The reference behind every number we report.

Three ways in

Most of what is written about AI search is either a vendor selling something or a guess with a confident tone. This is the part of Baseline Labs with nothing to sell: a reference you can look things up in, a blog that shows its measurements, and the answers to the questions people actually write in and ask.

Which one do you need?

I keep seeing GEO, AEO and LLMO and I cannot tell them apart

Start on the wiki. It is ordered as a map rather than A to Z: the engines first, then how they fetch, what they cite, what content wins, what that does to your traffic, and how any of it is measured. There is a numbered reading path on the hub if you would rather be told where to begin.

Something is crawling my site and I do not know what it is

Look the user agent up in the AI bot encyclopedia. Each entry names the operator, says what the bot feeds - model training, a live answer, an assistant fetching one page for one person - whether it honours robots.txt, and how to verify a request really came from that operator instead of something wearing its name.

I am comparing tools and I do not trust vendor blogs

Fair. The resources list includes our competitors with their real prices and what each one actually does, and the blog posts show the queries, engines and dates behind their numbers so you can rerun them against us. Where a free tool does the job better than a paid one, it says so.

How a fact gets onto the wiki, and stays there

A reference is only worth reading if somebody is checking it after it is published. This is the machinery that does that, and it runs every night whether or not anyone is looking.

1. The claim is recorded with its evidence. A source URL and a verbatim quote from that source, not a paraphrase. A claim nobody can quote cannot be checked by machine, and the report says so rather than quietly letting it pass.
2. The article has to cite it. Claims that nothing in the prose points at are dropped before the page is built, so the reference list at the foot is what the article actually leans on rather than a pile of further reading.
3. Each claim gets a recheck cadence from how fast it moves. Ninety days for a definition, thirty for something drifting, five days for the figures that change weekly, counted from the date that claim was last read.
4. What falls due is refetched and diffed. If the quote has gone from the live page, the archived copy is tried; when only the archive still carries it, the article links the archive rather than a source that no longer says what we said it said.

Why every figure carries a date

Screenshots get the same treatment as quotes. A picture of an AI Overview, or of a robots.txt blocking a named crawler, is evidence rather than decoration, so each one records when it was captured and is recaptured on the same cadence its subject moves at. A shot that is past due says so under the image instead of passing itself off as current. The counts on the wiki hub work the same way: how many articles there are, how many sourced claims they hold, how recent the newest check is, are all computed when the page is built and never typed in by hand. A typed number is wrong the week after somebody types it, which is an odd way to open a page whose whole claim is that its numbers are checked.

Common questions