ai mention share

Ask an AI assistant about your category. Count who shows up, in which situations, and how often.

Your brand (as written in the list below) Date you collected the answers Confidence

Brands to track, one per line. Add other spellings after an equals sign

Example line: Fizzo = fizzo cola, fizzo zero. Matching ignores case and matches whole words only. Include every brand you would count as a competitor.

1. Make the prompts

Each prompt should be a situation where someone might reach for the category, written without any brand name in it. Those situations are your category entry points.

2. Paste the answers

Start each answer with a line like === Assistant name | the prompt, then paste the answer text

The sample uses invented brands and invented answers.

verdict

Answers mentioning you

Share of mentions

Your mentions as a share of all tracked-brand mentions.

Named first

Situations you appear in

The link carries your settings and brand list. It does not carry the pasted answers.

Every brand

Brand	Answers mentioning	Range	Share of mentions	Named first	Situations

By situation

Each cell is the share of answers to that prompt that mention the brand.

By assistant

Brands you may have left off

Capitalised names that appear in at least 2 answers and are not in your list. This is a rough guess and includes non-brands.

How it works

The measures

A brand counts once per answer, however many times the answer repeats it. Answers mentioning is the share of answers that name the brand. Share of mentions is the brand's count divided by the total count across all tracked brands. Named first is how often it is the earliest tracked brand in answers that name any. Situations is the share of prompts where the brand appears in at least one answer.

These borrow from mental availability: how many of a category's buying situations bring the brand to mind. An assistant's answer is a different thing from a buyer's memory, so treat it as a proxy and a new one.

The verdict

Your brand is compared with the best-mentioned rival, answer by answer. The difference is the count of answers that name you and not them, minus the reverse, over all answers. Its range uses the paired difference method:

d = (b - c) / N SE = sqrt(b + c - (b - c)^2 / N) / N range = d +/- z x SE

ahead: the range sits above zero. behind: below zero. level: it covers zero, so the data cannot separate you. invisible: you appear in no answers. Fewer than 10 answers is flagged as a first look.

Read this before you quote a number

Answers are not stable. They change with the model version, the day, the location, chat memory, whether the assistant searched the web, and the wording of the prompt. Date every result, name the assistant and version, and re-run it later before you call anything a trend.

Repeat answers to one prompt are not independent. Ten runs of one prompt are closer to one data point than ten. The ranges here treat answers as independent, so they are narrower than they should be. The Situations measure counts each prompt once and is the safer one.

A mention is not praise. The tool cannot read tone. "Avoid Fizzo" counts as a mention. Skim the answers.

No brand names in the prompts. A prompt that names your brand tells you nothing about whether it comes to mind unprompted.

This does not show sales. No source found shows that being mentioned by an assistant moves sales. It is a measure of presence in a new place buyers look, nothing more.