Google is using AI to help detect more spam, while scaled content is becoming a larger problem than link spam, according to ROAST’s account of Search Central Live Deep Dive in Barcelona. The report points to an enforcement challenge created by the volume of low-quality material, rather than a blanket rejection of AI-assisted publishing.
The source distinction matters. These are statements attributed to Google speakers at the October 2, 2026 event in an attendee’s recap, not a newly published Google policy or an announcement of a specific upcoming rollout.
What the event report actually says
In ROAST’s day-three recap, John Campbell reports that Google Search Quality Analyst Duy Nguyen described the team’s use of AI to catch more spam and linked recent spam updates to growing content volume. The same account attributes the comparison between scaled content and link spam to Gary Illyes.
The discussion of user feedback has a separate context: ROAST places it in Eric Murillo’s session on Discover, describing feedback as important in developing spam filters. Those passages should not be merged into a claim that every user report directly changes general Search rankings.
The recap gives no public schedule for the next spam update or numerical threshold that publishers can use as a safe production limit.
The published policy focuses on purpose and value
Google’s official spam policies define scaled content abuse around generating many pages primarily to manipulate rankings rather than help users. The policy applies regardless of whether content is produced through automation, human work or a combination.
That is the firmer operational reference for an editorial team. A large publishing operation is not automatically abusive, and an individual AI-assisted article is not automatically spam. The useful review concerns why the pages exist and what they contribute.
Audit the publishing workflow, not only the prose
An editorial response would begin with groups of pages created through the same process. Identify the reader’s problem, the evidence used and the additional value supplied by each page. If hundreds of pages differ mainly in a place name or keyword, inspect whether those differences correspond to genuinely different information.
This is an audit recommendation, not a disclosed Google detection formula. A human approving every draft is useful only if the review actually catches unsupported claims, repetition and missing information. A checkbox alone does not establish quality.
For example, a local service guide can name a city without explaining anything specific about that location. A stronger editorial process would verify the service details and explain the practical choices relevant to the reader. Adding more city pages should follow the availability of that evidence.
Do not turn the warning into a ranking diagnosis
A traffic decline does not prove scaled content abuse. Review the affected pages and measurement evidence before attributing a loss to a spam system. The event report does not provide a test for identifying which mechanism changed an individual URL’s performance.
For AI Search strategy, the same editorial questions are useful when preparing material that assistants may retrieve: can a claim be verified, does the page answer a real question and does it contribute something specific? Those practices do not guarantee a citation or recommendation.
The practical implication is to make expansion accountable. Keep sources, corrections and substantive review connected to each publishing workflow, then judge pages by their contribution rather than the number produced. Google’s reported enforcement direction gives teams another reason to examine that process before scaling it further.