Citation Share is the percentage of AI answers to high-intent questions in your market that name your firm. We run 40 to 100 questions per market across ChatGPT, Perplexity and Google's AI Overviews, five times each, logged out and geo-set to your market, and report the mean and the spread.
Brand mentions and linked citations are counted and reported separately. This page is the complete specification, including the things the number can't tell you.
Version 1.0 · Last reviewed August 2026 · Changes are logged at the bottom of this page
Every vendor in this category will soon report some version of "AI visibility." Most will describe the method as proprietary. That's convenient, because an unpublished method can't be audited, can't be reproduced, and can be adjusted whenever the number needs to look better.
We publish ours for the same reason we publish the AEO scoring rubric: a competitor can copy a metric's name, but copying a published methodology means committing to the standard in it. And if you ever want to challenge a number we've given you, the answer should be a URL rather than an assurance.
| Parameter | Value | Why |
|---|---|---|
| Engines | ChatGPT Perplexity Google AI Overviews |
The three surfaces where prospective clients currently ask for a lawyer in volume. Each is reported separately as well as combined — they behave differently and averaging them hides that. |
| Questions per market | 40 · 60 · 100 by plan |
Drawn from real query data for your practice area and metro — the questions people actually ask, not a keyword list. The full question set for your market is visible in your dashboard and doesn't change between runs unless we tell you. |
| Runs per question | 5 | These engines are non-deterministic: the same question asked twice can return different firms. A single run is an anecdote. Five runs give a usable mean and let us report how much the answer moves. |
| Conditions | logged out no memory geo-set to market |
Personalization and chat history change answers. We measure one clearly defined baseline rather than pretending a single number describes every user's experience. |
| What counts | brand mention linked citation |
Counted and reported as two separate numbers, never summed. Being named in prose and being linked as a source are different outcomes with different value. |
| Frequency | monthly · weekly by plan |
Consistent intervals, same question set, same conditions. A metric measured irregularly isn't a time series. |
| Reported | mean + spread | The mean across five runs, plus how widely the result varied. A firm named in five of five runs is in a materially stronger position than one named in one of five, and a single percentage would hide that. |
// The calculation
Citation Share = (answers naming the firm ÷ total answers sampled) × 100Where total answers sampled = questions × engines × 5 runs. A 60-question market across three engines produces 900 sampled answers per measurement cycle. Mentions and citations are calculated independently over the same denominator.
Real queries for your practice area and metro, weighted toward high intent — someone describing a situation and asking who to call, rather than someone researching a definition. The set is published to your dashboard before the first run.
Fresh sessions, logged out, no stored memory, location set to your market. Same conditions every cycle, so changes in the number reflect changes in the answers rather than changes in how we asked.
Every question, on every engine, five times. Full response text is retained so any figure can be traced back to the answers that produced it.
Each answer is checked for your firm's name appearing in the response text, and separately for your domain appearing as a linked source. Near-matches and misspellings of your firm name count as mentions; competitor mentions are recorded in the same pass.
Mention share and citation share, per engine and combined, with the spread across runs. Alongside it: which competitors were named in the answers where you weren't, and which of your pages were cited where you were.
When an engine changes how it generates or sources answers, the series breaks. We mark the break on your chart rather than smoothing across it. A metric that never shows a discontinuity is a metric that's hiding one.
Every measurement has a boundary. Ours are below. If a competitor's AI visibility metric doesn't publish a section like this, it isn't because they don't have limitations.
// On the third one. The honest position is that we're measuring a leading indicator whose relationship to revenue we expect to demonstrate rather than assert. When we have a cohort large enough to report on, we'll publish the numbers, the sample size and the method — including if the relationship turns out weaker than we expect.
Until then, treat Citation Share as what it is: a measure of whether the machines increasingly mediating your market know your firm exists.
Rank tracking tells you your position on a page of ten blue links. Citation Share tells you whether you appear in an answer that names three firms. The distribution is much steeper: a page of results has ten slots, an AI answer has roughly three. Two firms can hold identical rankings and have completely different Citation Share.
Because one run isn't a measurement. Ask an engine the same question twice and it can name different firms — that's how they work, not a fault. Five runs give a stable enough mean to track over time, and reporting the spread alongside it tells you whether your presence in the answer is consistent or intermittent. A firm named in five of five is in a much stronger position than one named in one of five, and a single percentage would hide the difference.
They're different outcomes. A mention means the engine named your firm in the answer text. A citation means it linked your domain as a source. Mentions typically reflect entity recognition — the model associates your firm with the market. Citations reflect a specific page being used to answer a specific question. Summing them produces a bigger number that means less, and it's the easiest place for a vendor to inflate a figure without technically lying.
Yes, and you should. Your question set is published in your dashboard. Open a logged-out session, ask any question from it, and compare what you see to what we reported. Because the engines are non-deterministic, expect your single run to differ from our five-run mean — that variance is exactly what the spread column is reporting.
Gemini and Microsoft Copilot are not in the current set. Both are candidates for a future version, and adding an engine changes the denominator — so when we add one, it starts a new series with the break marked rather than being retroactively mixed into your history.
We don't know yet, and we won't say otherwise. It's a leading indicator: it measures whether AI answers name your firm when someone describes their situation and asks who to call. Whether that converts at a rate worth the investment is something we intend to demonstrate with cohort data and a published method, not assert on a marketing page. See the limitations above.
Yes. In every answer where your firm isn't named, we record who was. Over a cycle that produces a ranked view of which firms in your market the engines currently favour, and — where they were cited rather than just mentioned — which of their pages did it. On the Command plan we track a defined competitor set explicitly.
Any change to the spec is logged here with the date and the reason. Changes that break comparability with prior data are marked on your charts.
v1.0 — August 2026. Initial publication. Three engines, 40–100 questions per market by plan, five runs per question, logged-out and geo-set conditions, mentions and citations reported separately.
Forty questions from your market, three engines, five runs each — exactly as specified above. You'll see where you appear, where a competitor appears instead, and which of their pages the engines are citing.