A leaderboard is only worth reading if you can check it. This article is the argument for publishing the whole scoring method, and a walkthrough of what indexwire measures, how a score is computed, and the parts it deliberately leaves out. The canonical specification, with the exact weights, lives at /methodology and is versioned; the current version is v1.
A score you can’t verify is a score you can’t trust. So the method is open, down to the arithmetic.
Why publish the whole method
Most AI visibility tools hand you a number. Fewer explain how they reached it. A description of principles and a glossary of terms is better than nothing, but it is not enough to reproduce a score.
Open methodology means publishing the actual inputs and the actual arithmetic, not just the approach. If you have the prompt set, the weights, and the rule for combining them, you can run the same measurement and get the same number. If you don’t get that number, one of us made an error, and the error is findable.
There is a second reason, and it is the harder one. An open method commits us to publishing results that undercut our own product. If a run shows a tactic we sell does not move the score, the method gives us nowhere to hide it. That constraint is the point, not a cost.
What one run measures
A run covers one category for one ISO week. It puts a fixed set of prompts to a set of AI assistants; each prompt-and-assistant pair is one probe, and each probe is recorded with the answer it produced. The prompt set and the assistant set are frozen when the run starts, so a week’s shape cannot drift while it is being measured.
In each answer, every brand named is recorded three ways: what it was named as, where in the answer it appeared, and whether the mention was backed by a citation.
What it was named as. Three kinds of mention carry weight: a primary recommendation (1.0), an alternative (0.7), and an honorable mention (0.4). Everything else, a comparison target or a passing bit of context, scores nothing. Being named unfavorably is neutral here, not a penalty.
Whether it was cited. A mention backed by a link scores 1.0, an implied-but-unresolvable source 0.9, and a recommendation with nothing to point at 0.8. The recommendation is the thing being measured, so an uncited one still counts for most of its value.
Where it appeared. Position decays gently: first place 1.0, second about 0.63, third 0.5. Order matters, but being named at all matters more than being named first.
How a score is computed
For each answer, indexwire takes a brand’s single best recommendation mention, its three weights multiplied together, and nothing else. Being named five times in one answer scores exactly the same as being named once, well. That cap is deliberate: it stops the score rewarding brands that flood an answer.
Those per-answer numbers are summed and divided by every answer the run asked for. The denominator is the part worth reading twice. An answer where a brand was not mentioned counts against it, and so does a provider call that failed. It is a share of the answers we went looking for, not a share of the answers we got back.
In v1 every prompt and every assistant counts equally. The result lands between 0 and 1. It is a share of one category’s prompt set in one week, comparable only against the other brands in that same run, never a percentage of what AI assistants say in general.
How brands are ranked
Highest score first. Exact ties break by brand identifier in ascending order: an arbitrary rule, but a fixed one, so a tie never resolves differently on a re-run.
Every brand tracked in a category is ranked, including the ones no assistant mentioned. They appear at zero. A leaderboard that quietly dropped them would flatter everyone left on it.
Alongside the score, each run publishes the count of answers a brand appeared in. That is a count of answers, not mentions: appearing three times in one answer counts once, in keeping with the cap above.
The same answers always give the same score
The scoring code reads no clock and uses no randomness, and it sorts its inputs before it sums them. The same observations scored under the same version produce a byte-identical result, every time. That is a property of the code, not an aspiration, and it is what makes a score checkable rather than merely published.
It also changes what a methodology change costs. Every answer an assistant gives is recorded once and kept, and the score is a pure function of those records. Improving the method means recomputing from the answers already on hand, not going back to ask the models again. A re-folded score carries the version it was computed under, so the number never changes silently.
What this deliberately does not measure
Naming what the method leaves out is part of publishing it.
It does not read personalized answers. Only public, non-personalized answers feed a score; an answer shaped by a location or a signed-in profile is left out rather than guessed at.
It does not weight assistants by popularity. Which assistants a run covers is configuration, not methodology: adapters exist for ChatGPT, Claude, and Perplexity, and each run records the set it asked. Adding one changes the answers, not the arithmetic.
It does not score sentiment. The method measures whether an assistant recommends a brand and how prominently, not how warmly it describes it.
It does not compare across categories or weeks. A score belongs to one category’s prompt set in one ISO week. Reading it against a different category, or a week with different prompts, is not something the number supports.
What we don’t know yet
Open methodology means being explicit about the holes.
We don’t yet have a precise measure of answer-to-answer variance. The same prompt does not always return the same answer, and how much a score moves on a re-run of the same week is something we are still measuring rather than asserting.
We don’t claim the prompt set is complete. A frozen set of prompts is reproducible, which is the point, but a category is larger than any prompt set. Publishing the prompts is what lets you judge whether they represent your category, and tell us where they don’t.
We’ll publish findings on both as the data supports them, in the same detail as the method above.
Check it yourself
The method is not a summary of something proprietary. It is the whole thing. /methodology carries the exact weights and the arithmetic; the glossary defines the terms; each category’s page on the index shows the scores that method produced, next to the count of answers each brand appeared in.
If you run the same prompts against the same assistants and reach a different number, that is a conversation worth having, and the deterministic scoring means the discrepancy has a findable cause. A visibility score is only as good as the method behind it, and a method is only as good as your ability to check it.