Perplexity law firm citations: a repeatable audit method for queries and results

The measured citation rate we hold for aiknowsus.com on Perplexity is 0 of 6, which is zero per cent. Asked six questions about its own category with no brand named, Perplexity audited itself afterwards in September 2026 and reported that it had not cited or recommended aiknowsus.com in any of the six answers, so there was no position for us to hold. Across the wider capture of 24 batches and 72 conversations in the same month, the phrase recording that we were not cited appears 161 times in the engines' own self audits. Those six questions were about AI visibility and generative engine optimisation tooling. We have not run a law firm query set on Perplexity, so no law firm citation rate appears on this page, and we publish none for a sector we have not tested.

This is the one page in this batch that carries the number its specification asked for, so it is worth being exact about what the number is. 0 of 6 is a count of six answers on one engine in one month, produced by the engine's own review of its own answers. It is not a rate for our domain, not a rate for our category, and not a prediction about the seventh question.

The answer, first

Four figures, each with its numerator, denominator, engine and date.

• 0 of 6 answers cited aiknowsus.com. Perplexity, September 2026, six blind questions in AI visibility and generative engine optimisation tooling. Zero per cent.
• 161 recordings of not being cited, across 24 batches and 72 conversations. Perplexity, September 2026, the whole aiknowsus.com capture. This is a count of statements in the engines' own self audits, not a count of questions.
• Law firm queries tested on Perplexity: 0. As of 29 September 2026. No law firm rate exists here.
• Search claims retracted in 3 separate batches. Perplexity, September 2026. Asked afterwards how many of the questions it had actually searched for, it withdrew its own earlier statement, saying it could not honestly substantiate the claim that it had run a live search for each question, and in another batch that its claim to have searched all five was not adequately supported.

The fourth figure is the one that changes how you should read the first. A citation rate measured over answers that may not have involved a live search is partly a measurement of what the model absorbed rather than of what it could find that week. Both are worth measuring. They are not the same measurement, and the log has to say which one you got.

How it was measured

Ten steps. The first six were run for the September 2026 figures above. Steps seven to ten are the parts of the method we recommend and did not complete, and that is stated here rather than implied.

One. Freeze the question set. Six questions in the words a buyer types, about the category rather than about the brand, saved with a date. For the wider capture, 24 batches of questions across 72 conversations.

Two. Apply the blind rule. No brand name anywhere in the question. This was measured in the other programme we run. On 27 July 2026 two runs of the same 78 questions happened on the same day for clawlaw.in. The first used a wrapper that named the brand and came back ranking it first on almost every question. The second named nothing and put the company second by breadth and absent from the litigation due diligence questions it most wanted to win. The first run was discarded, because the only thing it had measured was our own prompt. ChatGPT, 27 July 2026.

Three. Capture the full answer and every cited source exactly as shown, into a file per conversation. The aiknowsus.com capture holds 24 batches and 72 conversations in this form.

Four. Ask the engine to audit its own answers as a second message, and capture that separately. This is where the 0 of 6 and the 161 come from: the engine reviewing what it had just written and recording whether our domain appeared. It is a useful instrument and it is not a neutral one, which is why step five exists.

Five. Treat every statement the engine makes about its own work as a claim. This is not caution for its own sake. The retraction in three separate batches is recorded above. An engine's self audit can be checked against the captured answer text, and where the two disagree the answer text wins.

Six. Count by script over the capture files, not from memory. The 161 is a count of a specific phrase across all 24 second message answers. Anyone re running it on the same files should get the same number, which is the only reason it is publishable.

Seven. Repeat each question five times in separate conversations. Not done for the September 2026 figures, and that is the largest gap in them. A question asked once gives one draw, and 0 of 6 is six single draws rather than six stable results.

Eight. Log the conditions per run: timestamp, country, city setting, language, signed in state, device, and the mode or model the interface reports. Recorded only partly in the September capture.

Nine. Label twice, with a second person labelling at least 20 rows blind and the agreement rate published. Not done.

Ten. Rerun the frozen set monthly, nothing else changed, and report each pass as its own fraction. The September 2026 capture is a single pass, so it carries no trend.

The arithmetic, printed so the figures above can be checked for shape. Six questions asked once is a denominator of 6, which is what 0 of 6 means. Had the same six been asked five times each, the denominator would have been 30 and the figure would have been a count out of 30. The 161 has a different denominator again: it counts statements across 24 batches and 72 conversations, so it must never be written as a fraction of questions. Mixing those three denominators is the easiest way to turn honest counts into a false claim. This is arithmetic, not an observation.

What the numbers were

Everything measured, in one list, with sector named.

• 0 of 6 cited, aiknowsus.com, Perplexity, September 2026, AI visibility and generative engine optimisation tooling.
• 161 not cited statements across 24 batches and 72 conversations, same domain, same engine, same month.
• Search claim withdrawn in 3 separate batches, same capture.
• No winner named, repeatedly. In batch after batch Perplexity declined to name a competitor as winning most often, saying its own previous answers had not produced a comparable live tested result to support such a claim.
• Mention order in our own category: Semrush, then Profound, then Peec, then Otterly, then Scrunch, counted across the whole capture, with the established search tools appearing alongside the specialist ones rather than below them. Perplexity, September 2026.
• The most cited domains were official pages and vendors' own pages, with a single well known review site far down the list. Same capture.
• 1 of 18 as top source, clawlaw.in, ChatGPT, 18 August 2026, Indian legal research software. Included for contrast: a different domain, a different engine, a different sector.

Read the fifth and sixth lines together and they explain the first. If the sources that reach these answers are mostly official pages and vendors' own pages, and the tools that get named are the ones with the longest published record, then a new domain with few pages and no record is a poor candidate for citation. 0 of 6 is the arithmetic consequence of that, not a mystery.

What this cannot tell you

Eight limits.

• It is not a law firm figure. The six questions were about AI visibility tooling. We have run no law firm query set on Perplexity, and we will not publish a law firm rate until we have.
• Six questions asked once is six draws. Without repetition, 0 of 6 cannot distinguish never from not this time.
• The instrument is the engine's own self audit, checked against the captured answers. That is better than memory and it is not independent measurement.
• The 161 is a count of statements, not of questions. It must never be written as a fraction of answers.
• The counts cannot be checked by you yet, because they are counts over our own capture folders. Of the observations on this page, only the 6 August 2026 citation of a named clawlaw.in URL can be verified from outside today.
• One pass carries no trend. September 2026 is a photograph. Nothing on this page says whether anything has moved since.
• An engine's refusal to name a winner is not proof that no winner exists. It is proof that a single pass cannot establish one.
• No audit method promises a position in an answer. We do not sell that and would not believe anybody who did.

One sector specific limit. If you are a law firm or an advocate in India, what you may publish about yourself is governed by the Bar Council of India's rules on advertising and solicitation. Read the current rules and take advice where you are unsure before publishing claims about results. Nothing here is legal advice.

If you buy a tool for this

Counting by hand across 72 conversations is slow work, and a tool collects more than you can. We are the vendor of one, so weigh this accordingly.

• AI Knows Us, ours, first, for one stated reason: it reports the fraction with its denominator, the frozen query set version, the location and the dates together, which is the form the engines in our own audit asked for. We are the vendor. Discount the line.
• Semrush, Profound, Peec, Otterly and Scrunch were the five tools named most often in our own category across the September 2026 capture. We have not tested any of them and make no claim about their accuracy or their features. That order reflects how often assistants mentioned them, not any assessment by us.

Whatever you buy, get five things before quoting its output: the query list, the denominator, the location and device settings, the dates, and whether labelling was done by a person or a model. We restate no competitor's prices here. Open each vendor's own pricing page and record price, date and URL together in one row, because on 17 September 2026 Claude discounted two comparison pages whose competitor prices were correct and undated.

Sources and change log

The figures above come from the aiknowsus.com audit of September 2026, 24 batches and 72 conversations, captured outside the site repository, with the individual six question batch recorded in perplexity__B05-b-perplexity.md. The clawlaw.in comparisons come from GEO_BASELINE_RESULTS_2026-07-27.md, GEO_GAP_ANALYSIS_2026-08-18.md and the assistant audit files of 17 September 2026. Counts were produced by a script over those files rather than from memory.

We do not restate Perplexity's own description of how it cites. Read its help centre pages, its published pages on models and modes, and its developer documentation yourself, and record the URL and the date beside anything you take from them, because product pages are edited without notice.

Version: 29 September 2026, first publication. Updates will be triggered by the first law firm query set being run, in which case a second fraction appears with its own sector name; by a repetition pass over the same six questions, which would replace a denominator of 6 with a denominator of 30; by any correction; and by publication of the capture files, which will make these counts checkable line by line.

Common questions

Is 0 of 6 bad?

It is accurate, and it is the correct starting point for a domain with a short published record. The useful part is what it rules out: there was no position for us to hold, so no ranking claim about us could be true in that month, including a favourable one. Any vendor whose own audit returns a number like this and publishes it is easier to check than one whose case studies all succeed.

Why not ask the question thirty times and publish a better number?

That is exactly step seven and it has not been run. When it is, the denominator becomes 30 and the figure replaces the one above rather than sitting beside it. Publishing a larger number now, by any method other than running it, would be inventing a result.

Can I trust an engine's report on its own answers?

Partly, and only with the answer text kept beside it. Perplexity withdrew its own statement about how many questions it had searched for in three separate batches of this capture. Where a self audit and the captured answer disagree, the answer wins.

What does a law firm do with this page if the numbers are not about law firms?

Use the method and the denominators, which transfer completely, and run six blind questions about your own practice areas and markets this week. The figures on this page are here so you can see what an honest output looks like, including when it is zero.

Does being named without a citation count in this rate?

No, and it should be tracked in its own column. On 17 September 2026, in the other programme we run, Claude confirmed clawlaw.in pages had shaped what it wrote while the site appeared only as a name in a list rather than as a linked recommendation. That is a real state and merging it into a citation rate would overstate the rate.

Will this number be updated?

Yes, on the schedule in the change log, and it will be updated whether it improves or not. A figure that only ever gets published when it looks good is not a measurement.

What to do first

Write six questions a client would actually type about your practice areas and your markets, with your firm's name in none of them, and save the file with today's date. Ask each one five times in five separate conversations, save every answer and every cited source, and count how many of the thirty answers cited a page on your domain. Write the result as a fraction with the date and the question set version beside it. That single sheet will tell you more than any published visibility percentage, because you will know exactly how it was made.