Claude bans fake AI sources, AI Overviews take 80% of brand searches

The biggest change this week came from an AI lab, not Google: Anthropic now bans using Claude to fake the sources AI answers cite. Google spent the week pushing AI Overviews onto almost every brand search while its own reporting broke: Search Console ran days late and rank trackers got blocked. The spam update is done, and the cleanup looks slow.

1. Anthropic now bans using Claude to fake the sources AI answers cite

TL;DR From November 12, Anthropic's usage policy bans using Claude to seed search and AI answer sources with content that hides who is behind it, to run networks of fake "independent" sites, or to build tools that do either.

Why it matters The previous version, from September 2025, did not mention search engines or AI answers at all. The penalty is losing access to Claude. It lands while the answer engines themselves are getting more skeptical of self-promotion. The "listicle yourself onto the shortlist" play now has a written rule against it.

What people are saying

• Lily Ray on what the new section bans, and that it "feels a lot like the early days of Google's spam policies". Search Engine Journal adds that the update drops automated publishing from the high-risk list.
• Lily also notes that Claude and ChatGPT have been calling out listicles and comparison pages since earlier this year, and shared another ChatGPT disclaimer about self-serving rankings.
• Dejan says small brands that list their competitors make those competitors stronger in LLM training data. He calls it the association paradox.
• On Reddit, an SEO testing local queries found ChatGPT cites aggregator "best dentist" lists, not the practices' own sites.

Do this List every "best X" page on your site that ranks you first. Rewrite the worst one so the criteria are public and a competitor could plausibly win it.

2. AI Overviews went from 26% to 80% of brand searches in five weeks

TL;DR DemandSphere now sees AI Overviews on more than 80% of the branded queries it tracks, up from 26% on September 1. Google has not said why.

Why it matters Last week's Ahrefs number showed the scale. This one shows the speed. A brand search used to be a near-certain click to your own site. Now an answer box, built from whatever pages describe you, sits in front of it.

What people are saying

• Semrush on the DemandSphere numbers, and Search Engine Journal on AI Overviews spreading to most big-brand searches.
• Lily Ray: Google seems "hell-bent on showing as many AI Overviews for as many queries as possible". Barry Schwartz found them even on typos and queries with no results.
• Aleyda Solis on "baggy jeans": searches up 60% year on year, organic clicks down 30%, with organic results in the top 10 cut from 9 to 5 by Shopping packs, a local pack and People Also Ask. No AI Overview at all.
• Jake Ward's case for prioritising Google: AI Overviews show on 5.9 billion searches a day, against about 100 million search-like queries a day on ChatGPT.

Do this Pick your 5 most valuable brand searches. Screenshot each result page today and again in a month. Track how many organic results are left above the fold, not just your position.

3. The spam update is done. Recovery takes months, and manual actions are back

TL;DR Google's September spam update finished on October 8 after nearly 14 days, its longest this year. Manual actions are up, and Google's own recovery timeframes run from 3 months to 2 years.

Why it matters If you were hit, there is no quick bounce. Marie Haynes compares the rise in manual actions to 2011, just before Penguin: Google collects examples by hand, then automates the demotion.

What people are saying

• Search Engine Journal on the update finishing, plus crawl timing ranges and site moves measured in months.
• Chris Long calls Google's new numbers on how long crawling, indexing and core update recovery take a goldmine.
• Lily Ray on recovery in "3 months to 2 years" matching what she sees, and on Marie Haynes' manual action surge. She also found another B2B SaaS blog wiped out with 38 self-promoting listicles.

Do this Check Search Console's Manual Actions tab today. It takes 30 seconds, and most people only look after the traffic is gone.

4. Your SEO data went dark this week

TL;DR Search Console's performance report ran more than 64 hours late, and third-party rank trackers are failing as Google blocks scrapers. AI traffic is still mostly invisible in analytics.

Why it matters You can't judge an update when the data arrives three days late and your tracker shows gaps. Ahrefs found AI search ahead of YouTube as a source of their own signups, second only to Google, once they looked past referral data. Referral numbers alone would have badly undersold it.

What people are saying

• Barry Schwartz on the Search Console delay and on rank tracking struggling. Glenn Gabe: "65 hours and counting".
• Cyrus Shepard's state of AI SEO attribution: direct traffic is up, and nobody can tell you what it is.
• Ahrefs on measuring AI visibility when referral data undersells it, and on AI search ROI, citing Forrester's finding that 94% of business buyers use AI search when buying.
• Brodie Clark's fix for one blind spot: submit each child sitemap, not just the index, so Search Console reports indexing by section.

Do this Add "How did you hear about us?" as a free-text field on your signup or contact form. Self-reported answers are where AI influence shows up first.

5. ChatGPT is reading your subdomains and help docs

TL;DR ChatGPT runs more site: searches now, so help centers, docs and other subdomains feed its answers about you.

Why it matters Those pages are written for existing customers, rarely reviewed by marketing, and sometimes include staging sites. If they describe your product to buyers now, they deserve homepage-level care.

What people are saying

• Lily Ray: take inventory of your subdomains and review the help center, docs and forums for how they frame your product.
• Chris Long: help centers and docs are often a huge source of LLM citations, and Claude's browser extension can draft them by running your tool and taking screenshots.
• Ahrefs studied 66.6 million ChatGPT responses and 15 million fan-out queries to see how it retrieves sources.

Do this Search site:yourdomain.com -www in Google to list what is public on your subdomains. Close anything that should not be there.

By the numbers

From SeekWinners' US shopping-search sample: 2,876 category terms across 12 categories. This is our data.

• 60% of shopping terms show an AI Overview, the same as last week. The jump this month is in brand searches (theme 2), not category searches.
• Reference sites take 21% of all AI Overview citations, more than Shopify stores at 13%. Wikipedia is the #2 cited host in 11 of 12 categories.
• The reference share is highest in Health and Beauty at 30% and lowest in Furniture at 12%. Apparel is where Shopify stores do best, with 19% of citations.
• YouTube is still the top cited host in all 12, with 22% of citations, and 32% in Vehicles and Parts.

Worth a click

• Google wants to turn Discover into an AI concierge, and is hiring teams to generate AI content for it.
• ChatGPT gets GPT-6 and Intelligent UI, so answers can include visuals and interactive tools.
• Google's UGC Fresh Data Program lets forums push content to Search faster.
• 10 SEO MCP servers for Claude, plus Manifold, a pay-per-call SEO data MCP.
• The minimal SEO setup: the Search Console API plus Claude.
• A backlink research prompt that asks for opportunities a small site can realistically get.
• GEO Search Console, a new tool aiming to be Search Console for AI answers.
• Tibo's ChatGPT citation playbook is free for AI to read. Humans pay $10.
• A plain on-page content checklist.

From CiteCycle

Theme 1 is about who gets to be a source, and theme 4 is about how hard that is to see. CiteCycle asks ChatGPT, Gemini, Claude and Google AI Overviews your buyers' questions every week, and records separately whether you were mentioned and whether your pages were cited. It doesn't depend on referral data or a rank tracker scraping Google.