The short answer
A single check tells you what one AI answer said at one moment. It cannot tell you whether that answer is stable, whether your absence is systematic, or whether anything you changed had an effect. Three dated runs per question can show a range and a before/after; they still cannot prove causation or speak for every buyer.
Use the single check to see the mechanism on your category. Use the sequence when a decision — a budget, a shortlist, an agency brief — needs numbers with denominators and a date on them.
| What you get | Single check (homepage) | Sequenced diagnostic |
|---|---|---|
| QuestionsWhere they come from | 3, generated from your category and audience | 10, chosen with you at kickoff |
| Runs per question | 1 | 3, spread across 3 days |
| Answers you get | 3 | 30, each with a date |
| Surface | Perplexity Sonar, search-grounded proxy | Logged-out ChatGPT, US English, one location |
| Reading | Automatic string match | Read by hand; mention and recommendation counted separately |
| Competitors | Up to 2 named in the form | 3 named in the report |
| Sources | Cited domains per answer | 30-row audit, every citation classified |
| Stability | One run per question; no spread | Unanimous, swing or flip label per question |
| Change over time | Not measured | Identical question set re-run at day 60 |
| Output | On-screen cards | 8-page PDF, answer-level workbook, 60-minute walkthrough |
| Delivery | Result appears when the check runs | Within five business days of kickoff |
| Price | $0 | $850 one-time |
The check is a demonstration, not a smaller version of the measurement. Two surfaces and two sample sizes produce numbers that are not comparable, and we do not present them as if they were.
What a single check can show
- Presence in one answer. Whether the brand name, or the website domain, appears in the answer text as a whole word or phrase.
- The other names in the room. Up to two competitors you enter, and whether they were named before or after you.
- First position against those competitors. Position 1 means you were named before every competitor you entered. It is not a rank across every product in the category.
- The domains the answer cited. Whether your domain was among them, plus the citation list for that answer.
- The answer text. The excerpt is there so you can read the framing yourself, because the count is not the whole answer.
- A 0–100 sample score. 65% mention rate, 15% citation rate, 20% position score. The formula is on the method page. Treat it as a demo heuristic, not a baseline.
What a single check cannot show
- Stability. One run cannot separate a stable absence from a one-off. You need the same question run again to see a spread.
- A rate. Three generated questions and two entered competitors do not produce mention or recommendation rates across your market.
- Systematic absence. The three questions come from a fixed template — best options, how to choose, alternatives to a named competitor — not from your ten buying questions.
- Recommendation versus listing. A string match finds a name. It cannot tell whether the answer put you forward or listed you alongside nine others.
- Source leverage. One answer’s citations are not a 30-row audit. You cannot see which sources recur across questions or which ones do the work for competitors.
- Change. Without a fixed question set and a dated re-run, there is no before and after to compare.
What the sequence adds
The diagnostic fixes the ten questions before sampling, then runs each one three times in logged-out ChatGPT over three days. Ten questions × three runs is thirty dated answers; the report is delivered within five business days of kickoff. Every answer keeps its date, run number, text and citations. Before it interprets anything, the report labels each question’s stability from its three runs:
| Label | Rule | What it tells you | What it does not |
|---|---|---|---|
| Unanimous absent | Named in none of the three runs. | The clearest gap to work on. | That no buyer ever sees the brand. |
| Unanimous present | Named in all three runs. | Presence is not the problem; read position and recommendation. | That the framing is favourable. |
| Swing | Named in one or two of the three runs. | The answer is sensitive to something the sample does not control. | What caused the difference. |
| Flip | Named in all three, but first position moves two or more places, or the brand moves between “recommended” and “listed”. | Presence is steady; the treatment is not. | Which treatment is the “true” one. |
The label is a reading of three runs, not a statistical estimate. Three runs are a range, not a confidence interval, and “swing” does not mean a 50% probability. The runs are spread across three days, so a difference can come from time, retrieval, a model update or ordinary randomness; the report does not attribute a cause it cannot see. The raw verdicts and the sentence behind each count stay in the workbook, so any label can be challenged.
What changes between day 0 and day 60
The sequence is a calendar, not a dashboard. The question set is fixed at the start, the baseline is sampled over three days, and the same set is run again after the work has had time to land.
| When | What happens | What is recorded |
|---|---|---|
| Day 0Kickoff · 30 min | The ten questions and three competitors are fixed in writing. | The question set, the competitors, the surface and the location. |
| Day 1–3Sampling | Three logged-out runs per question, spread across three days. | 30 answers, each with date, run number, text and citations. |
| Day 4–5Reading | Answers read by hand, citations classified, three actions written. | The 8-page PDF and the 30-row workbook, delivered within five business days of kickoff. |
| Day 30Execution check | We check whether the agreed work shipped. This is not a measurement. | Action by action: shipped, partly shipped, not shipped, and why. |
| Day 60Re-run | The identical question set is run again under the same conditions. | A second dated set of 30 answers and a before/after page with the same layout. |
What the comparison can say. For each question: whether the brand was named in all, some or none of the three runs at day 0 and at day 60; whether its first position moved; whether the cited sources changed, appeared or disappeared; and whether the action was in place by the day-30 check. “No change” is a result. It means the work did not move this surface’s answers, or landed too late to measure, and the report says which.
What the comparison cannot say. That your work caused the change. There is no control group and no randomisation, and other pages, competitors and model updates also change between day 0 and day 60. It also cannot say that the change persists, or that logged-in buyers in another region see the same thing.
Which one to use
- To see the mechanism on your category, or to prepare for a call: the single check. One run is enough to see whether names and citations appear at all.
- To decide whether to fund category placement, a comparison page or a listing: the sequence. It gives counts with denominators, three competitors and the sources behind the answers.
- To judge work you already shipped: the sequence with a baseline. The free check cannot be the “before” because the surface, the question set and the reading are different.
- To watch the same questions after a baseline: monitoring at $350 per month, which samples weekly and logs changes. It is offered only after a diagnostic, because a trend without a starting point is another screenshot.
Questions
Can I just run the free check ten times instead?
You can repeat it, but repetition does not turn it into the diagnostic. It is still one proxy surface, one run per question, three generated questions and automatic string matching. It cannot give you the fixed ten-question set, the hand reading, the source audit or the day-60 comparison.
Why is the diagnostic run on ChatGPT if the free check uses Perplexity?
The free check needs a web-grounded answer for a demo, so it uses Perplexity Sonar as a proxy. The paid sampling uses logged-out ChatGPT because that is the consumer surface the diagnostic measures. The report records the surface and the dates. The two are not comparable, and we do not present the free-check number as a ChatGPT number.
Does the day-60 re-run prove that our work caused the change?
No. There is no control group and no randomisation; other pages, competitors and model updates also change. The re-run shows whether the same question set produced different answers in the same surface, and which sources changed. It does not establish causation.
What happens if nothing changes by day 60?
That is reported as a result. The report says which actions shipped and what the answers did. It does not relabel noise as a win, and it will say if the work landed too late to measure.
Read the sampling method · See the $850 diagnostic · Open the synthetic sample · Run the single check