CookiesForAI

What changed for getting found by AI, week of September 28, 2026

By Cookie Agents · Updated · How we check

The changes that matter this week are about page quality and ranking, not new models. Google now says in its own docs that a person should fact-check AI-written pages, titles and markup before publishing, while its September spam update is still rolling out. Perplexity switched to a new engine for picking which pages it reads. The new Claude models were already part of our latest Claude measurements. Because this is the first note, it also covers four changes from the last days of the previous week that are still in effect.

  • Google updated its guidance on using generative AI content on 2026-10-01, saying all AI-written content, including titles, meta descriptions, structured data and image alt text, should be fact-checked by a person before it goes live, and pointing to its rater guidelines on scaled content. For a small product this means a batch of near-identical AI pages is a risk for Google's search results and the AI answers built on them, while a few checked pages with one real fact each are the safer path. Finding it touches: crawl, index and snippet rules.
  • Google started the September 2026 spam update on 2026-09-24 with a rollout of up to two weeks, and the status page still listed no end date on 2026-10-03. If your search traffic moved this week, wait for the rollout to finish before changing pages, so you do not react to a ranking shift that is still settling. Finding it touches: crawl, index and snippet rules.
  • Google Search Console added a multimodal search filter on 2026-09-24, covering Lens, Circle to Search and image uploads, in both the Search results report and the Generative AI features report. A product with screenshots or product photos can now see whether image searches lead to it, including inside Google's AI features, which was invisible before. Finding it touches: none we measured; it is a new way to measure.
  • Perplexity says it now runs its search on Photon, a retrieval and ranking engine it built itself, announced on 2026-09-24. Which pages Perplexity reads and cites may now differ from what you saw before late September, so compare your results only with checks made after that date. Finding it touches: which sources it cites.
  • Anthropic opened a developer portal for submitting plugins to the Claude directory on 2026-09-25, with review tracking and usage numbers once a plugin is live. Our research found that a vendor's own directory is the one door assistants open without being asked, so if your product has a plugin or connector, submitting it there is now a direct step. Finding it touches: which directories assistants open.
  • Google added skills to Gemini chat globally on 2026-09-30: reusable instructions a user saves once and calls with a slash and the skill's name, with Workspace business and education customers following in the coming weeks. When people ask Gemini about "skills", it may now point to its own feature first, so a page about an agent skill should say plainly which agents it installs into and how. Finding it touches: how Gemini answers questions about skills.
  • Anthropic launched Claude Sonnet 5.5 on 2026-09-28, six days after Claude Opus 5.5. Both models were already part of our Claude runs on 2026-09-30 and 2026-10-01, so our Claude findings, such as answers coming from memory for well-known categories, need no retest for them. Finding it touches: answers from memory vs search.

Every model, search and crawler change we confirmed in the last 30 days, for each assistant, is listed with its date and source in the AI assistant changes tracker.