How often does Gemini recommend law firms? A reproducible test across named markets and practice areas

As of 29 September 2026 the number of Gemini answers we have tested for a citation to aiknowsus.com is zero, so the citation count is 0 out of 0 and no rate can be calculated from it. We will not publish a rate we have not measured. The nearest measured figure we hold on our own domain is 0 of 6: 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. That was Perplexity, not Gemini, and the questions were about AI visibility tooling, not about law firms. This page gives the whole test, the named markets and practice areas it would run across, and the arithmetic, so that anybody can run it before we do.

There is one more thing to say before the method, because leaving it out would make the rest of the page dishonest. aiknowsus.com is not a law firm. It is a measurement product for AI visibility. Counting how often law firm questions cite our domain would be a test designed to return zero, and reporting that zero as a finding about law firms would be worse than publishing nothing. So the citation count on this page is about our domain on our own category questions, and the law firm query design below is given for law firms to run on their own domains.

The answer, first

Three numbers, and each is exactly what it says.

• Gemini answers tested for a citation to aiknowsus.com, as of 29 September 2026: 0. No runs exist. There is no numerator and no denominator.
• The nearest measured citation count on our domain: 0 of 6. Perplexity, September 2026, on six blind questions about AI visibility and generative engine optimisation tooling. There was no position for us to hold, so no ranking of ours could be reported either.
• The nearest measured count at scale: 161 across 24 batches and 72 conversations. Across the whole aiknowsus.com capture of September 2026, the phrase recording that we were not cited appears 161 times in the engines' own self audits of their answers. Perplexity.

If you were looking for a share of law firm answers that name a law firm, or a share that cites a particular firm, no such figure exists on this page and we have not seen one published anywhere with a query list and dates attached.

How it was measured

This is the protocol. The parts that have been run are named as run and the parts that have not are named as not run, in the results section below.

One. Build the query set from named markets and named practice areas, then freeze it. The market and the practice area are the two variables that decide a law firm answer, so they are stated rather than left implicit. Seven markets worth testing in India: Mumbai, Delhi, Bengaluru, Hyderabad, Chennai, Pune and Ahmedabad. Six practice areas that buyers ask about in different words: company and commercial litigation, insolvency under the Insolvency and Bankruptcy Code, employment and labour, intellectual property, real estate and RERA, and direct and indirect tax. That grid is 42 market and practice area pairs, and you write one to three queries per pair in the words a buyer types.

Two. Apply the blind rule. No firm name in any query. The reason is measured. On 27 July 2026 two runs of the same 78 questions happened on the same day in the clawlaw.in programme. 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 altogether 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. Set the repeat count before you look at anything. Five independent runs per query is our floor. One run cannot tell a firm that appears every time from one that appeared once.

Four. Keep runs independent. A new conversation each time, memory and personalisation off and recorded as off, runs spread across the day.

Five. Fix and record the conditions on every row. Surface, model version as the interface reports it, signed in state, country, city setting, language, device, timestamp.

Six. Log these thirteen fields per run, then one row per firm named: query set version, query id, exact query text, market, practice area, run number, timestamp, surface, model version, signed in state, city setting, device, screenshot filename; and per firm, the firm name, whether a supporting link was shown, the domain of that link, and whether the destination page actually mentions the firm.

Seven. Record any search claim as a claim, not a fact. In the aiknowsus.com audit of September 2026 Perplexity withdrew its own earlier statement about how many of the questions it had actually searched for, saying it could not honestly substantiate the claim that it had run a live search for each one, and in another batch that its claim to have searched all five was not adequately supported. Three separate batches produced that retraction.

Eight. Label twice, with a second person labelling at least 20 rows blind, and publish the agreement rate.

Nine. Rerun the frozen set on a fixed interval, nothing else changed, and report each pass as its own fraction.

The denominator arithmetic, printed so any figure we publish can be checked for shape: 42 pairs at one query each, five runs per query, is 210 runs, and 210 is the denominator for a study wide count. Each pair also has its own five run denominator, and each market has 30 runs across its six practice areas. A domain cited in 6 of 210 runs might be 5 of 5 in one pair and absent from the other 41, which is a completely different situation from six thin appearances. Report the pair level first. This is arithmetic, not an observation.

What the numbers were

Gemini: not run. 0 answers tested as of 29 September 2026. No estimate goes in that cell, no range, and nothing borrowed from another study.

Every measured figure we hold, with its engine, its date, its sector and its denominator.

• 0 of 6, aiknowsus.com. Perplexity, September 2026, six blind questions in AI visibility and generative engine optimisation tooling. Not cited or recommended in any of the six answers, by the engine's own account of its own answers.
• 161 statements across 24 batches and 72 conversations, aiknowsus.com. Perplexity, September 2026. The count of times the engines recorded, in their own self audits, that we were not cited.
• 1 of 18, clawlaw.in. ChatGPT, 18 August 2026, eighteen blind commercial questions in Indian legal research software. Top source on exactly one, with the company's own comparison page named in the answer as still being the vendor's own editorial page.
• 28, 21 and 18 out of 78, clawlaw.in. ChatGPT, 27 July 2026, blind breadth order: ProVakil 28, CLAW 21, Legistify 18.
• 0 of 78 searched, clawlaw.in. Same run, same date. ChatGPT confirmed it had performed no live web search for any of the 78 questions.
• 0 of 6 for third party sources, clawlaw.in. Claude, 17 September 2026. No third party review or directory source reached any answer across six questions, and one well known review site was dropped from the raw results because the list it offered was of American products.

One finding from an engine rather than from us belongs here, because it is the best argument against the page you were probably looking for. In batch after batch of the September 2026 capture, 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. An assistant refusing to rank from a single pass is a clearer statement of the limit than anything we could write.

What this cannot tell you

Seven limits.

• Nothing here is a Gemini measurement. Every figure came from ChatGPT, Claude or Perplexity. The systems select sources differently and a result from one is not evidence about another.
• Nothing here is a law firm measurement either. The two programmes behind these numbers are a legal research software company and a measurement product. The sector is named on every line for that reason.
• 0 of 6 is a fact about six questions in September 2026. It is not a rate for our domain and not a rate for any category.
• One observation is not a median. The only timing figure we hold is one citation of a named clawlaw.in URL by ChatGPT on 6 August 2026, fourteen days after publication. Fourteen days is a single observation, not a median time to first citation, and we will not present it as one.
• Most of these counts cannot be checked by you yet. The 6 August 2026 citation of a named URL can be verified from outside today. The counts over our capture folders cannot, until the captures are published.
• A change after a change is not a cause. An engine can change its own behaviour in the same month a firm publishes new pages.
• No test here can promise a position in an answer. We do not sell that and we would not believe anybody who did.

One more limit that applies specifically to law firms in India. What a firm or an advocate may publish about itself is governed by the Bar Council of India's rules on advertising and solicitation, and those rules matter more to your page than any visibility method. Read the current rules yourself, and take advice where you are unsure, before publishing claims about results or clients. Nothing on this page is legal advice.

Sources and change log

Every figure above traces to one of two recorded bodies of work. The clawlaw.in programme from July 2026, recorded in GEO_BASELINE_RESULTS_2026-07-27.md, GEO_GAP_ANALYSIS_2026-08-18.md and the assistant audit files of 17 September 2026. And the aiknowsus.com audit of September 2026, 24 batches and 72 conversations, captured outside the site repository. The counts were produced by a script over those files rather than from memory.

We restate no vendor documentation on this page. Read Google's Gemini Apps help pages and Google's developer documentation for the Gemini models yourself, on the day you read this, and record the URL and the date beside anything you take from them. We also restate no competitor's prices anywhere on this site: open the vendor's own pricing page and record price, date and URL together in one row, because on 17 September 2026 Claude discounted two clawlaw.in comparison pages whose competitor prices were correct and undated.

Version: 29 September 2026, first publication. The page will be updated when the first Gemini run set is completed, in which case the 0 out of 0 becomes a real fraction with a query set version and dates; when any figure is corrected; and when the capture files are published, which will move several rows from unverifiable to checkable. Corrections are made on the page and noted here rather than edited quietly.

Common questions

Why publish a page with no number in it?

Because the alternative is a number we did not measure, and that is the thing this whole site tells readers to distrust. A protocol with an honest empty cell can be checked, copied and run. A rate with no query set behind it cannot be checked by anybody, including us.

Can a law firm be cited by an AI assistant at all?

A page can be used as a source if it is indexed and readable in the form a crawler receives. Whether it is selected is a separate matter with no published rule. The nearest evidence we hold that a specific page can move from absent to cited is one record: ChatGPT cited a named clawlaw.in URL on 6 August 2026, fourteen days after publication, in a zone where the same question set had named the company nowhere at the 27 July baseline. One page, one engine, two dates.

Which practice areas should a firm test first?

The two or three you actually want more of, in the markets you actually take instructions in. A grid of 42 pairs is a complete study and a poor first week. Six pairs, five runs each, gives you 30 runs and a dated baseline you can act on.

Does the engine's own account of what it searched count as evidence?

No. Log it and mark it unverified. Perplexity withdrew exactly that kind of statement in three separate batches of the September 2026 capture, saying it could not honestly substantiate its claim to have run a live search for each question.

Our firm is named but never linked. Is that progress?

It is a different state and it should be counted separately, not added to your cited count. On 17 September 2026 Claude confirmed clawlaw.in pages had shaped what it wrote while the site appeared only as a name in a list, and in the same run admitted it had used two arguments from those pages and dropped the attribution both times. Being used without being linked is real, common, and invisible to tools that watch only for your name.

How long before a new page shows up in answers?

We do not know, and the one data point we hold is fourteen days for one page on one engine on 6 August 2026. That is not a median, not an average and not a promise. Anybody quoting you a typical time to first citation should be asked for their denominator and their dates.

What to do first

Pick three practice areas and two markets, write six queries in the words a client uses, and make sure your firm's name appears in none of them. Run each query five times in separate conversations today, logging the thirteen fields, and record for each firm named whether a link supported it and whose domain that link was on. Then read the Bar Council of India's current rules before you publish anything new about results. By the end of the day you will hold 30 runs and a dated fraction that belongs to your firm, which is the only kind worth quoting.