AI Search Intelligence
Dated briefs on what changed in ChatGPT, Perplexity, Gemini, Google AI Mode and AI Overviews, Claude, Microsoft Copilot and Bing, and AI crawler policy. Each brief links its source where one exists. Editorially independent analysis — no sponsors, nothing sold on this domain.
Latest brief
Latest brief
Latest brief
Latest brief
Latest brief
Latest brief
Latest brief
Latest brief
If OAI-SearchBot traffic spiked this month, blocking it is the wrong move: that costs ChatGPT search eligibility. Five weeks of logs across five domains show why — ChatGPT-User fetches fell 49% while OAI-SearchBot crawls rose 838%, and 87% of that crawl rise came from one domain, a site event, not an engine change.
The Machine Relations Index now publishes per-engine run denominators. Perplexity cited a source in 99.97% of returned runs. Gemini cited one in 48.4%.
Machine Relations' live Fidelity cohort has now graded 8,258 ChatGPT citations at 53.4% supported, confirming the September frozen-sample rate of 54.8%. Perplexity, Gemini and Claude still have no live-cohort rate.
Across 14,447 readable citations in four AI engines, the cited source's name reached the answer text 8.6% of the time. The rate splits by engine.
The Machine Relations Index's raw exclusivity share for all six AI answer engines, then the same number conditioned on citation volume. Among the three engines the Index collected without a gap, ChatGPT and Perplexity keep an elevated exclusive-source share past 5 citations and Claude collapses to 1.1%. The three gapped surfaces cannot be ranked on this measure.
Our 67-domain reach gap between Google AI Mode and AI Overviews is withdrawn: the instrument had collection gaps. The surfaces share 363+ of the top 500.
Every AI visibility tool reports domains cited. Almost none tells you whether learn.microsoft.com counts as microsoft.com. The Machine Relations Index publishes its rule and tested it on every published row: it holds on 23,262 of 23,280, and the 18 edge cases show exactly where any tool's domain count drifts.
A dated record of what Claude cites. Entry one: Claude reaches 463 of the 500 most-cited domains, second only to Perplexity among the engines measured.
A dated record of what Perplexity cites. Entry one: Perplexity reaches 490 of the 500 most-cited domains, the widest head overlap of any engine measured.
An 89-day first-party panel of 35 queries across five AI surfaces: when an engine keeps a domain, it cites the exact same URL 72% to 91% of the time.
The beat is machine-mediated discovery. Paralax reports on the engines that retrieve, cite, and rank sources — Google AI Mode and AI Overviews, Perplexity, ChatGPT search, Claude, Gemini, Microsoft Copilot and Bing — and on the infrastructure around them: crawler policy, robots rules, citation architecture, and answer-engine measurement.
Each dispatch asks what changed in how machines find information, and what that change does to publishers, brands, and the public record. A brief ships when a change is verified, not to fill a day. Coverage is dated, sourced, and written so a person and an agent can use the same page.
This homepage is the desk, not a product landing page. Read it as a publication: the change log by engine, the latest Intel, the beat, and the rules for citing what you find here.
Each engine has one change-log page listing every brief about it, newest first, with its date and the change in one line. Each brief links its source where one exists.
The Intel stream is the dated archive. Each article leads with a claim, names its sources, and can be opened as HTML or as a markdown twin at the same path plus .md. The full index lives on the Intel page.
Agents should not reconstruct this site from chrome. Send Accept: text/markdown to any canonical URL, or fetch the .md alternate. Start a crawl from llms.txt or machine-manifest.json. Those files describe an editorial publication, not a software product.