On 27 July 2026, Claw Legaltech was among the first five products named for 12 of 200 buyer questions in its category. On 26 September 2026 it was 171 of 200. Mentions of any kind went from 21 of 200 to 186 of 200, with 377 pages published in between.
Where it started, and why the first reading was misleading
The baseline run asked 78 of the 200 questions on ChatGPT, and ChatGPT confirmed afterwards that it had run no live web search for any of them. That reading measured what the model remembered, not what it could find. The company came second by breadth in that memory, behind one competitor on 28 questions and ahead of another on 18.
The distinction matters more than the numbers. A business can look reasonable in a model's memory and be absent the moment the same model searches, because memory is months old and search is today. Anybody measuring this without checking whether the assistant actually searched is reading the wrong instrument.
The useful part of the baseline was the gap it exposed. On the questions that matter most commercially, about checking a company's court cases and doing litigation due diligence, the company was named nowhere at all. Competitors owned that zone with pages built for those exact questions, and one of them was running the same playbook deliberately, publishing dated pages aimed at individual queries.
The first citation, and the one date with a URL behind it
The baseline run of 2026-07-27 asked 78 of the 200 questions and ChatGPT confirmed it ran no live search for any of them, so that reading is what the model remembered rather than what it could find. The first live citation is the one dated fact in this study that has a URL behind it: on 2026-08-06, fourteen days after the page went up, ChatGPT cited clawlaw.in/blog/how-to-check-a-companys-court-cases-in-india for a vendor due diligence question, in a zone where the same set had named the company nowhere at baseline.
That is the most useful fact in this study, because it is checkable and dated. Not a rate and not a percentage: a named page, a named assistant, the question it was cited for, and 14 days between publishing and being quoted.
It also tells you something about sequencing. The page was published into a zone where the company had no presence at all. Fourteen days later it was the source an assistant used to answer a due diligence question. Nothing about the company's reputation, size or age changed in those two weeks. What changed was that a page existed which answered the question.
What actually did the work
One page per question, not one page per keyword. The pages that got cited answered a single question a buyer types, in the first paragraph, then earned the rest with specifics.
Named counts instead of comfortable words. A page that lists which courts and tribunals are covered, by name and as a count, can be quoted. A page that says "comprehensive coverage" cannot. This was the clearest single pattern in the whole run.
Sourced competitor pricing. Two comparison pages stated competitors' prices with no link and no date. An assistant was able to verify those figures independently and they turned out to be correct, but it had no way to know that at read time, so it treated the pages as advocacy and used official sources instead. Unsourced competitor pricing cost more citations than any other single mistake.
Fixing the site around the pages. Part way through, one assistant searched the company's own site, found a headline figure on the main pages that could not be true, and told a buyer not to buy on the strength of it. The new pages were accurate and the older site around them was not, and the site is what got checked. Correcting that mattered as much as publishing anything new, and it is the reason we now treat a customer's existing claims as part of the job.
The four problems the engines found that we had not looked for
An audit is more useful for what it finds than for the rate it reports. These four came out of the assistants' own answers and none of them was on anybody's list beforehand.
- A pricing page that crawlers could not read. The prices only appeared after scripts ran, so what a crawler received was effectively an empty page. The prices an assistant did quote came from an app store listing instead.
- Two different price lists in public. The website and the app store listing disagreed on both names and amounts. An assistant noticed the contradiction and said so.
- A headline claim that could not be true. Provably impossible against public figures, and used against the company in a recommendation.
- No independent review presence at all. Rated poorly on independent evidence by a neutral engine, in a market where the usual review sites barely operate, so the realistic answer was to be the vendor whose own pages are most verifiable rather than to chase reviews.
Honest limits of this one
This is our own sister company, so treat it as a vendor describing itself. The baseline and the current reading were not taken under identical conditions, because the baseline engine did not search and the later ones did, which means part of the movement is a change of instrument rather than a change in position.
The informational questions in this category, about how to look up a case, still go to government portals, which is correct and will not change. No commercial product should rank above an official court portal for a question about using that portal, and we stopped trying.
Common questions about this run
Can I see the 200 questions?
Not yet, and that is a fair thing to hold against the numbers until we publish them. The sets were built and run by hand before this product existed, and they are being prepared for publication. Until they are, treat these figures as a vendor's own account. The scan on this site is free and it will give you your own number, which you do not have to take on trust.
Why should I believe a vendor reporting on its own client?
You should discount it, and so does every assistant we tested. That is why the most useful number in these studies is not a rate at all. It is the first citation: a named page, a named assistant, the question it was cited for, and 14 days between publishing and being quoted. A dated fact with a URL behind it survives scepticism in a way a percentage does not.
How much of this was the pages, and how much was the assistants changing?
We cannot separate the two with this data, and we are not going to pretend otherwise. The assistants changed what they do during every one of these periods. What we can say is that the movement showed up first on the questions the new pages were written for, and that the first citation in each run was of a specific new page.
Would this work in my sector?
The method worked in all 4 sectors we tried it in, and the ceiling was different in each. The limit is set by who is already trusted for your category. If your questions resolve to government portals or large analyst firms, expect to do well on the buying questions and to keep losing the purely informational ones.
How long before anything moves?
The first citation of a new page arrived between 9 and 16 days after publishing across these 4 runs. A change in the overall rate takes longer, because a rate over 200 questions only moves when many questions move. Expect weeks for the first signal and months for the rate.
What would you do differently now?
Fix the existing site before publishing anything new. In the run below, an assistant searched a client's own site, found a headline figure that could not be true, and advised a buyer against them. The new pages were accurate and the old site around them was not, and the site is what got checked.
What to do first if you are in a similar position
Check whether your own pricing and product pages can be read without scripts, because that is the cheapest and most common failure. Then find every public statement of your prices and make them agree. Then check your largest claim and ask whether somebody could disprove it in one search. Only after those three would we write a new page.
How this was counted
The tracked set is 200 questions, and that is the denominator under every number on this page. The questions are the ones a buyer in this sector actually types, and not one of them names the company. That rule is the whole measurement. A question that names the business gets a polite answer about the business, so it tells you about your prompt rather than about your position.
Each question was asked on 5 assistants, because they answer the same question differently on the same day from different sources. "Named" means the company appears anywhere in the answer. "In the first five" means it appears among the first five products the answer lists, which is the number that matters, because an assistant that names three or four options has no sixth place to give you.
Baseline reading: 27 July 2026. Current reading: 26 September 2026. Pages published in between: 377.
The four things we recorded for every question
- Whether the company was named at all, anywhere in the answer, including in a list rather than as a recommendation.
- Whether it was in the first five products named, and in what position.
- Every source the answer cited, by domain, tagged as official, the company's own site, a competitor's own site, a directory or listing, news, or a community thread.
- Whether the assistant actually ran a live search, asked directly and recorded as a count, because an answer given from memory and an answer given from a live search are two different measurements and averaging them produces a number that means nothing.
What this is, and what it is not
These runs were operated by hand, before this product existed, using the same method the product now runs on a schedule. They are our own results on our own clients, so read them as a vendor reporting its own work. The prompt sets are being prepared for publication so that anyone can repeat the run and check us.
They are also not a controlled experiment. Nobody ran a matched group that published nothing, so we cannot separate our work from everything else that happened in those months, including the assistants changing what they do. What the numbers show is that companies which did this work moved, not that nothing else could have moved them.