Since 24 September 2026, SQSEO publishes a page nobody else in the category publishes: a live, daily measurement of which AI visibility and SEO tools the AI assistants name when a buyer asks. It lives at /ai-visibility/ai-search-tools. This post explains what is on it, how the numbers are made, how to read a share with its interval, and what the first day of the ledger already says about the category.
Why a public leaderboard, and why now
Every vendor in this category sells the same promise: we tell you how often the AI assistants name your brand. Almost none of them show their own number. That is odd for a measurement product. If the method is sound, it should survive being pointed at its maker.
The second reason is practical. Buyers compare tools by reading listicles written by the vendors themselves. A daily ledger with sample sizes and intervals is a better basis for that comparison than any listicle, including ours. So the page is public, the JSON behind it is public, and the method is written down in full.
The third reason is that the category is moving fast. peec.ai’s pricing page lists plans of 50, 150 and 350 prompts tracked across three models daily, which is the volume a serious buyer expects (peec.ai pricing). A public page at fifty prompts across eight assistants is the same order of magnitude, measured in the open.
What the page shows
Every day SQSEO asks a fixed set of fifty prompts. They are the questions a buyer of an AI visibility or SEO tool types, across five topics: AI visibility tools, generative engine optimization, AI search analytics, competitor tracking in AI answers, and SEO tools for small businesses. Each prompt is asked of every assistant SQSEO can observe through an official API, and every answer is stored as evidence before anything is counted.
From those answers the page derives four views.
- A visibility leaderboard. For each brand, the share of answers that name it, with its 95 percent interval, its share of voice among all brand mentions, how often the answers cite a page on its own domain, and its average position when it is named.
- The same measure per assistant. Claude, Gemini and the others do not agree with each other. The matrix shows where each brand is strong, with the sample size in every column.
- The pages the answers cite. The domains the assistants pulled their answers from in the window, with the count of answers citing each. This is where the citations come from, and it is the list any GEO plan should start with. It is the same list you would build by hand when you compare brand share in Perplexity.
- A seven day trend. Daily visibility for the leading brands. A gap is a day without measurements, never a zero.
How a prompt becomes a number
The chain from a prompt to a number on the page has five links, and each one is checked by a test in the codebase.
First, the prompt is scheduled. Fifty prompts, each with a target per assistant for the United States daily, and weekly targets for five more countries on the assistants whose search follows the country. The scheduler writes one job per prompt, assistant, country and day, keyed so that a retry can never create a second job for the same day.
Second, the job is captured. A worker calls the assistant’s official API with the prompt and nothing else: no system prompt that nudges the answer, no examples, no brand names in the request. The full response, including every page the assistant retrieved and every search it ran, is stored as an artifact. The cost of the call, at the provider’s list price for the tokens and searches the response itself reports, is written on the job whatever the outcome.
Third, the answer is parsed. A parser per assistant reads the response into one shape: the answer text, the sources retrieved, the sources cited, and the search queries the assistant issued, which SQSEO calls fan-outs. Every parser is pinned to real recorded responses, so a change in a provider’s format shows up as a failing test, not as a silent zero.
Fourth, the brands are matched. A brand counts as mentioned when the answer names it or its domain, using the aliases the project keeps for each brand. Peec AI is also “Peec” and “peec.ai”; Semrush is also “Semrush AI Toolkit”. The matcher has its own evaluation corpus and a precision target it must hold on every deploy.
Fifth, the numbers are rolled up. Visibility is mentions divided by answers, share of voice is a brand’s mentions divided by all brand mentions, and each carries a Wilson score interval at 95 percent. The Wilson interval is the standard choice for a proportion at small samples, because it stays inside zero and one and does not collapse to nothing when the count is zero (Wilson score interval, Wikipedia).
The assistants, and what each one costs to ask
The assistants on the page differ in how they search, whether the search follows the country of the prompt, and what one answer costs. The table below is measured, not quoted: the searches per answer and the list price per answer come from the first real captures in SQSEO’s ledger on 24 September 2026, and the pricing rules come from each provider’s documentation.
| Assistant | How it searches | Search follows the country | Searches per answer (measured 24 Sep 2026) | List price per answer (measured) |
|---|---|---|---|---|
| Claude Sonnet 5 | Anthropic’s web search tool | Yes, by user location | 2 | 0.08 USD |
| Gemini 2.5 Flash | Google Search grounding | No | 4 fan-out queries | 0.04 USD |
| GPT-5 (through OpenRouter) | OpenAI’s native search | No | 6 | 0.11 USD |
| Grok, DeepSeek, Qwen, Mistral, Llama (through OpenRouter) | The Exa web plugin, 5 results | No | 1 | 0.01 to 0.02 USD |
Two things in that table matter for anyone building or buying a tracker. The first is that the search results dominate the cost, not the answer. Anthropic’s documentation lists web search at 10 USD per thousand searches on top of token costs, and a single answer pulled roughly 21,000 input tokens of search results (Anthropic web search tool). GPT-5 at its default reasoning effort ran six searches and used its whole output budget before writing a word, which is why the page’s OpenAI column runs the smaller model at low effort.
The second is that only Claude’s search follows the country of the prompt. Google’s grounding documentation describes grounding as a search over the web at answer time with no location parameter in the request (Grounding with Google Search). So the weekly targets for Britain, the Netherlands, Germany, Poland and Denmark run on the assistants that can honour a country, and the page says which columns those are.
How to read a number with its interval
A visibility of 62 percent with an interval of 50 to 72 means: of the answers in the window, 62 percent named the brand, and at this sample size the true share is very likely somewhere between 50 and 72. Two brands whose intervals overlap are not reliably in a different place. A brand whose interval sits entirely above another’s is.
The interval narrows as the sample grows, which is why the page shows the number of answers next to every figure. The table below shows the same 62 percent at four sample sizes, computed with the Wilson formula the page uses.
| Answers in the window | Brand named in | Visibility | 95 percent interval |
|---|---|---|---|
| 10 | 6 | 60 percent | 31 to 83 percent |
| 70 | 43 | 61 percent | 50 to 72 percent |
| 300 | 186 | 62 percent | 56 to 67 percent |
| 1,000 | 620 | 62 percent | 59 to 65 percent |
At ten answers the interval is so wide that the number says almost nothing. At seventy it separates the leaders from the tail. At a thousand, which is about a week of the page’s daily volume on one assistant, it separates brands a few points apart. This is the same reading discipline a buyer should apply to any tracker, free or paid. The spot-check method for AI trackers is the afternoon version of it.
A brand that no answer named yet reads “not named yet” rather than a zero. At small samples a zero is not a measurement of absence. The interval’s ceiling says how high the true share could still be, and the page says so instead of drawing an empty bar.
What the first day already says
The first seventy answers of the ledger, measured on 24 September 2026, put Otterly, Profound, Ahrefs, Semrush and Peec AI in the top five, each named in roughly half to nine tenths of the answers, with Scrunch and Rankscale named in about a third. SQSEO was named in none of the seventy. That is the honest starting line for a product launched this year, and the reason it sits in the table with everyone else.
Three patterns are already visible and worth watching as the sample grows.
The generalist SEO suites are named as often as the specialist trackers. Ahrefs and Semrush appear in half of the answers to prompts about AI visibility tools, because the assistants lean on the listicles that rank for those queries, and those listicles include the suites. A specialist that wants to be named has to be present on the pages the assistants read, not only on its own site.
The assistants disagree. The same brand can be named in most of Claude’s answers and few of Gemini’s, because the two retrieve different pages. A tracker that reports one blended number hides this, and the blend moves whenever the mix of assistants moves. The page keeps the per assistant matrix next to the blended leaderboard for that reason.
The cited domains are concentrated. A handful of comparison pages supply most citations across all assistants. That list is the working brief for anyone who wants to move their number, and it is the part of the page most worth reading every week.
Why SQSEO is in its own table
Because the measurement is the product. SQSEO is a new entrant in a category with established names, and the honest way to show that is to sit in the same table, measured by the same rule, from the first day. The numbers will move. The rule will not.
There is a second, quieter reason. A vendor that publishes its own zero has removed the easiest objection a buyer can raise about a tracker, which is that the vendor’s numbers are chosen to flatter someone. The page has no such freedom. It reads the ledger, the ledger reads the answers, and the answers are stored before anyone sees them.
What the page does not do
The page does not sample from anyone else’s data. It does not estimate a number for a day without answers. It does not solve a challenge page or use a proxy to reach a consumer surface that refuses automated visitors. The consumer versions of ChatGPT, Perplexity and Google’s AI Overviews are not on the page today for exactly that reason. Each one refused every hosted address SQSEO tried, and the refusals are recorded as the channel’s truth. When one of them serves a logged-out visitor from an address SQSEO is allowed to use, its column appears.
The page also does not claim precision it does not have. The window is seven days, the sample sizes are printed, and the intervals are honest about what fifty prompts can and cannot separate.
How to use it if you are a buyer
Read the per assistant matrix before the leaderboard. The assistant your customers use is the column that matters, and the blended number can hide a column where your brand is absent.
Read the cited domains as a brief. If three comparison pages supply most citations, those pages are where your brand needs to be described accurately, and where a request for correction or inclusion is worth the email.
Watch the trend for a month before drawing a conclusion. The first week of any measurement is noisy, and the interval says so. If you are wondering why a competitor is named where you are not, start with why brands disappear from AI product reviews.
Track your own brand the same way
The page is the same measurement SQSEO runs for its customers: your prompts, your competitors, your countries, the assistants that matter to you, every number with its interval, every answer kept as evidence. It starts free.
Key takeaways
The leaderboard is a daily, public measurement of which AI visibility and SEO tools the assistants name, built from fifty buyer prompts across eight assistants, with every number carrying its sample size and a Wilson 95 percent interval. Only Claude’s search follows the country of the prompt, and search results, not answers, dominate what an observation costs. SQSEO is in the table from day one, at zero of the first seventy answers, measured by the same rule as everyone else.
