What an unsourced figure costs. Without a denominator a claim cannot be wrong, so the risk of repeating it lands on you, and never on the vendor who made it. Every figure in these CiteAngle reports is sealed to a registered claim with its window, which is the reason you can cite one and defend it.
The second cost is a market-wide one. Without published, checkable reports, evidence-free claims set the price for everyone, and a vendor doing careful work has no way to show the difference — so buyers pay the same for both. These documents exist so that difference is visible before the invoice.
3
series open on this page — market maps, demand and citation evidence, and the pre-purchase checks
Every issue keeps its own permanent address
4
numbers to ask any vendor for before you sign: how many questions, on which surfaces, over how many runs, and measured against what
Answer them from any report here
95
percent Wilson confidence band printed around every measured rate in these documents
Unread observations stay in the denominator and print as a count
Why this market needs published reports
Almost everything written about AI search visibility is written by somebody selling AI search visibility, and almost none of it names a denominator. A deck says mentions tripled. Across how many questions, on which surfaces, over how many runs, measured against what? Without those four numbers the claim cannot be wrong, which is another way of saying it cannot be checked. A buying team ends up choosing between vendors on confidence alone.
These documents exist so there is something in the room that can be checked. Each report says what was asked, where it ran and how many times, and reports every measured rate with the Wilson 95% interval around it. Our own measurement panel is published as arithmetic: 50 buyer questions across 15 engines, 7 runs each, landing on 17 US surfaces. Every figure quoted in these reports is registered with its source and as-of date in the claims registry. Readings the grid planned but could not take are printed as a count, because a denominator that quietly shrinks makes any rate look better than it was. Before you sign, ask a vendor for those four numbers and check the answer against any report here. The gap shows up in a minute, and that is why we publish them.
What to fix on your own pages first
Start with one figure you already publish about your own brand — on a pricing page, in a deck, anywhere a buyer can see it. Most figures published in this market carry none of the three marks that make one checkable: the denominator it was measured against, the window it was measured in, and a source someone else can open. Fix that one figure first. An unsourced number of your own is the first thing a careful buyer will test. The test takes them a minute. Then move on to the next figure, and the one after that.
A report you can check beats a deck you cannot. Every figure here comes with its window and its Wilson 95% band, and that is why one of these holds up when a buyer asks where it came from.
The library — reports by series
Everything here keeps a permanent address. A link you cite today points at the same document next year, and revisions happen at the same URL with their change dates on record.
Market maps & baselines
Where US buyers actually meet AI answers, and what the ground looked like before the conditions changed — the documents you return to when someone asks "what was it like before".
- The 2026 US AI Search Surface Map: Where Buyers Actually Meet AI Answers
- Ads Have Arrived in AI Answers — Here's What Money Can and Can't Buy There
- A Falling Ad-to-Revenue Ratio Often Means Revenue Moved Faster
- Twenty-Nine Quarters, Three Phases: What an Ad Budget Looks Like When You Rebuild It Quarter by Quarter
Demand & citation evidence
What the controlled evidence says about how AI search expands questions and selects sources — reviewed against the published studies, not vendor decks.
- One Keyword Is Now Dozens of Questions: What Query Fan-Out Means for Demand Measurement
- What Actually Gets Cited by AI Search: A Sober Review of the Controlled Evidence
- Which Domains US AI Answers Actually Cite — 15 Surfaces, 1,987 Domains, and the Fifth You Can Publish To
- A Single Run Is a Point Estimate, Not a Number: The Published Evidence
Verification before you buy
The comparison grid, the denominator questions and the official telemetry rails — the documents a buying team reads before signing a measurement contract.
Our industry visibility index series — sealed query sets re-measured on a recurring schedule, published with named brands — currently ships for the Korean market and is archived on the Korean library page.
Publication policy — fixed addresses, updates in place
One address, for the life of the report
- This hub keeps its address. Series grow and issues accumulate, but the library URL does not move.
- Every report keeps its own URL. Published documents stay up after publication — retro-reading past issues is what makes published figures checkable.
- Revisions happen in place. Annual and recurring updates replace the document at the same address, with the update date and a correction block on the page — no new URL per edition.
- Citing us. Journalists and researchers are welcome to cite these reports with a link to this hub or the report's own URL. The evidence behind published figures sits in the public claims registry and the methodology.
Every figure on this page came from a measurement we ran on ourselves. Running the same grid on your own brand starts with an audit. See the audits and what they cost →
New reports are listed here the day they publish. For analysis and commentary, see the research index.