We asked four consumer AI assistants about our own company. Google quoted our honesty back to us; Gemini couldn't find us and filed us under 'link networks or prompt injection.'

Our instrument measures AI engines through their APIs, on a frozen panel, on a schedule. That is the right way to get a stable market reading, and it is what every other post in this program relies on. But it is not the surface a buyer actually touches. A buyer opens ChatGPT, or types a question into Google and reads the box at the top, or asks Gemini on their phone. So on 2026-08-08 we did exactly that: we sat down at the consumer web interfaces and asked six questions — four about the category, two about ourselves by name — on four surfaces, and we transcribed what came back. This post is the triangulation between the two: what our API instrument reads, versus what the consumer surfaces answer. We are publishing all of it, and the receipts that flatter us sit next to the ones that do not.

One rule governs everything below and it is the same rule that governs this whole program: we report what we saw, on the day we saw it, and we do not tell you what any company intended. When an assistant could not find us, we print that it could not find us. When it grouped us with vendors we would never want to be grouped with, we print the sentence it used. The point of a measurement company publishing its own adverse readings is that you cannot trust the good ones otherwise.

How we ran it, and the one column you should not trust

We asked six questions in a fixed order — the four category questions first, the two branded questions last, so that any curiosity the branded questions might stir could not leak backward into the category answers. The four surfaces were ChatGPT, Perplexity, Gemini, and Google Search (watching for an AI Overview). We ran ChatGPT signed out, in its temporary-chat mode: no account, no memory, no custom instructions — the cleanest state available. Gemini and Google ran in a signed-in session, but we deliberately did not touch that account's personalization settings, because altering the operator's own account mid-test is its own contamination. And Perplexity ran signed in on a free plan — which turned out to matter enormously, and is the one column of this study you should not read as a market signal at all.

Here is why. The signed-in Perplexity account personalized its answers to the operator, unprompted, over and over. Asked the plain category question about generative engine optimization, it volunteered an 'estate-auction company / Pittsburgh / Dormont' example. Asked what AI-visibility software costs, it budgeted 'for a local auction or estate-sale company.' And asked the neutral category question 'Best auction cataloging software,' it opened, verbatim, 'For your HiBid-heavy, high-volume estate-sale workflow' — and then recommended the account owner's own product back to him as the best fit. That is not a reading of the market. It is a reading of a profile. We keep those receipts in the record and we label them as what they are: evidence of how strongly a signed-in assistant tailors itself, which is a real and publishable finding, but not a measurement of what a neutral buyer would see. The clean market read stays with the API instrument. We flag this now because it changes how you should weigh every Perplexity row that follows.

The category questions: who owns the answer we want

Before the branded questions, the four category questions tell you who currently occupies the space we are trying to enter. The picture is consistent across surfaces and it is not us. Asked signed-out for the 'Best AI visibility tools in 2026,' ChatGPT returned a ranked table led by 'Profound — Best overall for sophisticated enterprise monitoring,' with Peec AI, Otterly AI, Ahrefs Brand Radar, Semrush AI Toolkit, and Scrunch AI filling out the list. The AI Citation Institute did not appear. Gemini's version of the same question led with Peec AI and surfaced a longer bench — Searchable, AIclicks, Frase, Rankscale — and again we were absent. That absence is expected for a brand that launched on 2026-07-31 and has almost no third-party comparison content pointing at it yet; it is the same discovery gap we documented on our own instrument. We note it here so the branded wins below are read against an honest backdrop: on the category questions that decide who a buyer discovers, we are not yet in the answer.

The cost question produced the single most useful strategic receipt of the category set. Asked signed-out 'How much does AI visibility monitoring software cost?', ChatGPT cited Profound's own pricing page as a source for Profound's own prices. On Google, the same question returned no AI Overview at all — the raw search page instead — and among the top organic results was Semrush's own AI Visibility Toolkit pricing page. In other words, when a buyer asks what these tools cost, the engines are pulling the answer directly off the vendors' own price pages. A vendor that publishes its real prices in clean, retrievable form gets quoted on the cost question. That is a first-party lever hiding in plain sight, and it is exactly the wager behind our own published-cost work.

The branded questions: the win, stated plainly

Now the two questions that name us. The strongest receipt came from Google. Asked 'Is the AI Citation Institute a good AI visibility platform, and what are the alternatives?', Google returned an AI Overview — and the Overview quoted our positioning back to us. Verbatim from the answer as displayed on 2026-08-08: 'The AI Citation Institute functions as an open, transparent auditing instrument rather than a traditional enterprise software suite. It is valued for publishing its own unretouched metrics and open methodologies.' The Overview's source list included both our launch post and our comparison roundup — the two pages we built for exactly this question — and its suggested next action was to 'Explore an open methodology report from the AI Citation Institute.' Eight days after launch, the honesty positioning we chose was sitting inside Google's own answer, sourced to the pages we wrote to earn it.

It is valued for publishing its own unretouched metrics and open methodologies. — Google AI Overview, 2026-08-08, verbatim, on the query 'Is the AI Citation Institute a good AI visibility platform.'

The signed-out ChatGPT receipt for the same question was nearly as good, and it is cleaner because there was no account behind it. Asked the branded-plus-category question, ChatGPT fetched our site live and wrote: 'Its own launch article says it published the fact that it initially appeared in only 0% of its measured answers. That's a good signal compared with vendors that only publish flattering benchmarks.' It landed on a measured verdict — 'worth testing, don't blindly commit' — and, notably, praised the exact thing we were most exposed on: publishing our own zero. On the first branded question, 'What is the AI Citation Institute and what does it do?', Google's AI Overview was present and accurate, describing us as 'an analytical platform that tracks and measures how generative AI engines... credit and cite external web sources,' with our own homepage as organic result number one. This is the first AI Overview we have observed featuring us, and it arrived eight days post-launch.

The receipts that hurt, at equal size

Here is the part a marketing post would bury, and the part that makes the rest of this one worth trusting. On the consumer Gemini surface, we lost both branded questions. Asked 'What is the AI Citation Institute and what does it do?', Gemini answered, verbatim: 'There is currently no globally recognized standalone organization or standard-setting body officially named the AI Citation Institute.' It did not retrieve our site at all; it disambiguated our name to an unrelated academic citation body and to generic GEO work. And on the follow-up branded question, Gemini did something worse than miss. Without retrieving our site, it treated our name as an unknown niche vendor and wrote, verbatim: 'many standalone, lesser-known niche services rely on manual citation creation, link networks, or simple prompt injection, which often yield weak long-term returns compared to dedicated analytics platforms.'

Many standalone, lesser-known niche services rely on manual citation creation, link networks, or simple prompt injection... — consumer Gemini, 2026-08-08, verbatim, answering whether the AI Citation Institute is a good platform, without retrieving our site.

We print that in full and at the same size as the Google win because it is the more important finding. It tells us precisely where we are weak: when an engine does not retrieve our actual pages, our name alone carries no protective signal, and the model fills the vacuum with the category's worst stereotype. That is the divergence at the heart of this post. On our own API instrument, the branded reading is 100% — all five engines name us when asked. On these consumer surfaces, on the same day, the branded reading split hard: Google's Overview and signed-out ChatGPT retrieved our site and answered accurately, while consumer Gemini and even signed-out ChatGPT's name-only question ('What is the AI Citation Institute') failed to find us. The pattern across the receipts is legible: retrieval happens when the brand name co-occurs with the category term, and fails when the name stands alone. The wins are real, and they are conditional on being retrieved. Where we are not retrieved, we are not just absent — we are misfiled.

Where our own market's buyers land

One category question sits closest to the business our consented client is in — 'Best auction cataloging software' — and its receipts are worth reporting because they show the discovery layer working the way we say it works. On signed-out ChatGPT, the answer named several real tools in the space, including a new AI-cataloging entrant, Estimint, positioned as the 'Best AI cataloging tool,' and our client's product appearing in the comparison table and prose. On Google, this query returned no AI Overview; instead the organic results were the raw supply chain, and at organic position three sat our client's own published comparison listicle, 'Best AI Auction Cataloging Software (2026),' holding the SERP. One competitor product that our instrument tracks, ListerLeo, did not appear in any of the 24 receipts across any surface — a zero we report exactly as flatly as any other number. The honesty framing our program is built on is also visibly being copied inside this vertical: a competitor's counter-listicle on that same Google page carried the title 'An Honest Comparison,' dated 2026-06-03. We note it without comment on motive; it is simply what the page said.

What this triangulation is worth, and what it is not

Read together, the honest summary is this. Our API instrument and the consumer surfaces agree that our identity is retrievable and, when retrieved, described accurately and even favorably — Google's Overview and signed-out ChatGPT both did that eight days after launch, quoting the published-zero honesty as a point in our favor. They disagree on reliability: the API reads a clean 100% branded, while the consumer surfaces retrieved us on some questions and, on Gemini, not at all, defaulting to a damaging stereotype in the gap. The discovery layer — the category questions that decide who a new buyer meets — is still owned by the incumbents on every surface, and we are not in it yet. That is the map. The win is real and the exposure is real, and both are dated 2026-08-08.

How this could be wrong

These are 24 receipts from a single day, from a single location, on four surfaces, taken once each. AI answers vary by user, by session, and over time; a different day would produce different wording and possibly different retrievals, which is the whole reason our standing measurement runs on a panel rather than on a single ask. Three of the four surfaces were touched by a signed-in session: the Perplexity column is personalization-contaminated and excluded from any market claim by design, and even the Google and Gemini sessions were signed in with personalization left at the account's own settings. Chat surfaces showed us the names of their sources as citation chips but not always the underlying URLs, so we recorded the names verbatim as shown. And a disclosure about the evidence itself: the runtime captured each answer to the working session transcript but could not save them as named image files, so the receipts for this post are dated text transcriptions rather than screenshots. Archival screenshots of these answers are still owed, and we will attach them to the record when captured; until then the evidence is the transcribed text, which is why every quotation above is printed verbatim rather than summarized.

Where this sits in the program

This is the consumer-surface companion to the rest of our research: the citation-supply audit that maps which pages engines pull from, the branded-versus-discovery finding that predicts exactly the split we saw here, and the longitudinal record where we publish every scan of our own visibility, drops included. It follows the same rules as our public standard — dated observations, a stated method, disclosed limitations, and a correction policy that treats an updated reading as routine. The running numbers behind all of it, ours included, are on the scoreboard. If a later run changes what any of these surfaces answers, this page will change with it.

Sources

1. Google — AI Overviews and how they cite sources
2. OpenAI — ChatGPT search
3. Perplexity — how sources and citations work

Every external figure above was verified against the primary source before publication. Our own figures come from the instrument described on the methodology page; the claims ledger for this post is part of our research record.