The SEO industry has an ancient tradition: whatever the search algorithm ranks for, that's how content gets produced.
In Google's PageRank era, everyone traded backlinks. In the mobile-first era, everyone built responsive designs. In the Core Web Vitals era, everyone desperately optimized LCP and CLS.
Now it's the AI Overviews era.
BBC reports that someone is systematically manipulating Google's AI search results. Not with black-hat backlinks or keyword stuffing — that's last generation's playbook. The new method is content structure engineering: carefully designing article paragraph structures, citation formats, and data presentation to make Google's AI models prioritize their content when generating search summaries.
New Manipulation, Same Old Logic
Specific techniques include:
- Using highly structured Q&A formats at article beginnings, so AI extracts them directly
- Using specific data formats (tables, lists, statistics) to cater to AI summary extraction preferences
- Embedding citation sources and authoritative expressions that AI models are frequently trained on, right in key paragraphs
This is essentially SEO for AI models — not optimized for crawlers, but optimized for large language models.
The interesting part is that this works precisely because of how LLMs are trained. Models absorbed massive amounts of web content during pre-training, and certain content structures appear more frequently and carry more weight in the training data. People who understand these patterns can produce content targeting them.
This isn't a vulnerability. It's a natural byproduct of how the model was trained.
Google's Counterattack
BBC says Google is "quietly fighting back."
Google hasn't publicly discussed specific countermeasures — which makes sense. If you tell the cheaters what you're catching, they'll know how to get around it.
But we can infer Google's likely strategies:
- Source credibility weighting: giving known authoritative sources higher citation weight, reducing the impact of newly registered sites
- Content pattern detection: using another model to detect "AI-optimized" content patterns, flagging suspicious sites
- Real-time information verification: cross-referencing multiple independent sources to reduce the influence of any single content source on AI summaries
These methods are theoretically feasible. But in practice, each has side effects. Boosting authoritative source weight entrenches existing power structures. Content pattern detection might accidentally flag high-quality structured content. Real-time verification adds latency and cost.
This Is an Infinite Game
The relationship between AI search and AI search manipulation is destined to be an infinite game.
Google improves detection → manipulators adjust strategy → Google improves again → manipulators adjust again.
This cycle won't end. Because as long as AI search results influence traffic distribution, someone will invest resources in influencing them.
The logic is identical to traditional SEO. The tech stack just upgraded from HTML meta tags to reverse-engineered prompt engineering.
The difference is that AI-era manipulation is harder to detect. Traditional keyword stuffing can be caught with regex. But "content structures optimized for AI models" and "genuinely good structured content" — the distinction between those may require human editors to judge. And the last thing Google wants to do is hire human editors to review search results.
My Take
This exposes a fundamental contradiction in AI search: the predictability of AI models is both its strength and its weakness.
AI search works well because it can quickly extract, integrate, and generate answers from massive information. But "extractable" means "manipulable." A team that understands the model's behavior well enough can systematically guide it to output specific results.
What Google can do isn't eliminate manipulation — that's impossible. It's raise the cost of manipulation so that cheating becomes less cost-effective than honestly creating content.
But that requires continuous algorithm iteration and adversarial investment. And as AI models get more capable, the technical barrier for manipulation may actually decrease — more mid-size players can join this game.
One metric worth tracking: changes in the coverage rate of Google AI Overviews in search results. If coverage suddenly drops, it likely means Google has detected large-scale manipulation and is tightening its strategy.
Main sources:
- BBC - Google's AI is being manipulated. The search giant is quietly fighting back
- Hacker News discussion threads