite.

How do I report AI spam or a false AI answer?

For Google, use the spam report form in Search Console; Google says since April 2026 such reports may lead to manual actions. For a wrong AI Overview, click the thumbs-down icon and choose “Report a problem.” In ChatGPT, thumbs-down the answer or use OpenAI’s content-report form. Also publish a clear correction on your own site, check which sources the answer cited, and ask those publishers to fix errors.