Reading a loss: not named by either app on a coverage question

Losing one question while winning another, on the same day, points at a claim rather than at visibility. In the scan on our proof page, one question asks which platform searches Indian court and tribunal records most completely. Put blind to ChatGPT and to Perplexity on 2026-10-16, it came back from both without clawlaw.in in it. ChatGPT named no business at all after reading 6 sites. Perplexity named Manupatra and SCC Online after reading 8. The same company was named first by both apps on a pricing question in the same run, so this is not a company the apps cannot find.

Why an uneven result is diagnosed differently from a flat zero

A business that is named nowhere has a presence problem: there is too little written about it in the places these systems read. A business that is named on some questions and not others has something more specific, because the retrieval clearly reaches it. On the questions it loses, the pages that get read do not contain the claim the question is asking about.

That distinction decides what to do next, and the two courses of action have almost nothing in common. If you are named on nothing at all, the work starts with the fact base rather than the blog, and what to fix first when you are named on nothing sets out that order. If you are named on some questions, the work is on the specific claim the lost question asks for.

What a comprehensiveness question is actually asking for

A question about the most comprehensive coverage in a category is asking for a list. Not an adjective, not a positioning line, but an enumeration: which courts, which tribunals, which record types, how far back, updated how often. A system answering that question needs something to compare, and it can only compare what is published.

So the two outcomes in this row are both explicable. Perplexity named the two publishers whose coverage is most widely documented. ChatGPT named nobody, which usually means it found no source it was willing to treat as settling the comparison. Neither of those is a verdict on any product. Both are readings about what the category has written down.

The asymmetry with the pricing question is instructive. A price is a single figure that a seller publishes about itself, so a page carrying it is easy to find, easy to quote and hard to dispute. A coverage claim is a set of facts about somebody else's data, it goes out of date silently, and it is the kind of claim a careful system will not repeat from a seller's own marketing page. The question that was won needed one number. The question that was lost needed a maintained list, and almost nobody maintains one.

The four things a lost question tells you, in order of usefulness

Reading	What to do with it
Who was named instead	Read their pages. The claim they publish is the one the answer used
How many sources were read	A high count with no business named suggests the comparison is unsettled
Whether the same question loses on both apps	Agreement on a loss points at the published record rather than one system
Whether you win adjacent questions	If you do, this is a claim gap and not a presence gap

In this case all four point the same way. Both apps failed to name the company, the source counts were 6 and 8, and the same company won the pricing question on both apps minutes earlier.

Turning that into work

• Publish the enumeration the question asks for, as a list with a date on it, rather than a sentence describing it.
• Keep that list current and say when it was last checked, because a restated coverage list goes stale into a false claim faster than almost anything else on a site.
• Get the same facts stated somewhere you do not own, because an answer assembled from several sources is harder to assemble without you.
• Re-run the identical question on both apps after the page exists, rather than a reworded version of it, so the second reading is comparable to the first.
• Keep the losing question in the tracked set permanently. A question you remove because you lose it is a question you have stopped measuring.

The record of which domains held the slots, question by question and app by app, is the asset that makes the second reading meaningful, and keeping the log rather than the verdict lists the fields worth recording for each answer.

What this reading is not

It is one question, on two apps, on one date. It is not a rate, and 6 of the 10 answers in that scan named the company, which is a count travelling with its denominator rather than a percentage. It also says nothing about Gemini, Claude or Google AI Overviews, which were not in this scan. Our separate ChatGPT and Gemini audits are different datasets with their own dates and sizes, and none of the four gets added to another.

Questions about losing a question an AI app answers

Does a loss mean the product is worse than the named alternatives?

No. It means the sources that were read did not contain a comparison that put it first. A product can be the best in a category and absent from the answer if nobody has published the specifics the question asks about.

Why would an app name nobody at all?

Usually because the sources it read disagree or none of them makes a comparative claim it is prepared to repeat. A generic answer is a different failure from an answer naming a competitor, and it needs a different fix.

Should I rewrite the question so it is easier?

No, and this is the main way a tracked set stops being useful. Keep the question a buyer would actually type, and keep it fixed, or your next reading measures a different thing.

When can a new page be judged on a result like this?

It depends on whether the app searches the web while answering or uses what the model already holds, and those are two pipelines with different waiting times. Re-run the question on a stated interval rather than watching for a day.

What do you do about this one?

We sell the measurement and the pages that follow it, so the honest answer is that a lost coverage question becomes a published coverage list with a date on it, and then the same question gets asked again and the count is reported whichever way it comes back.

The loss is the instruction

A won question tells you to keep doing something. A lost question tells you what to build, and it names the thing: in this case an enumeration somebody can compare, in a category where two publishers already have one. That is a more actionable output than any aggregate score, which is why a scan that reports question by question is worth more than one that reports a single figure, even when the single figure is higher.