I Read 25 AI-Visibility Vendors' Public Pages. Five Say Whether They Query ChatGPT or an API.

TL;DR: On 15 August 2026 I fetched the public pages of 25 AI-visibility vendors and scored each on six methodology facts a buyer needs: which assistants are queried, how many prompts per topic, refresh cadence, whether the tool queries the consumer interface or a model API, geography, and whether a price is published without a sales call. Only 5 of 25 disclose the interface-versus-API choice, which is the one that decides whether the number describes ChatGPT or something adjacent to it. Four vendors disclose none of the six. Two disclose all six. Everyone sells a share-of-voice number regardless.

Two years ago this category did not exist. Now dozens of vendors sell the same dashboard tile: your brand's share of voice in AI answers, as a percentage, over time. That percentage is not an observation. It is the output of six choices somebody made and did not write down.

What did I actually check?

The entire evidence base is public pages. I have no accounts, no trials and no demos with any of these vendors, and nothing below describes using a product. For each vendor I fetched the homepage and pricing page, then /methodology, /how-it-works, /docs, /faq, /help and /about, plus any docs subdomain I could find: over 200 URL fetches, most of them 404s. All 25 return 404 at /methodology, and exactly one publishes a methodology page anywhere. Scoring is binary and generous, and no vendor is marked ✗ until every one of those paths came back empty.

Two vendors defeated the method rather than failing it. Bluefish AI served zero bytes of readable text on all eight paths, so it is scored as disclosing nothing while actually being unmeasurable. Peec AI and Similarweb serve pricing pages that list plan tiers but return no price figure to a plain HTTP fetch, so their price column reads "n/r" (not readable without a browser), not "no".

Which vendors say whether they query ChatGPT or an API?

Five. Peec AI's docs are the clearest: "Instead of APIs, Peec AI uses advanced UI scraping technology to interact with AI models exactly as real users do," and they name the persona, "the reality for a logged-out user." ZipTie makes the same claim as a competitive jab: "Unlike many tools that rely on APIs, which can produce artificial results, ZipTie mimics real user behavior." SE Ranking uses "direct UI-based monitoring." Superlines collects "data from the real AI Search interfaces."

Evertune goes the other way and has the most interesting page in the set, the only real methodology page among the 25. It says it uses "direct API access" deliberately, to separate what a model knows from training and what it looks up live. It also publishes its sampling design: 100 prompts asked once returns a 29% visibility score at ±9 points, while the same 100 prompts asked 100 times returns 29% at ±1. That is a vendor telling you its own number is noise at low sample sizes.

The other 20 do not say. Their share-of-voice figures could come from either system, and the two are not the same product: a model API call has no user account, no memory, no ads, no shopping module and usually no web search unless it is explicitly switched on.

What did the six-column count show?

Vendor	Engines	Prompts	Cadence	UI/API	Geo	Price
Profound	✓	✓	✓	✗	✓	✓
Peec AI	✓	✓	✓	✓	✓	n/r
Otterly.AI	✓	✓	✓	✗	✓	✓
Scrunch	✓	✓	✗	✗	✓	✓
AthenaHQ	✓	✗	✗	✗	✓	✓
Evertune	✓	✓	✓	✓	✓	✓
Brandlight	✗	✗	✗	✗	✗	✗
Goodie	✓	✓	✓	✗	✓	✓
Rankscale	✓	✓	✓	✗	✓	✓
Hall	✗	✗	✗	✗	✗	✗
Trakkr	✓	✓	✓	✗	✓	✓
Knowatoa	✓	✗	✓	✗	✓	✓
LLMrefs	✓	✓	✓	✗	✓	✓
Bluefish AI	✗	✗	✗	✗	✗	✗
xFunnel	✓	✗	✓	✗	✓	✗
Superlines	✓	✓	✓	✓	✓	✓
ZipTie	✓	✗	✓	✓	✓	✓
Semrush	✓	✓	✓	✗	✗	✓
Ahrefs	✓	✓	✓	✗	✓	✓
Similarweb	✓	✗	✓	✗	✓	n/r
SE Ranking	✓	✗	✗	✓	✓	✓
Nightwatch	✓	✓	✓	✗	✓	✓
Writesonic	✓	✓	✓	✗	✓	✓
Conductor	✓	✗	✗	✗	✗	✗
BrightEdge	✗	✗	✗	✗	✗	✗

Column totals: 21 of 25 name the assistants, 19 disclose geography, 17 state a refresh cadence, 17 publish a price, 14 state prompt volume, 5 disclose interface versus API. Evertune and Superlines hit all six.

The pattern maps onto what fits in a pricing table. Prompt counts, model counts, country counts and cadence are plan limits, so sales needs them published. Query method is not a plan limit. Nobody is hiding anything cleverly; the methodology is missing because there is no commercial reason to type it.

Who discloses nothing at all?

Four: Brandlight, Hall, Bluefish AI and BrightEdge. Brandlight sells to "Fortune500 leaders" and has no pricing page (brandlight.ai/pricing returns 404); everything ends at "Get a demo." BrightEdge, selling enterprise SEO since 2007, 404s on /pricing, /docs, /faq and /methodology, and names Google AI Overviews as a target without ever saying which assistants its product queries. Bluefish AI serves an empty document to anything without a JavaScript engine.

Hall counts for a different reason: usehall.com is now one page saying the company is "now part of Tracksuit." A category eighteen months old is already consolidating, and a vendor shortlisted in spring is a farewell note by August.

What does this not prove?

It does not prove any of these tools is inaccurate. A vendor can measure carefully and market badly, and several of the silent ones almost certainly do good work. Non-disclosure is not evidence of a bad number, it is the absence of evidence of a good one.

It also does not cover what a buyer sees after signing an NDA. Several vendors will hand over a methodology deck on a sales call, and the honest read of a blank row is "you will have to ask," not "they cannot answer." My argument is that the ask should not be necessary, because two of the 25 have shown it is not. This is also a snapshot: pricing pages here change monthly, so the ✓ marks hold for 15 August 2026 and nothing else.

The pattern is familiar. It is the same one behind eight AI note-taker vendors selling action items with not one published accuracy number: a category prices and demos itself into existence before agreeing what its central number means. If one of two shortlisted tools tells you which interface it queries, that beats the feature gap, because its number is the only one you can interpret. The same disease runs through the field. It is why the crawl-to-refer ratio gets quoted at four different values for the same month, and why so few of the 90 GEO tactics I classified across 26 checklists survived having their citations followed.