Google Has Declared War on Bought AI Citations

On 15 May 2026, Google quietly updated its spam policy documentation. One line changed. The definition of spam now explicitly includes “attempting to manipulate generative AI responses in Google Search”. In June, Google followed up with formal guidance on optimising for AI search, including a direct warning against manipulating or buying citations.

Most of the industry read it as housekeeping. It is not. It is the formal end of the grey period in AI search, and it puts a meaningful share of the “GEO” and “AI visibility” market on notice. Any brand that has bought its way into AI Overviews, AI Mode answers or LLM citations over the past 18 months now holds the same risk profile as a brand that bought links in 2011.

We have watched this exact cycle play out before, with links. The brands that came through Penguin untouched were the ones whose links existed because they had earned them. The same will be true here.

What changed in Google’s spam policies

Until May, there was a documentation gap. Google’s spam policies governed rankings. AI Overviews and AI Mode sat in an ambiguous space where nobody could point to a written rule that said manipulating citations was a violation. Plenty of vendors built businesses inside that ambiguity. The update closes it. Three things now sit on the record.

The practical consequence is the one that matters: a site demoted for any of these violations is excluded from the AI citation pool, not just the rankings.

The tactics now classed as spam

The last 18 months produced a wave of agencies and tools selling AI visibility as a product. Some of that work is legitimate. A lot of it is not, and the people buying it often cannot tell the difference. The illegitimate version follows a recognisable pattern. Each of the following is now, in writing, a spam violation:

• Paid placements on listicle sites that LLMs are known to cite
• Networks of “best of” roundup pages built on expired domains
• Scaled AI-generated content seeded across third-party blogs to manufacture brand mentions
• Markdown shadow sites served only to AI crawlers such as GPTBot

Several of these were already producing visible pollution. Industry analysis of AI Overview citation quality over the past year found expired domains, thin affiliate sites and scaled AI content appearing in AI answers more often than in the corresponding organic results. Google has noticed. Enforcement is expected to ramp through the coming quarters, not arrive in one dramatic update.

If you have engaged a vendor whose deliverable is “guaranteed citations” or “placements in AI answers”, ask them one question: how? If the answer involves paying a publisher, a network or a platform for the mention, you have bought links with a new label on the invoice.

The Penguin parallel

In 2012, Penguin landed and a decade of bought-link equity turned into a decade of disavow files, manual actions and recovery projects. The brands hit hardest were not the spammers themselves. They were legitimate businesses that had outsourced their link building to vendors who never explained the risk, and who were long gone when the penalty arrived.

The structure of the AI citation market in 2026 is identical to the link market in 2010. New surface, no written rules, vendors selling shortcuts, buyers who assume that because everyone is doing it, it must be safe. The May update is the moment the rules went on the record. What follows, on past form, is a period of quiet algorithmic enforcement, then something with a name and a rollout date.

Why regulated brands have the most to lose

For most businesses, losing AI citations is a traffic problem. For operators in gambling, finance, legal and health, it is worse, for three reasons.

YMYL scrutiny stacks on top of spam enforcement

Google already holds gambling and finance content to a higher evidential standard. A regulated brand caught manipulating citations is not just violating a spam policy, it is confirming the exact trust deficit that YMYL systems are built to detect. Recovery from that combination is slower and less certain than either problem alone.

The citation pool is the acquisition channel

In verticals where paid channels are restricted, regulated or priced into absurdity, organic visibility carries a disproportionate share of acquisition. Exclusion from the AI citation pool in a market where a growing share of player and customer research starts in an AI answer is not a dip in a traffic chart. It is a structural hole in the funnel.

Regulators are watching the same signals

Licensing regimes increasingly examine marketing practice, not just product compliance. A spam penalty for deceptive visibility tactics is precisely the kind of finding that surfaces in due diligence, affiliate audits and licence reviews. The compliance cost of a shortcut is no longer limited to Google.

The legitimate citation playbook

None of this means AI visibility work is dead. Google’s own guidance is explicit that improving clarity, strengthening expertise signals and earning credible citations remain legitimate. The line is intent: content and mentions that exist for users survive, content and mentions manufactured to game a retrieval system do not. The durable playbook has four parts.

01  Original data and reporting

LLMs and AI Overviews cite sources that contain information nobody else has. Proprietary research, market data, named expert commentary. This is the single highest-yield citation asset and the hardest to fake.

02  Entity clarity

Consistent, unambiguous information about who you are, what you do, who you serve and why you are credible, structured so that machines can extract it. Most regulated brands are still invisible at the entity level, which is why thin affiliates outrank them in AI answers.

03  Earned coverage in sources LLMs already trust

Digital PR into publications with genuine editorial standards, not placement networks. The mention has to survive a human editor to be worth anything to a machine.

04  Compliance architecture as a ranking asset

Licensing signals, regional legal pages and responsible practice content are exactly the trust evidence that YMYL systems reward. For regulated brands, the compliance layer is not overhead. It is the moat.

The enforcement window and next steps

There is a short period, likely two to four quarters, between this policy clarification and serious enforcement. Brands carrying manufactured citation footprints should use it to audit and unwind. Brands that have done the work properly should use it to extend the gap, because every competitor that gets filtered out of the citation pool makes the remaining citations worth more.