What a mention rate is, and why it needs a range

A mention rate is how many of your tracked questions named your company, out of how many you asked, and there is no universal good number. A rate of 2 of 5 on easy questions is worth less than 2 of 5 on the questions your buyers really ask. The rate is only readable next to its question set and next to your competitors on that same set.

What is actually happening underneath

Four decisions sit inside any rate, and different tools make them differently. Two tools can watch the same company and publish very different numbers without either one lying.

What counts as a mention. Named anywhere in the answer is one thing. Named in the first few products the answer lists is another, and it is the one that matters to a buyer, because people act on the first names they read. Report both, separately.

What the denominator is. Questions asked? Questions times assistants? Questions times assistants times runs? All three are defensible. Only one of them should be used, and it should be printed next to the number.

Whether brand questions are in it. If questions that name your company are counted, the rate can only go up, which makes it meaningless. Keep them out.

Whether searched and unsearched answers are mixed. An answer built from fetched pages and one recalled from training are different measurements. Blended, they produce a figure that moves when the app changes its behaviour rather than when your visibility changes.

Why a rate needs a range

A rate comes from a sample, so it carries uncertainty, and with a small set that uncertainty is larger than the movements people report as progress.

The practical version, without the arithmetic. If you asked five questions and were named in two, the honest reading is that your true rate is somewhere quite wide, and the next reading could show one or three with nothing having changed. If you asked two hundred questions and were named in eighty, the range is much narrower and a movement of a few points starts to mean something.

This produces one rule that saves a lot of embarrassment. Two rates whose ranges overlap have not been shown to be different, even when one headline number looks much bigger. Batch 01 has the arithmetic of how the range is calculated. Read it before you report a movement to anybody, because the most common misleading report in this field is a small sample run twice.

A worked example

Here is one of our own readings written as a rate. On 18 August 2026 we asked ChatGPT eighteen blind commercial questions in the category of clawlaw.in, our sister company, with no brand named in the set.

It was the top source on exactly one of the eighteen. In one answer what got named was the company's own comparison page, and the answer described it as what it is, the vendor's own editorial page.

So the honest line is a count with its denominator: top source on 1 of 18 blind commercial questions, ChatGPT, 18 August 2026. And a second line for the other definition, because being the source an answer rests on and being a name mentioned inside it are two different events. A single rate hides which one you had.

Note what we do not write. Not a percentage, because one of eighteen implies a precision eighteen questions cannot carry. Not "AI visibility", because the number describes one assistant, one day and eighteen questions we chose ourselves. And no claim of movement against an earlier reading, because on a set that size the two ranges would overlap and the comparison would prove nothing.

What it means for a business

Write the rate as two numbers and a set description: "named in 2 of 5 answers, across 5 buying questions on 2 assistants, no brand questions included". That sentence can be audited. A bare percentage cannot.

What should you compare against, then? Four things, in this order. Your own previous reading on the same set. Your named competitors on the same set, which is the only fair comparison available. The gap between named anywhere and named in the first few. And the per assistant split, because a strong total can hide being absent from one app entirely.

Report per zone as well as in total. If you serve three cities and two customer sizes, one blended rate averages away the fact that you are strong in one zone and invisible in another, which is the finding you needed.

What a mention rate does not measure

It is a reading of your question set on your chosen apps on the day you asked. Five things it is not.

It is not a share of the market or of the conversation. Nobody knows how often real buyers ask these questions or in what words, so no honest denominator exists for that claim.

It is not revenue. Being named is not being chosen. A rise in mentions with no change in enquiries is a normal and disappointing result, and it usually means you are named late in the list or described in a way that does not fit the buyer.

It is not comparable between tools. Different definitions of a mention and different denominators. Do not put two vendors' numbers in the same chart.

It is not comparable across sets. Add five easy questions and the rate rises with nothing improved. This is the easiest way to mislead yourself and the most common.

It is not stable at small sizes. Below a couple of dozen questions, report counts and say plainly that the set is too small for a percentage.

Common questions

What is a good mention rate?

There is no published benchmark worth quoting, and any single number offered as an industry standard is invented. The useful comparisons are your own previous reading on the same set, and your competitors on that same set. If you must have a target, make it a gap: be named more often than the competitor you actually lose deals to.

Should I report a percentage to my board?

Report counts with the denominator, plus the direction over three or more readings. If the board insists on a percentage, print the two numbers next to it and one line saying the set is a sample. That line is what stops a normal wobble being read as a collapse.

Does being named fifth count?

Count it, and count it separately. Named anywhere and named in the first three are both real and they behave differently. People act on the first names they read, so the second number is the one that tracks commercial value more closely.

Why did my rate fall when I did nothing wrong?

Three ordinary reasons before you assume damage: normal variation between runs, the app choosing not to search for some questions this time, and a third party page that changed or dropped out of the results. Check whether the citation lists changed before you change anything on your site.

Can I compare my rate with a competitor's published figure?

No. Unless they used your set, your assistants and your definition of a mention, the two numbers are not the same measurement. Measure them yourself on your own set, which is the comparison that is actually fair.

How do I count an answer that describes us without naming us?

Count it as not named, and note it in a comment column. A description with no name is not a mention a buyer can act on, and treating it as one inflates the number in a way you cannot defend.

What to do first

• Fix your question set and write it down. It is the denominator, so it is part of the number.
• Split your log into named anywhere and named in the first three.
• Move brand questions to a separate sheet with a separate total.
• Add a citations column so searched and unsearched answers never get blended.
• Stop reporting a percentage until the set is large. Report counts instead: 2 of 5 reads honestly, 40% does not.

A last note, since we sell a tool in this space. A low rate on a hard question set is more useful to you than a high rate on a set you designed to be flattering. The second number feels better for one meeting and then stops helping.