Control, measurement, and policy all moved this week. We got a first-party switch for AI surfaces, fresh reporting to see where those surfaces show your pages, hard numbers on how often engines cite and stick with sources, and what happens when a major forum slams the door. On top, Google expanded its spam frame to cover AI answers. Start with the new Search Console tools, because they change what you can opt into and what you can track, and because they affect how stakeholders will judge exposure.
Google adds site controls and insights for AI answers
Google said website owners now have a Search Console toggle to include or exclude their links and content from AI Overviews, AI Mode, and AI features in Discover. The company said this setting is not used as a ranking signal outside those features. A new insights report shows impressions for your content in these AI experiences, and breaks out pages and countries. Google described the rollout as global and complete as of Aug 31, 2026. For teams fielding questions about AI visibility, this is a product change, not a policy carve-out: control is scoped to the AI features named above, and reporting is confined to those impressions.
What it changes: Set a position per site, brief stakeholders on scope, and watch the impressions report before changing content bets.
Cross-engine tests show citations unstable and fragmented
Ansengine ran a methods-first measurement of 10,639 answers across 9 AI search surfaces from July 11 to Sept 2, capturing 77,526 citations from 14,477 domains. Perplexity cited a source on nearly every answer and kept 89% of sources stable run to run; Google AI Mode retained 24%. Agreement across engines was thin—about 1 source in 5 overlapped. Six in ten citations came from outside the 200 most-cited domains. The researchers published CSVs for replication. For anyone testing supplier exposure, these numbers quantify what you feel in the field: frequent churn, low consensus, long-tail coverage.
What it changes: Track by engine and rerun often; judge success on ranges, not single runs.
AI Mode cites google.com far more than AI Overviews
BrightEdge reports that google.com appears in 44.9% of citations in Google AI Mode versus 9.9% in AI Overviews—about 4.5× higher—based on a U.S. prompt set in consumer product and finance queries measured Jan–Aug 2026. To reduce noise, they kept only prompts with stable returns across repeated runs and reported citation share rather than raw counts. The gap predates August redirect news. This is a vendor study, not a platform statement, but the size of the spread makes the surfaces not comparable on linking behaviour.
What it changes: Segment your reporting: treat AI Mode as a distinct surface with heavier google.com self-citation.
Reddit’s crawl block coincides with ChatGPT’s Reddit citations hitting zero
BrightEdge examined ChatGPT citations after Reddit blocked crawling in August. In an Education prompt panel, reddit.com presence fell week over week from 17.2% to 3.3% to 0.0%, and their account-wide view also collapsed. Some of the gap was filled by government, university, and professional-association sources, though several spikes faded by week 37. Methods and limits are documented, with the usual caveats on prompt sets and stability. The direction is clear: cut off access and your citations can disappear fast, with replacement sources shifting as the model or retrieval adjusts.
What it changes: If you restrict crawling, plan for loss of presence in AI answers and track which institutions replace you.
Google says spam policies also apply to AI answers
Google updated its Search documentation to say that the web spam policies also apply to generative AI responses in Search. That means attempts to manipulate AI answers—not just classic rankings—can trigger automated or manual actions. The entry frames this as aligning treatment of AI answers with other Search results. The change sits in a docs changelog rather than a blog post, but the policy effect is direct: spam enforcement can reach tactics aimed at AI surfaces.
What it changes: Review any AI-answer targeting schemes under spam policy; if it would be spam in web results, expect the same risk here.