Method
How the audit measures, and how sure it is.
The measures behind the Intelligence Audit, written so you can judge the evidence before you rely on the findings.
01
Principles
An audit is only useful if you can see how each finding was reached and how much weight it carries.
- Evidence first. Every finding cites the data, the account or the test that produced it.
- Ranges, not false precision. Where a measure varies, we report the spread and how many observations it rests on.
- Graded confidence. Each finding says whether it was verified, reproduced or inferred.
- Ranked by value. Findings are ordered by expected effect on revenue and by dependency, not by how alarming they sound.
- Nothing changed. The audit reads your systems and does not modify them.
02
How findings are graded
Every line in the report carries one of three grades, so a decision-maker knows what to act on now and what to test first.
| Grade | Meaning | Example |
|---|---|---|
| Verified | Seen directly in data or configuration you gave us access to. | A purchase event fires twice in the tag container. |
| Reproduced | Observed repeatedly in tests we ran, with the method and dates recorded. | A tracking tag loads before consent in five of five clean sessions. |
| Inferred | Modelled from partial data. Reported with the assumption and a range. | Avoidable spend estimated from CRM-traceable conversions. |
03
Measuring AI visibility
Answers from AI systems change from one run to the next, so a single screenshot proves nothing. We measure repeated runs against a fixed prompt set and report the spread.
- Prompt set
- Typically 20 to 40 prompts written with you from real buyer questions in your category and markets, across discovery, comparison and decision. The set is fixed before the first run and shared with you.
- Engines
- ChatGPT, Gemini, Perplexity and Google AI Overviews as standard; others where your buyers use them.
- Runs
- Each prompt is run several times per engine over different days, in clean sessions with the language and country of the target market.
- What we record
- Whether your brand is named, its position among recommendations, the sources cited, and whether the facts stated about you are correct.
- How it is reported
- Median and range across runs, with the number of runs. Competitors are measured with the same prompts and runs.
- Re-measurement
- The same set is re-run at the interval you choose, so change is compared like for like.
Limits: these systems are personalised, update often and are not fully observable from outside. The measurement describes what a typical buyer in that market is likely to see, not what any one person will.
04
Measuring attribution and spend
Platforms mark their own homework. We reconcile what each platform reports against what your CRM and bookings can trace.
- Tag and consent audit: every tag, when it fires, what it sends, and whether consent was given first.
- Conversion reconciliation: platform-reported conversions against enquiries and revenue traceable in the CRM, channel by channel.
- Modelled data is labelled. Where consent mode or platform modelling fills a gap, we report observed and modelled figures separately.
- Avoidable spend is stated as a range with its assumptions, and is never presented as a guaranteed saving.
05
What we do not claim
- We do not guarantee rankings, citations or a share of AI answers. We measure them and work on the sources that influence them.
- We do not give legal advice. Where a finding touches a regulation we cite the article so your DPO or counsel can decide.
- We do not publish client results without written consent. Examples on this site are labelled as examples.