A finished audit, page by page.
This is a specimen audit for a fictional early-stage SaaS — "a session-replay tool for small product teams." Category, prompts and structure are exactly what clients receive; the data is a demonstration set, marked as such on every page. Your audit looks like this, with your numbers.
The number everything hangs on.
Forty-two prompts, five engines, 210 logged answers. The scorecard reduces them to one honest figure per engine: how often you appear at all, and how high in the answer when you do.
In this specimen, the subject appears in 3 of 42 prompts (7% share of answer) while the category leader appears in 31 (74%). That gap — not a keyword ranking — is what the engagement exists to close.
| Engine | Prompts cited | Share |
|---|---|---|
| ChatGPT | 1 / 42 | 2% |
| Perplexity | 2 / 42 | 5% |
| Gemini | 0 / 42 | 0% |
| Claude | 0 / 42 | 0% |
| AI Overviews | 0 / 42 | 0% |
| Blended share of answer: 7% · category leader: 74% | ||
Every prompt, on the record.
The log groups prompts into four clusters. A sample of each — in the real document, every row links to the full captured answer and screenshot.
- "best session replay tools 2026"absent ×5
- "session replay for startups"absent ×4
- "affordable hotjar-style tools"cited ×1
- "[leader] alternatives"absent ×5
- "[leader] vs [rival]"absent ×5
- "cheaper than [leader]"absent ×5
- "watch user rage clicks tool"cited ×2
- "replay + privacy for EU users"absent ×5
- "is [subject] legit?"thin answer
- "[subject] pricing"outdated
Where the answers come from.
Across the 210 answers, the engines leaned on 61 distinct URLs. Ranked by how many answers each fed, the top of the map decides the work:
- Review platform, category pagefed 38 answers · subject absent
- "Best tools" listicle (updated 2026)fed 31 · subject absent
- Community thread, "what do you use?"fed 19 · subject absent
- Leader's own comparison pagesfed 17 · hostile framing
- Subject's websitefed 2 · nothing quotable
The diagnosis memo translates this: the subject isn't being skipped by accident — it's absent from every source the engines trust, and its own site gives them nothing extractable to quote.
Ranked fixes, with owners and dates.
| Fix | Why | Effort |
|---|---|---|
| Create review-platform profile + first 10 honest reviews | feeds the #1 source | 2 wks |
| Publish "[subject] vs [leader]" comparison page with current pricing | quotable, owns displacement prompts | 1 wk |
| Fix Organization/Product schema + entity disambiguation | engines mis-resolve the name | 2 days |
| Add llms.txt + verify AI crawler access | three crawlers currently blocked | 1 day |
| Pitch inclusion in the 2026 listicle (it updates quarterly) | feeds the #2 source | outreach |
| Answer the community thread honestly, with disclosure | feeds the #3 source | 1 hr |
| Full worksheet: 23 lines · 9 client-side · 14 we can run in a Sprint | ||
The delta is the report.
Sprint engagements re-run the identical prompt log monthly. The specimen below shows the format: same prompts, same engines, movement stated plainly, every figure re-checkable in a chat window.
We don't promise a specific delta — engines are probabilistic and categories differ. We promise the measurement will be honest, and the founding-client terms on the rate card put teeth behind run 01.
| Measure | Run 01 | Run 04 |
|---|---|---|
| Blended share of answer | 7% | illustrative ↑ |
| Engines citing ≥1 prompt | 2 / 5 | illustrative ↑ |
| Reputation prompts answered correctly | 0 / 6 | illustrative ↑ |
| Specimen format — your report carries your verified numbers | ||