The most practical
AI search writing
you'll find.
Daily reporting on what is actually changing in AI search. Findings first, sources named, written for people who run this work.
Yeşilyurt rebuilds his site so only AI agents can read it
The practitioner replaced his WordPress homepage with a Markdown, API-first version served to about 30 agent families.
Graphite asks converting users which prompt they typed into which LLM
Ethan Smith reports 30% of inbound leads self-report AI as the channel, with Claude heavily underreported by last-touch.
Posting new URLs to Reddit got them indexed in six minutes, Bobbink says
A 32-URL test by Jan-Willem Bobbink used Google's real-time pipes into Reddit and X to skip the sitemap wait.
ChatGPT now flags unverified vendor claims six times more often after GPT-6 switch
David Konitzny's analysis of nearly 3 million chats shows warnings about listicle rankings jumped from 0.56% to 3.46% within two days of the default change.
Outbound links in AI answers rarely match the sources they cite, Konitzny finds
A four-week study of nearly 9 million answers shows citation URLs and clickable links diverge sharply, especially on Google's AI surfaces.
ChatGPT replaces citations with direct homepage links in 42% of answers
Tomek Rudzki's one-million-prompt test finds the new links carry no referrer, hiding AI traffic as direct visits.
ChatGPT stopped tagging outbound links on Oct 9, inflating GA4 Direct
Ethan Smith estimates a ~45% drop in ChatGPT-attributed organic sessions across sites he tracks, with the traffic landing in Direct instead.
GPT-6 reverses GPT-5.6 search patterns, Konitzny finds in 2.2M responses
Comparison queries, listicles and editorial domains climb again after the default model switch.
Google is poisoning 8% of rank tracker SERPs, Nozzle finds
Derek Perkins says Google now serves normal-looking results pages with swapped rankings to suspected scrapers.
Fake-stats site pulled 72K ChatGPT-User requests in 30 days
Metehan Yeşilyurt's StateGlobe experiment shows LLMs reach for structured numbers when their training data falls short.
Anthropic's new usage policy bans using Claude to seed AI answers with hidden-origin content
An update announced on 8 October and effective 12 November adds search engines and AI tools to Claude's prohibited uses.
ChatGPT's GPT-6 system prompt tells the model to rank sources by claim fit, freshness and authority
Tomek Rudzki surfaced the 128k-character system prompt and published the web tool call section on GitHub.
Google AI Overviews embed external links in 26% of answers, up from near zero in August
In-text links jumped from 0.09% in late August to 26.19% by 20 September, with English markets seeing them most.
Google tests AI Overview links that open AI Mode instead of websites
Links at the bottom of some AI Overviews now trigger follow-up prompts inside AI Mode rather than sending clicks to publishers.
A fake stats site pulled thousands of LLM citations for seven months
Metehan Yeşilyurt's made-up statistics site collected 330+ referring domains and over 10,000 ChatGPT sessions while Google deindexed it inside a month.
ChatGPT routes web searches through four separate pipelines, Konitzny says
The standard search_query sits alongside image, business and product fan-outs, each running on its own backend.
Google expands AI Overviews on branded queries, inflating GSC impressions
Lily Ray reports rival sites now citing inside brand-name AIOs register as #1 rankings in Search Console.
Google patent describes fan-out queries built from embeddings, not text
Roger Montti flagged a Google filing outlining R4T-Diffusion, a trained model that would generate fan-outs directly from embeddings.
ChatGPT reads YouTube descriptions, not transcripts, when it retrieves a video
David Konitzny, analysing the ChatGPT retrieval leak, finds the model receives the full description but no transcript when it pulls a YouTube video.
Google is indexing 12,000 spam pages built on public Gemini Notebooks
Black hat operators are publishing NotebookLM pages on peptides, porn and coupon codes, and Google is ranking them.
YouTube loses 91% of its ChatGPT citation share versus August
David Konitzny's tracking shows several major domains cited far less often, though retrieval may be unchanged.
External links in Google AI Overviews jump from near zero to over 25%
Peec AI data shared by Malte Landwehr shows a sharp rise in outbound linking inside AI Overviews.
Peec AI publishes ChatGPT's internal retrieval engine list from Yeşilyurt leak
A GEO research team has released the full inventory of search engines ChatGPT calls behind the scenes, including dozens of Labrador variants for Reddit, news, finance, legal and medical queries.
ChatGPT now cites LinkedIn posts almost 5x more than Pulse articles
David Konitzny's URL-type breakdown shows a near-complete inversion across 2026.
Landwehr: under 5% of AI Search practitioners sit in communications
A Berlin practitioner argues brand perception in LLMs needs comms teams, not just marketing.
Konitzny maps 69 engines behind ChatGPT retrieval, 48 tagged internal
A registry pull shows OpenAI's own index handles every specialist vertical, while external engines cover only web, news and images.
Landwehr turns signup questions into a live prompt corpus
A second question on the signup form captures the exact prompts that brought users in, then feeds them to a visibility tracker.
Listicle share fell because ChatGPT's source pool grew 75%, Konitzny finds
Indexing absolute weekly volumes against a July baseline separates a dilution effect from a real retrieval drop for Reddit.
Malte Landwehr on why a competitor's 1,100% SEO jump did not last
A cautionary tale from 2021 that the former SEO lead now sees repeating in AI Search.
Specific figures beat vague copy in every one of 60 GPT-5.4 trials
John Lawson ran 15 paired sources through GPT-5.4 and the citable version won 100% of the time.
Google is paying some publishers when their content shapes AI answers
The AI contribution pilot pays on judged influence in AI Overviews, AI Mode and Gemini, not on raw usage.
Suganthan Mohanadasan makes Claude Code pay to crawl his site using x402
A working proof of concept revives HTTP 402 and turns crawler access into a priced transaction.
Google AI Mode cites 72% fewer sources for logged-out users since mid-August
David Konitzny's weekly tracking shows averages collapsing from 15 to 4 sources per answer across seven markets in three weeks.
ChatGPT fetched 223 pages and cited 16 on one prompt, Yeşilyurt logs
A stream capture from Metehan Yeşilyurt shows where the drop happens, and it is not at the crawl.
ChatGPT flags 'inauthentic mentions' on self-ranked listicles
A screenshot from Metehan Yeşilyurt shows a mechanism that could explain the citation drop Tomek Rudzki measured three weeks earlier.
Turley testified OpenAI planned its own index from day one
Tomek Rudzki surfaces sworn 2025 testimony that reframes the Bing licensing story as a bridge, not the destination.
