What we've measured about getting recommended in AI answers, written to be used rather than admired. Every number here re-derives to specific questions and specific engines, and where a mechanism is unproven we say so.
Search gives you a list; an AI answer makes a choice. A four-layer model - eligibility, demand, evidence, measurement - with the sourced numbers and the test to run on any vendor.
Read it →Tools measure retrieval: what engines answer today and which pages they read. No tool can see inside a model's training. The difference decides what you can fix and how fast.
Read it →Instagram pages showed up as cited sources in 34 of 276 answers in one consumer-retail measurement, and two of four engines did all of it. What to do about it, and what not to.
Read it →Six checks, in order: was the wave comparable, did the model change, is it inside the noise floor, did the fix actually get indexed, are you pushing the un-movable layer, and has it simply not been long enough.
Read it →The checks that decide whether AI engines can read your site at all - crawler access, JS dependence, answerable structure, verifiable facts - and the fashionable checks we refuse to score.
Read it →Visibility rate, top-3 rate, #1 pick rate, citation, grounding query, AI-sourced vs AI-influenced. What each number counts, what it can't say, and why a score out of 100 isn't measurement.
Read it →Before an engine answers, it rewrites the buyer's question into several internal searches and reads the winners. How fan-out works, why it varies by engine, and what it breaks about keyword thinking.
Read it →GPTBot trains models. OAI-SearchBot builds a search index. ChatGPT-User fetches live for a person mid-answer. Three different jobs, three different robots.txt decisions - most sites treat them as one.
Read it →A proposed standard for handing AI engines a curated map of your site. No major engine has confirmed consuming it. We publish one anyway - here's the honest case for a cheap hedge.
Read it →JSON-LD makes your facts machine-legible and costs little. What's actually evidenced, what's vendor folklore, and where schema sits in the readiness stack.
Read it →Segment the referrers, catch utm_source=chatgpt.com, and split AI-sourced clicks from AI-influenced customers who arrive looking like direct. Plus the answer-without-click traffic you'll never see in GA4.
Read it →The engine read pages that answered the buyer's question, and they were your competitor's. Which sources get trusted, what training memory adds, and which part of that you can actually change.
Read it →Same question, same engine, different day - different answer. Model updates, retrieval freshness, and nondeterminism all move the needle, which is why one-off checks mislead and waves don't.
Read it →Citation share tells you how often engines read you. Share of voice tells you how often buyers hear your name. Measuring the second takes a frozen question set and a schedule.
Read it →Measuring how AI engines answer real buyer questions, and what moves the counts, as a discipline distinct from search optimisation.
Read it →