Everyone spent late June 2026 watching the June spam update roll out — refreshing dashboards, comparing tracking tools, arguing about who moved. It was the loud event. But the loud event wasn’t the important one. Five weeks earlier, on 15 May 2026, Google made a quiet change to the wording of its spam policy that will shape SEO far longer than any single update: for the first time, it named the manipulation of generative AI responses as spam. No rollout, no dashboard entry, just a revised sentence in a policy document. And that revised sentence is the framework every future enforcement action will be judged against. The update got the headlines; the policy change is the story.
For anyone doing SEO or link building in 2026, understanding this shift matters more than tracking any one rollout, because it redraws the line between legitimate optimisation and punishable manipulation — and it does so in a way most of the “AI SEO” industry hasn’t fully absorbed. Here’s what actually changed, why it’s bigger than it looks, and what it means for how you work.
What The Policy Now Says
The change is a single clause, but a consequential one. Google’s definition of spam previously centred on the familiar idea of manipulating Search systems into ranking content highly. The revised wording adds a second target: attempting to manipulate generative AI responses in Google Search. In plain terms, trying to game your way into AI Overviews and AI Mode — the AI-generated answers Google now shows above the classic results — is now spam, carrying the same demotion-or-removal risk as classic ranking manipulation. Repeated or egregious violations can trigger the same automated and manual actions, up to deindexing.
Reported examples of the newly-named tactics make it concrete: recommendation poisoning (structuring content, or planting text, to instruct AI systems to treat a specific site as an authority) and biased best-of listicles engineered purely to be cited in AI answers. These are the manoeuvres of the emerging “generative engine optimisation” or GEO industry, which grew up promising to force brands into AI citations. As of 15 May, the deliberately manipulative end of that toolkit sits inside the spam boundary.
Why It’s Bigger Than The June Update
A spam update is an improvement to detection — Google tuning the engine to catch what the policy already prohibits. A policy change is different: it redefines what the engine is allowed to hunt for. The June update enforced existing rules better. The May change expanded the rules themselves. That’s the more fundamental move, and it’s why it will outlast the news cycle.
Consider the sequence Google ran, because it’s a template. On 15 May it named the abuse in writing. On the same day it published its first official AI optimisation guide — the legitimate playbook — so it released the rule and the sanctioned path in the same breath. Then, a little over a month later, it improved the detection engine with the June update. Name the abuse, publish the clean path, tune the enforcement: that’s the pattern to expect for AI surfaces going forward, and it’s a sharper, faster cadence than the old quarterly rhythm described in a decade of Google spam enforcement. The enforcement machine is iterating faster, and the May clause is the new rule it’s iterating on.
The Part Most People Miss: There’s No Separate “AI Spam” Rulebook
Here’s the nuance that actually matters for how you work, and it’s the opposite of how the change is often framed. Google did not create a new, separate category of AI spam with its own rules. It folded AI manipulation into the existing spam framework. The revised sentence simply extends the same logic — don’t deceive users, don’t manipulate the systems that decide prominence — to a new surface. Prominence used to mean ranking position; now it also means being cited in an AI answer. Same principle, wider scope.
That framing carries a direct, money-saving implication. If AI visibility is governed by the same rules and the same systems as classic ranking, then there is no separate GEO channel to game — and much of what’s sold as GEO is therefore unnecessary. Google has said plainly that the things a lot of vendors push for AI visibility — dedicated machine-readable files, content chunking, AI-specific rewriting, manufactured brand mentions, special schema — are not needed to appear in its generative results. The winning inputs for AI citation turn out to be the same inputs that win classic rankings.
This is well supported by the data. Analysts studying what actually predicts AI citation keep landing on the same finding: classic search visibility is one of the strongest predictors of whether an AI system cites you — in one widely-cited analysis, second only to basic URL accessibility. In other words, the work that ranks you is the work that gets you cited. That’s the throughline of link building in the AI search era: the surface changed, the fundamentals didn’t, and the shift from ranking to citation rewards the same genuine authority it always did.
Where The Line Actually Falls
So what’s prohibited and what isn’t? The distinction the policy draws is about intent and effect, not about tools. A few clarifications worth internalising:
- AI-assisted content is still fine. Google did not ban AI-written content. It targets content built to deceive or manufactured at scale to manipulate, regardless of whether a human or a model wrote it. Original, genuinely useful AI-assisted content is not the target; scaled filler with no value behind it is.
- A genuine comparison is fine; a rigged one is spam. An honestly useful best-of list is legitimate. A listicle engineered purely to steer AI citations toward a site, with the ranking predetermined by who benefits, is the newly-named abuse. The tactic is identical on the surface; the intent is the difference.
- Planting instructions for the model is spam. Recommendation poisoning — embedding text designed to tell an AI system to treat your site as authoritative — is squarely over the line. This is the AI-era equivalent of keyword stuffing: content written for the machine, not the reader.
- Real expertise, cleanly presented, is the whole game. Strong content, clear structure, and demonstrated expertise carry straight over to AI surfaces. There’s no trick layer on top; the citation goes to the source that genuinely deserves it more often than to the one that engineered for it.
The uncomfortable reality for the GEO gray market is that the manipulative tactics are both newly-risky and, increasingly, hard to make work — while the legitimate path is just … good SEO. That’s consistent with everything in AI visibility and link building: the durable way to show up in AI answers is to be the kind of source a careful system would cite anyway, not to manufacture signals that the SpamBrain system is being trained to discount.
What This Means For Your Work
For SEOs, link builders, and network operators, the takeaways are practical:
- Stop paying for GEO tricks Google says it ignores. Dedicated AI files, chunking, AI-rewriting, manufactured mentions — Google has said these aren’t needed for AI visibility. Effort and budget spent there is wasted, not rewarded.
- Treat AI manipulation as the same risk as link spam. The May clause means gaming AI answers carries real demotion and deindexing risk. If a tactic only exists because AI Overviews exist, and it wouldn’t make sense on a normal page, it now reads as spam.
- Keep investing in the signals that work on both surfaces. Genuine content, clean structure, real authority, and legitimate links win rankings and citations alike. There’s no longer a plausible argument for a separate AI strategy built on manipulation.
- Expect the name-then-enforce pattern to repeat. Watch the policy wording, not just the update calendar. The next expansion of what counts as spam will show up as a quiet clause before it shows up as a rollout.
The Bottom Line
The June 2026 spam update was the noise; the 15 May policy change was the signal. By naming the manipulation of generative AI responses as spam for the first time, Google redrew the enforcement line to cover AI Overviews and AI Mode — and, crucially, did it by extending the existing spam framework rather than inventing a separate AI rulebook. That means there is no secret GEO channel to game: the same clean signals that win classic rankings are what earn AI citations, and much of what’s sold as AI optimisation is unnecessary at best and now punishable at worst. The manipulative tactics — recommendation poisoning, rigged listicles — carry the same risk as any other spam, enforced by the same machine-learning detection that has spent years learning to read manufactured signals.
For anyone running a network, the lesson is the reassuring one it usually is: the work that survives is genuine content on properly-hosted, real-looking sites, earning links and citations because they deserve them. Build on PBN Hosting and keep your effort on the fundamentals that win on every surface — because Google just told you, in writing, that the shortcuts aimed at its AI are the next thing it’s coming for.