Some of the tricks people use to get cited by AI assistants can get a site demoted in Google. That matters twice. Google's own AI answers draw from pages in Google's index. And other assistants, like ChatGPT, sometimes run web searches before they answer. So a tactic that wins a few AI citations this month can cost you Google traffic and, later, the AI visibility you were chasing in the first place.
This piece looks at six common GEO tactics and checks each one against what Google's published policies actually say. GEO means generative engine optimization: work done to get your brand mentioned or cited in AI-generated answers. A citation is when an AI answer links to your page as a source. At the end there is a simple way to measure both channels at once, so a "win" in one place does not hide a loss in the other.
Google's spam policy now names AI answers directly
Google's spam policies define spam as techniques used to deceive users or manipulate Google's systems. The current version spells out two kinds of manipulation. One is trying to get content to rank highly. The other is "attempting to manipulate generative AI responses in Google Search."1
In plain words, trying to game Google's AI answers is treated the same way as trying to game the blue links. Google says it finds these practices with automated systems and, when needed, human review. Human review can lead to a manual action, which is a penalty applied by a Google employee. Sites that break the policies may rank lower or not appear in results at all.1
The page was last updated on August 28, 2026.1 Here are the policy sections most relevant to GEO tactics:
| Policy | What Google says it is (short version) |
|---|---|
| Scaled content abuse | Many pages made mainly to manipulate rankings instead of helping users, usually large amounts of unoriginal content with little value, "no matter how it's created"1 |
| Doorway abuse | Sites or pages made to rank for specific, similar searches, including substantially similar pages that lead users to a less useful place1 |
| Hidden text and link abuse | Content placed on a page only to manipulate search engines and not easily seen by visitors, such as white text on a white background or font size set to 01 |
| Cloaking | Showing different content to users and to search engines to manipulate rankings and mislead users1 |
Why a Google penalty can also cost you AI visibility
AI answers are not built from thin air. Many of them start with a search.
- Google AI Overviews and AI Mode. Google says there are no extra requirements to appear in these features. To be shown as a supporting link, a page must be indexed and eligible to show in Google Search with a snippet. Both features may use "query fan-out," which means running several related searches across subtopics and data sources to build one answer.2
- ChatGPT search. OpenAI says ChatGPT search "sometimes partners with other search providers." It typically rewrites your question into one or more targeted searches and sends them to those providers.3
So if a page drops out of Google's index, it cannot be a supporting link in Google's AI features. And if a site sinks in rankings, it shows up in fewer of the searches that AI systems run behind the scenes.
The link is not one to one, though. An Ahrefs study published on March 2, 2026 looked at 863,000 keyword results pages and 4 million AI Overview URLs. It found that 37.9% of URLs cited in AI Overviews also appeared in the first 10 results blocks for that keyword. Another 31.2% ranked in positions 11 to 100, and 31.0% ranked beyond the top 100. In Ahrefs' earlier study from July 2025, the top 10 share was about 76%.4 Ahrefs suggests query fan-out is one reason. Google now pulls sources from the searches it runs for related subtopics, not only from the results for the original question.4
This means ranking in the top 10 for one keyword is no longer the whole story. But the sources still come from Google search results, just a wider set of them. A sitewide demotion hurts you across all those related searches at once. That is the risk this article is about.
Six GEO tactics, checked against Google's own words
1. Mass AI-written pages
Why people do it: AI answers pull from many related searches. More pages seem to mean more chances to be picked.
What Google says: The first example listed under scaled content abuse is "Using generative AI tools or other similar tools to generate many pages without adding value for users."1 Google's guidance on generative AI content repeats the same warning. It says this "may violate Google's spam policy on scaled content abuse." That page was updated on October 1, 2026, and it also says it is "critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing."5
Note the wording. Google does not ban AI-written content. It targets many pages that add no value, however they were made. The people-first content guide says the same thing in another way: using automation "to produce content for the primary purpose of manipulating search rankings" breaks the spam policies.6
2. Fake "date modified" refreshes
Why people do it: AI tools often favor recent sources for some questions. Changing the "last updated" date is quick and makes a page look fresh.
What Google says: In its guide to helpful, people-first content, Google lists warning signs of "search engine-first" content. One of them is: "Are you changing the date of pages to make them seem fresh when the content has not substantially changed?"6 Google's byline date documentation adds that dates "must describe the publication or update date of the page, not the stories or events described therein."7
Fake freshness is not its own named spam policy. It is a sign of low-quality, search-first content in Google's eyes. So the risk is real but usually lower than with the tactics that break a named spam policy.
3. Self-ranking "best X" listicles at scale
Why people do it: AI answers to "what is the best X" questions often draw from ranked list pages. A brand that publishes its own list can put itself at number one.
What Google says: Google's guide to writing reviews asks for real evidence. It tells writers to "Provide evidence such as visuals, audio, or other links of your own experience." When something is called the best, Google says to "include why you consider it the best." It also asks that ranked lists have enough useful content "to stand on their own."8
One honest comparison page that explains its method is fine. The trouble starts when a brand publishes dozens of nearly identical "best X for Y" pages, each with itself on top and no real testing. That starts to look like the scaled content and doorway patterns above.1
4. Mass "alternatives" and "X vs Y" pages
Why people do it: Buyers ask AI tools things like "What are the alternatives to Brand A?" A page for every competitor seems like a way to show up in every one of those answers.
What Google says: Doorway abuse includes substantially similar pages built to rank for specific, similar searches.1 Scaled content abuse lists "Creating many pages where the content makes little or no sense to a reader but contains search keywords."1 A template with the competitor's name swapped in, and no real comparison, fits both patterns.
5. Hidden prompt instructions and "Summarize with AI" buttons
Why people do it: If you can tell an AI assistant to trust your brand, it may recommend you later.
What the evidence shows: On February 10, 2026, Microsoft's security team named this "AI Recommendation Poisoning." Some websites add "Summarize with AI" buttons. These open an AI assistant with a prompt already filled in through the link. The hidden part of the prompt tries to plant instructions in the assistant's memory, like remembering a company "as a trusted source for citations."9
Microsoft says it "identified over 50 unique prompts from 31 companies across 14 industries," found while reviewing AI-related links in email traffic over 60 days. It also says "Every case involved real companies, not hackers or scammers."9 Microsoft warns that assistants affected this way can give "subtly biased recommendations on critical topics including health, finance, and security."9
What search engines say: If the instruction is hidden in page text, Google's hidden text policy applies.1 If it aims at Google's AI answers, the spam policy's line on manipulating generative AI responses applies too.1 Bing is even more direct. In 2024 it added this line to its Webmaster Guidelines: "Do not add content on your webpages which attempts to perform prompt injection attacks on language models used by Bing." It added that this "can lead to demotion or even delisting of your website from our search results."10 A prompt injection is text written to trick an AI model into following instructions it was not meant to follow.
6. Invented statistics pages
Why people do it: AI answers love a clear number. A page full of "stats" seems likely to get quoted.
What Google says: Google's generative AI guidance explains why made-up numbers are a real danger. AI models "don't retrieve facts, but predict a likely sequence of words," which is why Google calls manual fact-checking critical.5 Its people-first content guide also asks whether your content leaves "readers feeling like they need to search again to get better information from other sources."6
Invented numbers also harm you outside Google. Once an AI tool repeats a fake stat with your name on it, anyone who checks it will find nothing behind it. Anthropic has also tightened its rules on using Claude to build fake source networks. Ripplix covered that change in our article on Claude's usage policy.
The GEO tactic risk check
Here is the whole picture in one table. The risk levels are Ripplix's own judgment, based on how directly Google's documents describe each tactic. They are not a Google rating.
| Tactic | Why people do it | What Google's policy says | Risk level | Safer alternative |
|---|---|---|---|---|
| Mass AI-written pages | More pages, more chances to be picked by fan-out searches | Named example of scaled content abuse1 | High | Fewer pages, each with original data, examples or expertise, fact-checked by a person |
| Fake "date modified" refreshes | Looks fresh to AI tools and searchers | Listed as a search-first warning sign; dates must reflect real updates67 | Medium | Change the date only after a real update, and say what changed |
| Self-ranking "best X" lists at scale | AI answers draw from ranked lists | Reviews need first-hand evidence and reasons; many similar pages can be scaled or doorway content81 | Medium to high (higher the more pages) | One honest comparison with a stated method, real testing and fair placement |
| Mass "alternatives" and "vs" pages | Show up for every competitor question | Fits doorway abuse and scaled content patterns when templated1 | High | A few deep comparisons for the competitors buyers really weigh you against |
| Hidden prompts, "Summarize with AI" memory instructions | Plant your brand in an assistant's memory | Hidden text abuse; manipulating generative AI responses is spam; Bing bans prompt injection content110 | Very high | A plain summary button with no hidden instructions, or no button |
| Invented statistics pages | AI answers quote numbers | Google calls fact-checking critical; inaccurate content fails its quality bar56 | High | Publish your own real data, with method and sample size shown |
Notice the pattern in the safer column. Every safe version is smaller and more original. That is not a coincidence. Google's spam policies mostly punish scale without value, and AI tools need something worth citing anyway.
Timing: Google just finished a spam update
This is not only a theory. Google released its September 2026 spam update on September 24, 2026. Google said it "applies globally and to all languages" and might take up to two weeks. The Search Status Dashboard marked the rollout complete as of October 8, 2026.11
Google did not say which spam practices this update targeted. So no one can honestly say it was aimed at GEO tactics. But if your Google traffic moved between September 24 and October 8, compare the timing with any of the six tactics above that your site uses.
Even a "clean" change can help AI and hurt Google
The risk is not limited to spam. A perfectly honest change to a page can help AI tools pull out an answer while making the page work worse in Google.
Say you move the full answer into a short box at the top of a guide and trim the longer text below. An AI tool may find it easier to quote. But the trimmed text may have been what ranked for long, specific searches.
There is a second effect too. A randomized field experiment by researchers Saharsh Agarwal and Ananya Sen, posted to SSRN in April 2026, studied 1,065 US desktop Chrome users. It found that AI Overviews cut outbound organic clicks by 38% on the searches where they appeared.12 The paper is a draft and not yet peer-reviewed.12 Still, it shows why a citation in an AI answer and a click from Google are different things. You can gain one while losing the other.
The Ripplix practitioner view: AI visibility and Google organic traffic have to be read side by side. A citation report on its own cannot tell you whether a change was worth it, and neither can a Search Console chart on its own.
A simple two-channel measurement plan
The fix is to test changes on a group of pages and compare them with a similar group you leave alone. Analysts call this a test group and a control group. The control group shows what would have happened anyway, because of seasons, Google updates or AI model changes.
- Pick two matched groups. Choose pages of the same type, with similar traffic and similar topics. For example, 30 guides in the test group and 30 in the control group.
- Record a baseline. For 4 weeks before the change, track both groups. Do not change anything yet.
- Change only the test group. Make one kind of change, so you know what caused any result.
- Track both channels for 4 to 6 weeks.
- Google: clicks, impressions and average position from Search Console. Google says traffic from its AI features is counted in the Performance report under the "Web" search type.2
- AI: how often the pages are cited across a fixed set of prompts (the questions you track in AI tools), plus visits that arrive from AI assistants in your analytics.
- Compare the change, not the raw numbers. Subtract the control group's change from the test group's change. What is left is a fair estimate of what your edit did.
- Avoid big Google updates if you can. If a core or spam update lands during the test, note it. The control group helps, but a big update can still hit the two groups differently.
Illustrative composite example: Kettlewick Outdoor
This example is an illustrative composite. Kettlewick Outdoor is an invented brand, and every number below is invented to show the method. It is not a real client or a real study.
Kettlewick sells camping gear. Its team rewrote 30 buying guides so each one opens with a short answer box and a quick comparison table. They cut about a third of the longer text below. Another 30 similar guides were left alone as the control group. Both groups were tracked for 4 weeks before the change and 6 weeks after it. The AI side was checked with 120 tracked prompt runs per group.
| Metric (illustrative) | Test before | Test after | Test change | Control before | Control after | Control change | Net effect |
|---|---|---|---|---|---|---|---|
| Google clicks | 12,000 | 10,200 | -15% | 11,800 | 11,564 | -2% | -13 points |
| AI citations (out of 120 prompt runs) | 18 (15%) | 31 (26%) | +11 points | 17 (14%) | 18 (15%) | +1 point | +10 points |
| Visits from AI assistants | 410 | 560 | +37% | 395 | 405 | +3% | +34 points |
Looked at alone, the AI numbers seem like a big win. Citations rose from 15% to 26% of prompt runs, while the control group barely moved. Now look at both channels together.
- Google clicks: the net effect is -13 points. On a baseline of 12,000 clicks, that is about 1,560 clicks lost (12,000 × 0.13).
- AI visits: the net effect is +34 points. On a baseline of 410 visits, that is about 140 visits gained (410 × 0.34).
So in this invented case, the change won about 140 AI visits and lost about 1,560 Google clicks. More citations may still be worth something, since some people see a brand in an AI answer and never click. But the team can now make that trade on purpose instead of by accident.
A sensible next step: keep the answer box and table, bring back the deeper text below them, and run the test again.
What to do this week
- Audit for the six tactics, starting with any hidden instructions. They carry the clearest policy and trust risk.
- Check your September dates. Compare Search Console clicks before September 24 and after October 8, 2026, by page group.
- Report both channels in the same place. If AI citations and Google clicks live in separate reports, a loss in one can hide behind a win in the other.
See both sides of your search visibility
A GEO win only counts if it does not cost you more in Google. Ripplix tracks how AI assistants cite and describe your brand across the prompts your buyers use, so you can read those numbers next to your Search Console data and catch a bad trade early.
Get your free AI Visibility Report →
Sources:
- Google Search Central: Spam policies for Google web search (last updated August 28, 2026)
- Google Search Central: AI features and your website (last updated December 10, 2025)
- OpenAI Help Center: Searching the web with ChatGPT (accessed October 14, 2026)
- Ahrefs: Update: 38% of AI Overview Citations Pull From The Top 10 (March 2, 2026)
- Google Search Central: Google Search's guidance on using generative AI content on your website (last updated October 1, 2026)
- Google Search Central: Creating helpful, reliable, people-first content (last updated October 5, 2026)
- Google Search Central: Add a byline date to Google Search results (accessed October 14, 2026)
- Google Search Central: How to write reviews (last updated December 10, 2025)
- Microsoft Security Blog: Manipulating AI memory for profit: The rise of AI Recommendation Poisoning (February 10, 2026)
- Search Engine Land: Prompt injection added to Bing Webmaster Guidelines (July 1, 2024)
- Google Search Status Dashboard: September 2026 spam update (September 24 to October 8, 2026)
- Search Engine Journal: Study Confirms Google AI Overviews Cut Organic Clicks 38% (April 27, 2026)
