Understand AI search:
How engines discover and select sources
AI engines turn a prompt into searches, retrieve pages, pick the passages that answer best and cite a few. Start here: see the full process, a worked example for "best whatsapp mcp", how each engine picks sources and how query fan-out works.
How AI engines pick sources:
One prompt narrows to a few citations
Why do only a few pages get cited for each prompt?
Each step narrows the field. The engine runs several searches, retrieves many pages, then keeps only the passages that answer best. Only a few sources make it into the answer.
- Read top to bottom: each band is a step an engine takes.
- The width shows how many pages are still in play.
- Orange is the outcome you want: being one of the few cited.
Good sign: your pages are retrieved for several fan-out searches, not just one.
Warning sign: you rank well on Google but never appear in the cited sources.
of searches Deep Search in AI Mode can run for one question
Google, May 20, 2025of AI Overview citations also rank in the top 10 for the same query
Ahrefs, Mar 2, 2026of Google visits with an AI summary ended on a click inside the summary
Pew Research Center, Jul 22, 2025of AI citations come from earned media, not your own pages (vendor claim)
Muck Rack, May 8, 2026Lessons in Understand.
- Research 1.1→
Compare AEO, SEO and GEO
SEO ranks pages, AEO gets answers quoted by AI and GEO is the research name for the same AI work. See how they differ and what to do first. - Lesson 1.2→
Learn the 8-step process
The AI search optimization process in 8 steps: keyword, prompts, fan-out, baseline, actions, fixes, owned, earned and paid, then attribution. - Lesson 1.3→
Walk through a worked example
We asked ChatGPT "best whatsapp mcp" on Nov 17, 2026. See the 5 tools it named, the 7 sources it cited and what each one teaches about AI search. - Lesson 1.4→
See how engines pick sources
How ChatGPT, Google AI Mode, Perplexity, Gemini and Claude find pages, pick sources and cite them, and what to publish for each engine. - Lesson 1.5→
Map the query fan-out
Query fan-out is how AI Mode and ChatGPT split one prompt into many searches. Learn to list the fan-out queries for a keyword and cover them with pages. - Research 1.6→
Get mentioned in Claude
Claude answers stable questions itself and searches only for current ones. See which prompts make Claude search, and how it picks sources to cite. - Research 1.7→
Budget for AI search
Budgets are flat, yet enterprises put 12% of digital spend into AEO and GEO. See where the money moves and how to prove it pays.
See this hub in ThriveStack citedby.
ThriveStack citedby turns this research into actions for your own brand: tracked prompts, ranked fixes and attribution to revenue.