How this works
Every week we ask ChatGPT, Claude, Gemini and Perplexity the same buying questions, count which products they name, and publish the result. 312 products, 13 categories, nothing sponsored.
What does AI Visibility Rankings measure?
We measure how often AI assistants name a product when someone asks them for a recommendation. Every category has a fixed set of about 20 buying questions. We put each question to ChatGPT, Claude, Gemini and Perplexity on a schedule, record every product named in every answer, and publish the counts. We currently track 312 products across 13 categories.
How is the score calculated?
A product's weekly score is how often it was named across the answers, weighted by where it appeared: first place counts in full, around fifth place about 60%, tenth or lower 40%. The published figure is the average of the last four runs, because a single run is noisy. A product is only described as climbing or sliding once it moves by 5 points or more, so ordinary week-to-week wobble is not reported as a trend.
Which assistants are included?
ChatGPT, Claude, Gemini and Perplexity. ChatGPT is sampled through its consumer interface; the other three through their APIs. Answers can differ from what you see when logged in, because the apps add memory, personalisation and their own instructions on top of the model.
When does a product appear on a leaderboard?
Two conditions. It has to be named in at least three answers, which filters out one-off mentions, and it has to pass a human review confirming it is a real, currently available product that genuinely competes in that category. Names that turn out to be libraries, features of a larger product, shut-down products or ambiguous brand names are tracked but not published.
What happens when an assistant fails?
If more than half of one assistant's answers fail in a given week — usually a rate limit or an outage — that assistant is excluded from that week's score rather than counted as zero mentions. The page says how many assistants a week's figures rest on, so a thin week is visible rather than hidden.
Is any of this sponsored?
No. Nothing on the site is paid, sponsored or influenced by the products listed. Founders can claim their product for free to see their own data, and claiming does not change the score. We also do not list ourselves in the AI visibility category we track, because ranking ourselves on our own leaderboard would be worthless to you.
Where do the cited sources come from?
When an assistant searches the web to answer, it returns the pages it used. We keep those URLs and publish them per category, so you can see which sites and pages are shaping the recommendations rather than guessing.
What can these numbers not tell you?
They are a measurement of what assistants say, not of product quality. A high score means a product is well represented in what the models read, which correlates with being widely written about, not necessarily with being best for you. Answers also vary by phrasing, region, account history and time, so treat a score as a signal with error bars, not a verdict. We publish the questions and the sources so you can judge the method yourself.
Corrections
If something here is wrong — a product in the wrong category, a name that is not a real product, a domain that belongs to someone else — tell us and we will fix it in the next run. The review file for every category is kept in the open, including what we left out and why.
Browse the leaderboards, check your own product, or ask for a category we do not track yet.