Google’s AI Spam Policy Clarification: What Changed for SEO Teams

The policy changed its wording, not the definition of a legitimate SEO project

Google expanded the opening description of spam in its Search spam policies. Previously, the definition focused on deceiving users or manipulating Search systems into ranking content prominently. It now explicitly includes attempts to manipulate generative AI responses in Google Search. The documentation says its policies apply to all web search results, not only traditional blue links.

On the same date, Google published a separate resource about optimisation for generative AI in Search. The two publications need to be read together: ordinary SEO practices still matter, while trying to manufacture prominence in an AI answer does not create an exception to spam rules.

What Google did not establish: the documentation change did not identify a newly deployed detection model, a rollout schedule for AI-specific penalties or named websites sanctioned because of the revised wording. The general spam policy describes possible ranking consequences and manual actions. Those are enforcement powers, not evidence that an AI-specific penalty had already been applied on May 15.

Where the line becomes difficult: content quality versus answer manipulation

Terms such as recommendation poisoning and prompt injection appear in industry discussion, but neither is a separately named violation in Google’s May 15 policy clarification. They can describe activities worth investigating. Whether a particular page violates policy depends on the actual content and behaviour, not the label attached to the tactic.

A comparison page that recommends its own product

A software vendor may publish a comparison that includes its own product. The format alone does not establish a violation. A useful editorial review explains what was evaluated, where the evidence came from, the limitations of each option and any material commercial relationship. A page presented as independent research becomes more concerning when its conclusions were predetermined, competitors are misrepresented or the recommendation is supported by fabricated evidence.

The first action is to review claims and disclosures, not to delete every listicle. Nor should a change in AI Overview citations be treated as proof that Google has classified a particular page format as spam.

Instructions designed to steer an AI answer

Instructions hidden in source content to make an AI system endorse a brand are a different proposition from ordinary product details, structured data or text written for people. Such tactics warrant scrutiny under Google’s existing rules, including hidden text, cloaking and other forms of manipulation where relevant. But the May 15 clarification did not publish a new, exhaustive list of prohibited prompt patterns. An editor should examine what the material does before claiming an official violation.

Publishing pages for every possible AI prompt

A large set of near-identical pages created chiefly to capture prompt variations presents a more directly documented risk. Google’s subsequent generative AI search guide specifically warns against generating pages for many query or fan-out variations primarily to manipulate rankings or AI responses. That falls within its existing scaled content abuse policy. AI-assisted drafting itself is not the deciding factor; purpose, originality and usefulness matter.

How I would prioritise the audit

For a content operation, an indiscriminate rewrite is the wrong first response. The narrower task is to identify pages and workflows whose claims or production methods cannot be justified. This is an editorial and operational assessment, not a claim that Google has investigated the site.

1. Identify potentially affected work. Start with AI visibility experiments, commercially motivated recommendation pages, bulk-produced prompt-targeting pages and vendors supplying paid mentions. An ordinary explanatory article about AI Overviews is not automatically in scope.
2. Check how each claim was produced. Can an editor trace a ranking, recommendation or quoted statement to evidence? Are comparisons based on a consistent method? Is the site’s ownership or affiliate relationship made clear where relevant?
3. Separate policy conflicts from editorial weaknesses. Hidden manipulation and scaled content abuse require a different response from a thin but honestly written comparison. A weak page may need better sourcing; that does not make it a confirmed spam violation.
4. Fix the smallest defensible set first. Remove manipulative instructions or invented endorsements, correct unsupported claims and combine genuinely overlapping pages. Preserve useful, independently justified information. Document the changes and the reason for each decision.
5. Check the outcome without inventing a cause. Use Search Console, landing-page data and editorial records to see what changed. A ranking drop alone cannot establish that this May 15 clarification caused an enforcement action.

MOCOBIN’s content inventory workflow provides the durable page-level process; this news analysis concerns what the May policy clarification changes in that review. For multilingual sites, apply the same evidential test to each language version. A translated recommendation should not acquire unsupported local claims merely because a publishing template calls for them.

Enforcement and performance are two different things to monitor

Google says it detects spam through automated systems and, when needed, human review. Its Search Console Manual actions report is the place to check for a confirmed manual action. If that report shows no issue, it does not prove that every page complies with policy, nor does it explain every loss of traffic. A manual action, an algorithmic change and a decline in AI citations are different observations.

There is now a separate visibility measure as well. Google announced dedicated generative AI performance reports in Search Console on June 3, 2026, and said they were available worldwide from August 31. These reports show how pages appear in generative AI features, but visibility data cannot establish why Google did or did not feature an individual source. It is not a spam-enforcement report.

Site owners should also distinguish policy enforcement from the Search generative AI control, available worldwide since August 31, 2026. This Search Console setting governs whether a property can appear in covered generative AI features. Excluding a site is a publisher choice, not evidence of a spam action. Check that setting before interpreting a disappearance from AI features.

Before purchasing an AI-placement service, ask the provider to distinguish measurable work on your own pages from predictions about Google’s systems. Google’s third-party SEO guidance warns that external tools do not have access to Google’s internal ranking or AI systems and cannot guarantee performance. That is a stronger basis for assessing a vendor than a promise of a particular citation position.

The decision after the clarification

The May 15 update closes a documentation ambiguity: trying to manipulate Google’s generative AI responses falls within Search spam policy. It does not turn legitimate AI-focused SEO, product comparisons or AI-assisted writing into prohibited activities by default. For an SEO team, the sensible order is to verify the method, correct genuine manipulation or unsupported claims, and only then interpret visibility changes.

For the underlying distinctions between manual actions, ranking changes and recovery steps, see MOCOBIN’s separate guide to Google Search penalties. This article is an analysis of a dated policy clarification, not a replacement for the broader technical guidance.