Baseline Labs wiki
50 articles on how AI search works, written claim by claim: 704 numbers and statements, each with a named source, a verbatim quote and a date it was read. 512 of them (73%) also carry a Wayback snapshot, so a source that moves or disappears does not take the evidence with it. A nightly job refetches the fastest-moving figures and diffs the quoted sentence against the live page. The most recent source read was 2026-09-05.
The numbers
| Measure | Count |
|---|---|
| Articles | 50 |
| Sourced claims | 625 |
| Sources saved to a public archive | 512 |
| Links between articles | 279 |
| Articles reviewed | 49 of 50, latest 2026-09-04 |
| Bot stubs in the registry | 150 |
| Station: Engines and operators | 10 articles, 126 claims |
| Station: How they fetch | 10 articles, 129 claims |
| Station: What they cite | 6 articles, 83 claims |
| Station: What content wins | 9 articles, 113 claims |
| Station: What happens to traffic | 5 articles, 53 claims |
| Station: How you measure | 4 articles, 50 claims |
| Station: Foundations and rules | 5 articles, 71 claims |
| Source kind: Documentation | 185 |
| Source kind: Vendor documentation | 182 |
| Source kind: Journalism | 112 |
| Source kind: Academic | 102 |
| Source kind: Empirical study | 44 |
| Recheck tier: Stable | 191 |
| Recheck tier: Moving | 220 |
| Recheck tier: Volatile | 214 |
Start here
Seven pages, in order, if you are trying to get your own pages cited and do not know where to begin.
- Generative engine optimizationWhat the discipline is, and how it differs from SEO.
- Answer enginesThe products that answer instead of listing links.
- AI crawlersWho comes to your site, and what you can tell them.
- AI citation sourcesWhere the answers get their citations from today.
- Content structureWhat makes a page easy to quote once it is found.
- Zero-click searchWhat being cited does to your clicks.
- AI visibility measurementHow to tell whether any of it worked.
The map
Left to right, how an AI answer gets made: who answers, how they fetch, what they cite, and what that does to your traffic.
Engines and operators
The products people ask, who runs them, and the surfaces their answers appear on.
10 pagesHow they fetch
How those engines reach a page, and what you can tell them at the door.
11 pagesWhat they cite
The sources an answer leans on when it picks who to quote.
6 pagesWhat content wins
What makes a page worth citing once an engine has it.
9 pagesWhat happens to traffic
What all of this does to clicks, referrals and share of search.
5 pagesHow you measure
Working out whether you are cited, and what the numbers mean.
4 pagesFoundations and rules
The disciplines underneath, and the rules that bound them.
5 pagesMoving right now
Figures we refetch every five days. The date is when the source was last read.
Show the numbers (3)
| Point | % |
|---|---|
| Perplexity | 28.6 |
| Gemini | 8.6 |
| ChatGPT (in-text) | 8 |
Last checked 2026-09-04
Show the numbers (4)
| Point | % |
|---|---|
| 2020, worldwide (Similarweb) | 65 |
| 2024, US (Datos) | 58.5 |
| 2024, US (Similarweb) | 60.45 |
| 2026, US (Similarweb) | 68.01 |
Last checked 2026-09-04
Show the numbers (6)
| Point | % |
|---|---|
| 91.1 | |
| Bing | 4.5 |
| Yahoo! | 1.23 |
| Yandex | 0.99 |
| DuckDuckGo | 0.7 |
| Baidu | 0.62 |
Last checked 2026-09-04
Show the numbers (4)
| Point | citations |
|---|---|
| Google AI Mode | 340721 |
| Perplexity | 146196 |
| ChatGPT | 25681 |
| Gemini | 49 |
Last checked 2026-09-04
Last checked 2026-09-04
Last checked 2026-09-04
Show the numbers (4)
| Point | Jaccard similarity |
|---|---|
| Same prompt, rerun (low) | 0.5 |
| Same prompt, rerun (high) | 0.61 |
| Cosmetic paraphrase | 0.29 |
| Constraint-adding paraphrase | 0.14 |
Last checked 2026-09-04
Show the numbers (4)
| Point | % |
|---|---|
| 2025-01 | 6.49 |
| 2025-03 | 13.14 |
| 2025-07 | 24.5 |
| 2025-11 | 15.5 |
Last checked 2026-09-04
Most cited sources
The domains the wiki quotes most, counted across every claim. Primary sources first is the house rule, so the blog and documentation an engine publishes itself outrank commentary about it.
- arxiv.org85 claims, 26 pages
- developers.google.com73 claims, 28 pages
- ahrefs.com32 claims, 11 pages
- en.wikipedia.org32 claims, 12 pages
- developers.openai.com28 claims, 11 pages
- searchengineland.com28 claims, 15 pages
- blog.cloudflare.com27 claims, 10 pages
- blog.google22 claims, 11 pages
- justice.gov20 claims, 1 page
- cjr.org15 claims, 8 pages
How it works: every number on these pages is a claim with a named source, a retrieval date and a volatility tier, and each source is saved to a public archive the moment we cite it. Volatile figures are refetched every five days and the quoted sentence is diffed, so a page tells you when it was last checked rather than asking you to trust it. The bots each engine sends are catalogued separately in the bot registry.
Every article
50 entries, alphabetical.
- AI agents
- AI assistant fetchers
- AI citation sources
- AI crawlers
- AI Mode
- AI Overviews
- AI referral traffic
- AI visibility measurement
- Answer engines
- Anthropic
- Backlinks
- Bing and Copilot
- ChatGPT Search
- Citation accuracy
- Cloudflare
- Content freshness
- Content licensing deals
- Content structure
- Core Web Vitals
- E-E-A-T
- Featured Snippets
- Gemini
- Generative engine optimization
- Georgestub
- Google antitrust remedies
- IndexNow and sitemaps
- JavaScript rendering
- Knowledge Graph
- llms.txt
- Local search and AI
- Multilingual GEO
- OpenAI
- Perplexity
- Query fan-out
- Retrieval-augmented generation
- Robots.txt
- Scaled content abuse
- Search crawlers
- Search engine optimization
- Search market share
- SEO and GEO
- Structured data
- Training data versus retrieval
- Training scrapers
- Unlinked brand mentions
- Wikipedia
- YouTube
- Zero-click search
Volatile figures are rechecked every 5 days, moving facts every 30, stable principles every 90.
