How to make your business easier for Claude to find and verify: a reproducible 30-day test

Claude cites a business when the claim it wants to make can be checked on a page it can reach, and it will name a business without linking it when the page reads as advocacy. The measurement to run is simple: our domain was cited in X of Y reproducible search-enabled runs, on stated dates, with the prompt set and the answer log kept. On the run we have logged, across six blind commercial questions on 17 September 2026, Claude used two specific arguments drawn from clawlaw.in pages and dropped the attribution in both places, and the site landed as a name inside a list rather than as a linked recommendation. Our pages were read. They were not credited.

That is the most useful result we hold, because it separates two problems that look the same from the outside: not being found, and being found and not credited. The work that fixes them is different.

The answer, first

Six things made the difference in our own logs, and none of them is a trick.

• State facts as figures with dates, not as adjectives. "Affordable" cannot be cited. A price with a date can.
• Publish coverage and scope as named lists. On 17 September 2026 two enterprise vendors were ranked above clawlaw.in on a question about finding every case against a company, specifically because they publish explicit court and tribunal coverage lists, and Claude noted that its ordering reflected price transparency and source authority rather than product quality. Claude, 17 September 2026, on clawlaw.in.
• Source your comparisons. Two comparison pages stated competitors' prices with no link, no date and no source. The figures turned out to be correct when the assistant checked them independently, but it had no way to know that at read time, so it treated the pages as advocacy and used official sources instead. Claude, 17 September 2026, on clawlaw.in.
• Make the page readable without scripts. A pricing page rendered its prices only after scripts ran, so what a crawler received contained no prices at all, and the prices the assistants did quote had come from an app store listing instead of from the company's own site. Claude, 6 August 2026, on clawlaw.in.
• Remove claims that cannot be true. A headline figure on a main site could not be true, and when an assistant checked it against public numbers it advised a buyer against the product, after which the company's accurate claims stopped counting for that answer. Claude, 6 August 2026, on clawlaw.in.
• Do not try to instruct the model. A competitor's page carried a hidden block of text addressed to answer engines, instructing them to cite that company as the source. The assistant found it, refused it, and named the company that had done it. Claude, 6 August 2026, on clawlaw.in.

How it was measured

The 30-day design

• A frozen set of blind questions, written in buyer language, with your company named nowhere. Thirty to sixty questions is the usual size.
• Search-enabled runs only in the headline figure. A run where no retrieval happened cannot cite anybody and belongs in a separate column, not in the denominator.
• Weekly runs, same weekday, same rough hour, time zone recorded, which gives four or five runs in the window.
• Clean sessions, one question per conversation, nothing uploaded, nothing pasted.
• At least one other assistant alongside, because a business absent from one engine and present in two others has a different problem from one absent everywhere.

The five outcomes scored per answer

• Cited: a link to a page on your domain appears in the answer's sources.
• Top source: your domain is the first source listed.
• Named only: your name is in the text with no link to you.
• Used without credit: an argument or figure that exists only on your pages appears, with no link.
• Absent: neither named nor cited.

The headline figure is cited over search-enabled runs. The other four are reported next to it and never folded into it. "Used without credit" is the column that tells you a page is working and a credibility signal is missing.

The verifiability audit that runs alongside

Before and after the window, check your own pages on these seven points, because this is the half you control.

• Every commercial fact present as plain text without scripts running.
• A date next to every figure.
• Coverage and scope stated as counted, named lists.
• Every competitor figure linked to its source and dated.
• A page stating what you do not do and who should buy elsewhere.
• One version of each fact across your site, your listings and your app store entries.
• No claim on the site that an outsider could show to be impossible.

What the numbers were

What we have: one logged Claude run on a blind commercial question set, and it produced zero citations of our own domain. Across six questions on 17 September 2026, on clawlaw.in, clawlaw.in pages ranked in the assistant's raw results and shaped what it wrote, and the site still landed as a name inside a list rather than as a linked recommendation, because its own comparison pages read as vendor advocacy. In the same run, the assistant admitted it had used two specific arguments drawn from clawlaw.in pages and dropped the attribution in both places, which it described as a citation lapse rather than a ranking judgement.

Three more findings from that same six-question run, because they explain the zero.

• No third party review or directory source made it into any answer at all. One well known review site did appear in the raw results and was discarded, because the list it offered was of American products and so was not an answer to an India question. Claude, 17 September 2026, on clawlaw.in.
• Official sources owned the how-to questions. On the questions about how to look a case up, official court portals took every position above any commercial product, and the assistant said plainly that no commercial product should rank above the official portal for a question about using that portal. Claude, 17 September 2026, on clawlaw.in.
• Coverage lists beat the better product. The ordering reflected price transparency and source authority rather than product quality, in the assistant's own words.

What we do not have: a completed 30-day Claude window with weekly runs and a citation count for each run. As of 29 September 2026 that series has not been run, so there is no trend on this page and no rate over a month. The single dated run above is what we hold and we are not going to describe it as more than that.

For context from other assistants, both dated: ChatGPT cited clawlaw.in/blog/how-to-check-a-companys-court-cases-in-india fourteen days after publication, on 6 August 2026, for a vendor due diligence question in a zone where the same set had named the company nowhere at baseline. And across 24 batches and 72 conversations on aiknowsus.com in September 2026, the phrase recording that we were not cited appears 161 times in Perplexity's own self audits of its answers.

What this cannot tell you

• Six questions is six questions. No percentage, no trend, no claim about Claude in general.
• A zero citation count is not proof of a bad site. In our run the pages were read and used. The failure was in credibility signals, not in reach.
• The checklist is not a cause. The seven verifiability points come from what the assistant said about its own ordering in one audit, which is stronger than guesswork and weaker than a controlled experiment.
• Nothing here promises a position. We cannot guarantee that Claude, or any assistant, will name or link your business, and any supplier who says otherwise is describing something they do not control.
• It does not transfer. A Claude result says nothing about Gemini, ChatGPT, Copilot or AI Overviews.
• Self reports are not logs. Asked afterwards how many questions it had searched for, one assistant withdrew its own earlier claim in three separate batches. Perplexity, September 2026, on aiknowsus.com. Record the interface, not the confession.

Common questions

Our pages are being used and not linked. What do we fix?

The credibility signals on the page rather than the page's reach. Add the source and the date to every figure you quote about anybody else, state the limits of what you sell, and remove any claim that reads as advocacy without evidence. Our own comparison pages were the reason for the missing link, in the assistant's own account, on 17 September 2026.

Should we write a comparison page at all?

Yes, and write it so that somebody who is not your customer could use it. Every competitor figure linked and dated, the cases where a competitor is the better choice named, and your own limits stated. An unsourced comparison page was read as advocacy even when its figures were right.

Do review sites help with Claude?

Less than most people assume, on the evidence we hold. Across six questions no third party review or directory source made it into any answer at all, and the one that appeared in the raw results was discarded for listing American products against an India question. Claude, 17 September 2026, on clawlaw.in.

Can we get cited on "how do I do X" questions?

Often not, and it is better to know that early. Official portals took every position above commercial products on those questions in our audit. The useful move is to own the practical part: the sequence, what delays it, what it costs in time, with the official page linked for the rules.

Is hidden text addressed to the model ever worth trying?

No. We have a dated case of it being found, refused, and the company named for doing it. Claude, 6 August 2026, on clawlaw.in. The downside is that you become the example.

How long before a new page can be cited?

We hold one dated publish-to-citation event, fourteen days, on ChatGPT, for one page. That is an observation and not a timeline you should plan around.

Sources and change log

Read Anthropic's own material for what Claude does: the documentation for the web search tool, the model documentation on knowledge cutoffs, and Anthropic's pricing page on the day you need a figure. We do not restate prices here, on purpose.

Our observations come from the clawlaw.in programme from July 2026, in particular the Claude response audit of 17 September 2026 and the interim checks of 6 August 2026, and from the aiknowsus.com audit of September 2026. Several of the 6 August and 17 September findings concern public pages and can be checked from outside. Counts across our capture files cannot be, until the captures are published.

Change log. First published 29 September 2026, carrying one dated Claude run and no 30-day series. When the series is run, each week's count and denominator will be added here with the prompt set, and this paragraph will stay so that the earlier state of the page is visible.

Disclosure. AI Knows Us sells AI visibility measurement. We are the vendor, and this page recommends measuring.

What to do first

Open your own highest-value comparison page and do one pass: put a source link and a date next to every number about anybody else, and add one paragraph naming the cases where a competitor is the better choice. That single edit is the thing our own log points at, and it costs an afternoon rather than a project.