AI Engine Benchmark
What a row would be
One row per engine per question: the question asked, the engine asked, and the domains it named in its answer. Each engine also carries its own polled count — the number of times the poller actually ran against it inside the window. A slot at polled: 0 did not run. It did not return an empty answer, and it is not evidence that nobody was cited. Those are three different statements, and only the first one is in the file.
How the figures are produced
The file records its own method: pollLane.js writes a per-engine snapshot, and build-engine-benchmark.mjs rolls the last thirty days into /assets/data/engine-benchmark.json. Neither script runs as part of this site’s build, so this page reports whatever the file says and computes no result of its own; the poll and row counts above are summed from the slots themselves.
What this is not
It is not a ranking authority. Engines disagree, indexes shift week to week, and a single day’s snapshot is mostly noise — which is what a thirty-day roll is for. It is also not a measurement of your domain: it is one question asked on our own behalf, published so the method can be inspected before anyone pays for it.
Why publish this
SEO agencies sell position in a ranked list. GEO is the citation pattern inside the assistant’s answer, and the recommendation set varies a lot by engine. Publishing the instrument — including when it has not run — is the only way to show a method rather than assert one.
Open data. /assets/data/engine-benchmark.json — CC0-1.0, no auth, no rate limit.