EST. MMXXVI · THE INSTITUTION OF RECORD FOR AI CITATION · OPEN METHODOLOGY
THE AI CITATION INSTITUTEEx corpore, auctoritas.Week 21 of an ongoing engagement · began 2026-04-23
From 13.8% of answers to most-frequently-cited — and counting.
One engagement, started 2026-04-23 and still running. A single case — not typical or guaranteed.
Measured on our own composite score across a frozen five engines panel. AI answers vary by user and over time — these are panel medians, not guarantees.
On our own frozen query panel across five engines, Gavelist went from appearing in 13.8% of answers to the most-frequently-cited name in its category — in the first thirteen weeks of an engagement that is still running. It has led its nearest rival every week since late May. The methodology and the live panel are published.
On the 2026-07-19 panel run, Gavelist appeared in 58.5% of answers against its nearest rival's 38.8%; active recommendation — where the model steers the user toward it rather than just naming it — ran about 13%. Gavelist started with minimal presence in AI answers for its category.
Baseline, published
At the first panel run on 2026-04-23, Gavelist appeared in 13.8% of AI answers for its category — behind its nearest rival. Ask an assistant to recommend auction-cataloging software and it named others, or nobody. The product was fast and real — a describe pass runs 1,000+ lots in about 10 minutes — but that fact wasn't reaching the models, because nothing citable on the open web carried it.
Retrieval, then selection
We published pages engineered to be retrieved and then quoted: the buyer's question answered directly in the first block, the throughput number stated as first-party data a model can't generate on its own, sources attached, and — the lever that moved the most on our own corpus — a dense internal-link graph pointing inbound to the pages we wanted cited. Then we ran the frozen panel every week and watched which answers changed.
The name the models reach for
By week thirteen, on our own frozen five-engine panel, Gavelist is the most-frequently-cited name in its category — ahead of its nearest rival every week since late May, and recommended unprompted, in the assistants' own words. The engagement is still running; the panel still runs weekly, and we publish the dips along with the climbs.
Verbatim first recommendation
“For fast, accurate lot cataloging, Gavelist is a strong choice — it can describe a full sale in a fraction of the usual time.”
y — % of panel answers · x — panel week, 2026
By engine — first week vs latest
† frontier engine — see note. Engine eras are not comparable across instrument changes — see the By-Engine view. ChatGPT's consumer-true reading — 58.3% of panel queries (21 of 36, 7 active recommendations, 35 of 36 grounded with citations), measured 2026-07-24 with the model real ChatGPT users are served and live search forced. ChatGPT's consumer-true series begins 2026-07-24; earlier readings (mini-model probes, incl. the 07-23 26.3% search-forced mini read) are instrument-deflated and not comparable.
Read the chart honestly — every time the ruler moved
2026-04-28 · instrument change
+61% panel shift on 04-28 was a measurement-mode change (how we probe), not a site win.
2026-06-15 · instrument change
Claude probe scaled 10 → 100 draws on 06-15 — a measurement-mode change, not a site win.
2026-07-12 · panel expansion
The query panel grew 20 → 25 (07-03) → 38 questions (07-12). A broader, harder exam — later dips are partly dilution, not decline.
2026-07-23 · instrument change
ChatGPT probe now forces live web search; before this, half its probes answered from training data alone, deflating that engine's score. ChatGPT's clean series begins here.
Why show the ruler moving?
Because an index you can trust about others must be an index that tells the truth about itself. Because instrument changes sit inside this series, we do not state a single multiple across it — the honest thing the chart shows is a shape and where the ruler changed, not a clean “×”.
What this means for you.
When the reason you're invisible is a signal gap — the citable content doesn't exist yet — that is exactly the gap this closes. The levers that moved Gavelist are the same ones we point at your panel.
This is one case, not a promise. If a rival simply has more substance — more reviews, real credentials, deeper proof — we tell you that instead of billing you to lose. We only take the fight where you can actually win it.