The reading, in one paragraph
A set of prompts stands in for the questions your buyers ask. Each prompt is run, the answers are read for a mention of your site, and the mentions are counted.
That is the whole measurement. Everything else is presentation.
A filled reading
Brightwater Dental, twelve prompts frozen on 2026-09-14, run on 2026-09-28.
prompt set 12 questions, version 1, frozen 2026-09-14
branded 6: questions that name the practice
unbranded 6: questions about treatment and cost
run on 2026-09-28
answers collected 12
mentions 7 of 12 58.3%
branded 6 of 6 100%
unbranded 1 of 6 16.7%
arithmetic mentions / prompts, per group
cited pages /emergency, /implants, /childrens, /pricing
absent prompts 3, 4, 5, 7, 9 all cost or safety questions
type estimate
limits small sample, model version not recorded
The branded split is the honest part. Being named for your own name is the easy half. It is reported apart from the half that matters.
Read it as an estimate
The answers change between runs, models and days, so a reading is a sample of a moving target. Record the method beside the number, or the number becomes an anecdote.
Acceptance: the reading is usable when
- the prompt set is frozen, versioned and recorded in full
- branded and unbranded prompts are counted apart
- the run date and the number of answers collected are written down
- cited URLs are listed, not summarised
- the report says estimate, and names each limit
- a later reading compares against the same set, unchanged
Troubleshooting
| Symptom | Likely cause | What to do |
|---|---|---|
| Every branded prompt names us | Expected, and not useful alone | Report the unbranded half separately |
| Two runs disagree wildly | A sample too small to compare | Add prompts before drawing a conclusion |
| Mentions drop to zero | The model, or the prompt set, changed | Re-run the frozen set unchanged |
| An answer quotes a deleted page | The index is stale | Check the page, then re-run |
| A client wants a citation guarantee | Nobody can promise one | Show the method and the sample |
Limits
It cannot promise a citation, and it cannot promise that a citation sends anybody. It also cannot see an answer that was never shown to a person.
The reading is an estimate on every platform, including this one.