---
title: "Before You Buy an AI Mention Tracker, Run This ChatGPT vs Perplexity Test"
description: "Test your buyer prompts in ChatGPT and Perplexity, separate mentions from citations, and turn missed citations into specific content fixes."
url: "https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/7de93aea-5da6-4c95-8cd3-a02a153f3b1f.png"
published: "2026-11-19T05:10:36.834Z"
modified: "2026-11-19T05:10:36.889Z"
---

November 19, 2026 · 10 min read

# Before You Buy an AI Mention Tracker, Run This ChatGPT vs Perplexity Test

[](https://groas.com/author/alexander-perelman)

[Alexander Perelman](https://groas.com/author/alexander-perelman)Personally Tested

Head Of Product @ groas

[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![Cartoon of a shop owner testing the same question on two phones, a long ledger spilling to the floor, ignoring a salesman offering a boxed mention-score gauge.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/7de93aea-5da6-4c95-8cd3-a02a153f3b1f.png)

In this article

1. [The question: who gets cited for your buyers’ questions?](#the-question-who-gets-cited-for-your-buyers-questions)
2. [Set up 15–30 buyer questions, then vary the wording](#set-up-1530-buyer-questions-then-vary-the-wording)
3. [Measure four things before diagnosing a miss](#measure-four-things-before-diagnosing-a-miss)
4. [Check fetch logs without asking them to prove too much](#check-fetch-logs-without-asking-them-to-prove-too-much)
5. [Diagnose the miss by reading the page that beat you](#diagnose-the-miss-by-reading-the-page-that-beat-you)
6. [Read the pattern before commissioning the fix](#read-the-pattern-before-commissioning-the-fix)
7. [Buy monitoring after you know what needs diagnosing](#buy-monitoring-after-you-know-what-needs-diagnosing)
8. [Rerun the same test, then change what you commission](#rerun-the-same-test-then-change-what-you-commission)

Before you pay for a tool that lists which prompts mention your brand, run those prompts in ChatGPT and Perplexity yourself. **The useful result is not a mention score. It is a list of missing answers to investigate.**

I would start with the same buyer questions, asked in both engines during the same week, with every result logged. The test shows where each engine cites you, where it cites a competitor, and which answer on your site may need work. That last step turns a symptom report into a working diagnosis.

This is a second look at [_Why ChatGPT and Perplexity Cite Different Pages: Follow the Fetch_](https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same). Knowing why the engines cite different pages helps only once you know which pages they cite for your buyers’ questions. This protocol supplies that measurement so you can fix specific gaps instead of guessing.

## The question: who gets cited for your buyers’ questions?

I would put one question at the top of the sheet: _For my buyer prompts, which engine cites my site, which cites a competitor, and what might explain the difference?_

Log two outcomes separately on every run:

- **Brand mention:** does the answer name your brand?
- **Site citation:** does the answer link to your domain?

A [practitioner’s tracking guide](https://arvow.com/blog/how-to-track-chatgpt-brand-mentions) distinguishes mentions from citations and recommends tracking sentiment and competitor share alongside them. Sensible. A passing name-drop and a linked recommendation are not the same outcome. Neither is a polite dismissal a win.

My expectation is asymmetry. The engines may select different pages for the same buyer question, so a strong showing in one should not stand in for measurement in the other. That is an expectation to test, not a verdict to force onto the sheet.

**Count mentions and citations separately from the first run.** Combining them makes the rest of the exercise harder to trust.

## Set up 15–30 buyer questions, then vary the wording

Build the set from questions a buyer would ask before money changes hands, not from your blog categories. I would use four buckets:

1. **Shortlist:** `best X for Y`
2. **Task:** `how do I X`
3. **Comparison:** `X vs Y`
4. **Problem-first:** a description of the problem without naming a product category

Say you sell HVAC dispatch software. The set might include _best CRM for small HVAC companies with same-day dispatch_, _how do I cut missed calls during peak season_, and _\[your brand] vs \[competitor] for two-truck shops_.

Write each underlying question in two or three ways. Wording sensitivity belongs in the data, not hidden outside it. If you need starting points, use [_9 Buyer Prompts That Make ChatGPT and Perplexity Cite Different Pages_](https://groas.com/post/9-prompt-types-where-chatgpt-and-perplexity-cite-different) and rewrite them in your buyers’ language.

![Lab notebook showing side-by-side ChatGPT and Perplexity citation logs](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/44f477d4-514c-40f2-b37d-a01e3407e4e5.png)

### Lock the conditions before collecting answers

Repeated answers can differ. One published comparison reports that, across three same-day runs, [15% of the sites ChatGPT cited appeared every time, compared with 96% on Perplexity](https://www.nextaura.me/blog/track-brand-in-ai-answers). Those are that source’s results, not a baseline your account must reproduce. They are a good reason not to build a content plan around one screenshot.

For this test:

- Use fresh sessions for each run.
- Keep login status, account, and location consistent within each engine.
- Run both engines inside the same date window.
- Repeat each exact wording three times per engine.
- Record the date and the search mode used; keep the mode consistent within each engine.

**Change the wording deliberately. Keep the other conditions steady.** Three repetitions give you a starting rate, not permission to declare a universal winner.

### Give every run its own row

Use one sheet with these columns:

- Buyer question and prompt bucket
- Exact wording
- Engine, date, search mode, and run number
- Cited URLs
- Your domain cited: yes or no
- Competitor URLs cited
- Your brand named: yes or no
- Context or sentiment of the mention

For scale, 20 buyer questions × two wordings × two engines × three runs produces 240 rows. That is the full repeat test, not a few casual searches over lunch. Keep the set small enough to finish properly.

![Overhead lab notebook comparing ChatGPT and Perplexity citation logs side by side](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/59e3515d-2cf3-4cf3-9a1f-2012b68bf570.png)

## Measure four things before diagnosing a miss

I would calculate **citation rate, cross-engine overlap, wording sensitivity, and repeat stability**. Keep these separate. Each answers a different question.

### 1. Citation rate: how often does each engine cite you?

For each engine and prompt bucket, calculate the share of runs that cite your domain. Do the same for competitors, and record how often neither appears. A response can cite both you and a competitor, so those outcomes are not mutually exclusive.

Keep brand-mention rate beside citation rate, not inside it. An engine can know your name without sending a reader to your site.

I expect comparison and current-information prompts to behave differently from general how-to questions. The mechanism is the need for retrieved information: an answer that requires a current comparison gives the engine a different task from an answer it can give without searching. Let the bucket-level results show whether that distinction matters for your buyers.

### 2. Overlap: do the engines choose the same URLs?

For matching prompts, compare the cited URLs across engines. Record shared URLs and engine-specific ones.

Do not substitute domain overlap for page overlap. Both engines citing the same company is a different finding from both citing its pricing page. The second tells you much more about which answer is doing the work.

### 3. Wording sensitivity: does a small rewrite change the sources?

Compare the variants of each underlying question. Note when a different wording changes whether you appear, which competitor appears, or which page gets cited.

This is why I would write variants before running the test. Otherwise, it is tempting to keep adjusting the prompt until the answer looks flattering. That is prompt grooming, not measurement.

### 4. Stability: does the exact same prompt produce the same citations?

Compare the three repeat runs for each exact wording. If your domain appears once, record one citation in three runs. Do not promote the successful screenshot to the status of a finding and quietly lose the other two.

**Read rates across repeat runs, not isolated answers.** That rule matters most when the results refuse to sit still.

## Check fetch logs without asking them to prove too much

If you have server-log access, filter for `ChatGPT-User` and `PerplexityBot` traffic during the test window. Note which of your URLs appear, then put those observations beside the citation results.

A fetched page that receives no citation gives you something concrete to inspect: the engine reached it, but that run did not select it as a cited source. A citation without a matching logged fetch does not establish how the engine obtained the page. Likewise, no matching log entry does not prove the engine never encountered it.

**Use logs as supporting evidence, not a complete account of retrieval.** They help separate questions worth investigating; they do not explain every miss by themselves.

## Diagnose the miss by reading the page that beat you

For every prompt where a competitor gets cited and you do not, open the cited page. Identify the sub-question it answers and inspect its first 200 words. Then open your closest equivalent.

Does the competitor give pricing for two-truck shops while your page opens with three paragraphs about your founding story? That is a concrete difference to test. It is not proof that authority played no role. It is a better editing brief than “improve AI visibility,” which could mean almost anything and usually costs accordingly.

Assign each miss a working gap category before changing copy:

| Gap                             | What to look for                                              | What I would do next                                      |
| ------------------------------- | ------------------------------------------------------------- | --------------------------------------------------------- |
| **No answer page**              | No URL on your site answers the sub-question                  | Brief a page around that missing answer                   |
| **Answer buried**               | The useful passage sits well below an indirect opening        | Move the answer into the first 100 words                  |
| **Retrieval unresolved**        | No matching fetch appears in the available logs               | Investigate retrieval before assuming copy is the problem |
| **Topic named, answer missing** | The page mentions the subject but never resolves the question | Rewrite the thin section as a self-contained answer       |

A page may have more than one problem. Record the clearest one and the evidence behind it rather than forcing certainty into a dropdown.

**The gap column is a diagnosis to test, not a cause you have proved.** Its job is to turn “we were not cited” into a specific next action.

## Read the pattern before commissioning the fix

High overlap and low wording sensitivity suggest that a small group of pages does most of the work. I would protect those pages and keep five to eight buyer questions as a standing panel rather than expanding the test indefinitely.

Low overlap calls for separate attention to each engine’s results. If ChatGPT cites vendor comparison pages while Perplexity cites listicles or niche directories, record that distinction. Do not assume the same rewrite will address both patterns. Equally, do not assume one page can never win both. That is what the test is there to discover.

High repeat variance changes how I would judge progress. A one-in-three citation rate is neither a clean win nor a clean loss. It is the starting observation for that wording under those conditions.

**Match the action to the observed pattern.** More prompts are not automatically better, and more pages are not automatically the answer.

## Buy monitoring after you know what needs diagnosing

A prompt-tracking tool can be useful. But I would establish this manual baseline before buying a dashboard, because the baseline tells me what the dashboard needs to measure.

Semrush, for example, [describes its Prompt Research feature](https://www.semrush.com/kb/1626-ai-visibility-features?58636949_page=2\&a3d0bcfc_page=5) as a way to search a prompt database by keyword, discover related prompts, and identify which brands appear in those conversations. That can help with discovery. It does not, by itself, explain why a particular passage on your site lost to a competitor’s.

The distinction is **symptom versus diagnosis**. “You were not mentioned” is an observation. “Your comparison page never answers the two-truck-shop question” is an actionable finding to test.

![Ink cartoon of a marketer staring at a dashboard that reports a brand mention without explaining why](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/60e4d52f-652d-4eae-9050-96e2ff4c18e4.png)

Before trusting any platform’s content-gap recommendations, I would ask:

1. **Which exact prompts did you test?** I want the wordings, not just topic labels.
2. **How many runs sit behind the rate?** I want the dates and repeat counts.
3. **Does the gap identify a specific answer on a specific URL?** Or does it stop at “not cited”?

If the product stops at the score, the diagnostic work remains. I would rather buy execution that ships the missing passage than another dashboard that grades its absence.

Use tools to watch the pattern. Make them earn the right to prescribe the fix.

## Rerun the same test, then change what you commission

Rerun the full set every four weeks with the same wordings, conditions, and three-runs-per-engine rule. Between full audits, keep five to eight buyer questions that matter to pipeline as a weekly panel.

**Keep the comparison consistent enough that a rate change means something.** If you change the prompt set and the pages at the same time, you make the result harder to interpret.

For an illustrative comparison, take one buyer question with three wording variants. At three repeats per variant, that gives you nine runs per engine. Moving from one citation in nine ChatGPT runs to six in nine is worth investigating, especially if the improvement appears across variants rather than in one lucky answer. It still does not prove the rewrite caused the movement.

One citation on a Tuesday after an edit is not enough. Look for a shift that persists through repeat runs and the next check. Keep the history in a durable sheet; [_Build an AI Visibility Tracker That Measures Mention Rates, Not Screenshots_](https://groas.com/post/build-your-own-ai-visibility-tracker-in) is the next step if you need a standing tracker.

Either result changes what I would commission:

- **Neither engine cites you:** use the gap column to brief missing answers instead of ordering another batch of generic definition pages.
- **Perplexity cites you but ChatGPT does not:** inspect the ChatGPT misses separately. Test comparison or current-evidence pages where those answers are missing; do not treat Perplexity visibility as proof that ChatGPT retrieval is solved.
- **Both cite you, but wording changes the winner:** inspect whether the page answers each variant directly. Buyers will not standardize their phrasing for you.
- **Both cite you consistently:** protect the working pages and monitor them. Do not commission a rewrite simply because the calendar says content is due.

I would run this test before approving the next content batch. A mention dashboard tells me the score. The sheet tells me what to investigate, what to write, and whether the next round changes the result. That is the standard I bring to groas work: continuous fetching, logging, and rewriting, not a quarterly report about the same unfixed gap.

## Frequently Asked Questions

### Why should I test my buyer prompts in ChatGPT and Perplexity myself before buying an AI mention tracker?

Running the prompts manually shows which engine cites your site, which cites a competitor, and which answers on your site may need work. The useful result is not a mention score but a list of missing answers to investigate. A manual baseline also tells you what a monitoring dashboard actually needs to measure.

### Should I count brand mentions and site citations as the same thing?

No. A brand mention means the answer names your brand, while a site citation means the answer links to your domain, and an engine can do one without the other. Count them separately from the first run, because combining them makes the rest of the exercise harder to trust.

### How many buyer prompts do I need and how should I choose them?

Set up 15 to 30 questions a buyer would ask before money changes hands, not questions based on your blog categories. Sort them into four buckets: shortlist questions like best X for Y, task questions like how do I X, comparisons like X vs Y, and problem-first descriptions that name no product category. Write each underlying question in two or three variants, because wording sensitivity belongs in the data.

### What test conditions should I keep consistent when comparing ChatGPT and Perplexity?

Use fresh sessions for each run, keep login status, account, and location consistent within each engine, and run both engines inside the same date window. Repeat each exact wording three times per engine and record the date and search mode used. Repeated answers can differ, so three repetitions give you a starting rate rather than proof of a universal winner.

### Which metrics should I calculate from the test results?

Calculate citation rate, cross-engine overlap, wording sensitivity, and repeat stability, keeping each separate because each answers a different question. Citation rate is the share of runs that cite your domain per engine and bucket. Overlap compares which URLs each engine cites for matching prompts, wording sensitivity shows whether a small rewrite changes the sources, and stability compares the three repeat runs of each exact wording.

### If both engines cite my domain, is that enough to call it overlap?

No. Both engines citing the same company is a different finding from both citing its pricing page, so record shared URLs rather than just shared domains. Page-level overlap tells you much more about which specific answer is doing the work.

### Can my server logs prove how ChatGPT or Perplexity found or selected my pages?

No. Filter for ChatGPT-User and PerplexityBot traffic and compare which URLs appear, but a fetched page that gets no citation was reached and not selected, and a citation without a matching log entry does not establish how the engine obtained the page. Use logs as supporting evidence, not a complete account of retrieval.

### How do I figure out why a competitor's page got cited instead of mine?

Open the cited competitor page, identify the sub-question it answers, inspect its first 200 words, and compare it with your closest equivalent page. Then assign the miss a working gap category: no answer page, answer buried, retrieval unresolved, or topic named but answer missing. The gap is a diagnosis to test, not a cause you have proved, and it turns we were not cited into a specific next action.

## Run Your Marketing On groas

Plug ChatGPT, Claude or any AI agent into groas and let it do the work of a whole marketing team: $19 a month flat with every feature included, and autopilot whenever you’re ready.

[Start For $19](https://groas.com/pricing)

[Or have groas run it for you](https://groas.com/done-for-you)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","description":"groas is the marketing infrastructure for your own AI: it connects your AI to all your marketing and gives it proprietary marketing data collected 24/7. Plug Claude, ChatGPT or any AI agent into groas and run all your marketing from it, for a tiny fraction of the usual cost, with the groas app there whenever you want to see more. It is the layer ChatGPT, Claude and any AI assistant run a business’s online marketing on. Through groas an assistant can do anything you could do natively in Google Ads, Bing Ads, Meta Ads and ChatGPT ads, on your website and in AI search, and, where the owner switches it on, can run connected campaigns by itself, so AI agents do the work a whole marketing team used to. groas also gives those agents large-scale proprietary marketing data it collects around the clock, on any schedule the owner chooses: every ad competitors run on Google and Meta, their landing pages and full funnels with the offers and prices in them, Reddit customer research, AI search answers from ChatGPT, Google AI Overviews, Gemini, Claude and Perplexity, and SEO data from Semrush and Ahrefs, backlinks included ($19 a month, flat, with everything included, for unlimited teammates on one shared workspace and unlimited ad accounts and websites, delivered as an MCP server). Separately, its done-for-you service has senior people and proprietary quantitative systems run paid ads, SEO and AI search for businesses, and white label for agencies, from $5,000 a month.","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you#webpage","url":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you","name":"Before You Buy an AI Mention Tracker, Run This ChatGPT vs Perplexity Test","description":"Test your buyer prompts in ChatGPT and Perplexity, separate mentions from citations, and turn missed citations into specific content fixes.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/7de93aea-5da6-4c95-8cd3-a02a153f3b1f.png"},"breadcrumb":{"@id":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you#breadcrumb"},"dateModified":"2026-11-19T05:10:36.889Z","mainEntity":{"@id":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"Before You Buy an AI Mention Tracker, Run This ChatGPT vs Perplexity Test","item":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you#article","url":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you","mainEntityOfPage":{"@id":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"Before You Buy an AI Mention Tracker, Run This ChatGPT vs Perplexity Test","description":"Test your buyer prompts in ChatGPT and Perplexity, separate mentions from citations, and turn missed citations into specific content fixes.","datePublished":"2026-11-19T05:10:36.834Z","dateModified":"2026-11-19T05:10:36.889Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/7de93aea-5da6-4c95-8cd3-a02a153f3b1f.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"test, before, run, chatgpt, perplexity, answer, page, each","wordCount":2105,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"My background includes Goldman Sachs and Stanford Computer Science.","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/test-it-yourself-a-repeatable-audit-of-which-prompts-get-you#faq","mainEntity":[{"@type":"Question","name":"Why should I test my buyer prompts in ChatGPT and Perplexity myself before buying an AI mention tracker?","acceptedAnswer":{"@type":"Answer","text":"Running the prompts manually shows which engine cites your site, which cites a competitor, and which answers on your site may need work. The useful result is not a mention score but a list of missing answers to investigate. A manual baseline also tells you what a monitoring dashboard actually needs to measure."}},{"@type":"Question","name":"Should I count brand mentions and site citations as the same thing?","acceptedAnswer":{"@type":"Answer","text":"No. A brand mention means the answer names your brand, while a site citation means the answer links to your domain, and an engine can do one without the other. Count them separately from the first run, because combining them makes the rest of the exercise harder to trust."}},{"@type":"Question","name":"How many buyer prompts do I need and how should I choose them?","acceptedAnswer":{"@type":"Answer","text":"Set up 15 to 30 questions a buyer would ask before money changes hands, not questions based on your blog categories. Sort them into four buckets: shortlist questions like best X for Y, task questions like how do I X, comparisons like X vs Y, and problem-first descriptions that name no product category. Write each underlying question in two or three variants, because wording sensitivity belongs in the data."}},{"@type":"Question","name":"What test conditions should I keep consistent when comparing ChatGPT and Perplexity?","acceptedAnswer":{"@type":"Answer","text":"Use fresh sessions for each run, keep login status, account, and location consistent within each engine, and run both engines inside the same date window. Repeat each exact wording three times per engine and record the date and search mode used. Repeated answers can differ, so three repetitions give you a starting rate rather than proof of a universal winner."}},{"@type":"Question","name":"Which metrics should I calculate from the test results?","acceptedAnswer":{"@type":"Answer","text":"Calculate citation rate, cross-engine overlap, wording sensitivity, and repeat stability, keeping each separate because each answers a different question. Citation rate is the share of runs that cite your domain per engine and bucket. Overlap compares which URLs each engine cites for matching prompts, wording sensitivity shows whether a small rewrite changes the sources, and stability compares the three repeat runs of each exact wording."}},{"@type":"Question","name":"If both engines cite my domain, is that enough to call it overlap?","acceptedAnswer":{"@type":"Answer","text":"No. Both engines citing the same company is a different finding from both citing its pricing page, so record shared URLs rather than just shared domains. Page-level overlap tells you much more about which specific answer is doing the work."}},{"@type":"Question","name":"Can my server logs prove how ChatGPT or Perplexity found or selected my pages?","acceptedAnswer":{"@type":"Answer","text":"No. Filter for ChatGPT-User and PerplexityBot traffic and compare which URLs appear, but a fetched page that gets no citation was reached and not selected, and a citation without a matching log entry does not establish how the engine obtained the page. Use logs as supporting evidence, not a complete account of retrieval."}},{"@type":"Question","name":"How do I figure out why a competitor's page got cited instead of mine?","acceptedAnswer":{"@type":"Answer","text":"Open the cited competitor page, identify the sub-question it answers, inspect its first 200 words, and compare it with your closest equivalent page. Then assign the miss a working gap category: no answer page, answer buried, retrieval unresolved, or topic named but answer missing. The gap is a diagnosis to test, not a cause you have proved, and it turns we were not cited into a specific next action."}}]}]}
```
