Google has confirmed that its search spam policies now apply to AI Overviews and AI Mode, the AI-generated answers that increasingly sit above traditional search results. The update, made on 15 May 2026, closes a gap that marketers and bad actors had been quietly exploiting.
It matters because AI answers are fast becoming the first thing a searcher reads, and often the only thing. Research has consistently found that people click through to websites less often once an AI summary appears at the top of the page. If those answers can be gamed, so can the reputation of any business they mention.
What Google changed
Google updated its Search Central documentation to state plainly that its spam policies apply to all of Google Search, including generative AI responses. In practice, the long-standing rules set out in Google’s spam policies, covering scaled content, cloaking, hidden text, link spam and site reputation abuse, now explicitly govern what can surface in AI Overviews and AI Mode.
Sites that break them may rank lower or disappear from results entirely, with enforcement handled by automated systems and, where needed, human review. The change does not invent new rules so much as confirm the old ones were never optional just because the output is generated by AI.
How AI answers get gamed
The tactics are simpler than most businesses assume. A BBC journalist spent about 20 minutes writing a fake article that ranked himself the world’s best competitive hot-dog eater among tech reporters, citing a championship that does not exist. Within 24 hours, Google’s Gemini app and AI Overviews were repeating the claim as fact, according to the BBC investigation. ChatGPT did the same; Anthropic’s Claude was not fooled.
That stunt exposed a serious problem. Self-promotional best-of listicles, manufactured brand mentions and artificial citation networks can all nudge an AI model toward a chosen answer, and none of it requires hacking anything, just publishing the right words in the right place. Google’s updated policy names exactly these tactics, which means the shortcut that worked last month is now a liability rather than a growth hack.
Why this matters for Australian businesses
For Australian businesses, the risk runs both ways. A competitor, or a careless marketing agency, could seed misleading content that an AI answer then repeats to your customers, with no easy way to correct the record. Equally, a business chasing visibility in AI results may adopt tactics that now carry a genuine penalty rather than a quiet advantage.
The safer path is to earn AI citations the way Google says it wants them earned: accurate, well-structured content from a credible source. That is the foundation of legitimate Generative Engine Optimisation, and after this update it is the only approach that does not risk a demotion. Visibility built on manipulation was always fragile; now it is openly against the rules.

Not every AI model behaves the same
One detail from the BBC test is worth holding onto: Claude resisted the fake content while Gemini and ChatGPT did not. Different models apply different levels of scrutiny to their sources, so a business’s exposure depends partly on which assistant a customer happens to use.
For teams weighing up tools, those differences are practical rather than academic, and our comparison of Claude and ChatGPT looks at how each handles sourcing and reliability. The wider lesson is not to rely on any single model behaving well, but to keep your own content accurate enough that any model repeating it tells the truth.
A new standard for AI answers
Google’s move signals that AI answers will be held to the same standard as the ranked links they are replacing. For businesses, the takeaway is straightforward. The manipulation shortcuts are closing, and the content that wins in AI search will be the content that would have earned trust without any trick at all. That is a harder road than buying a quick ranking, but a far more durable one, and it rewards the businesses that were doing the work properly all along.
