Week 10 · measuring it
The audit: measuring it without fooling yourself
Seven dimensions are only useful if you can measure them the same way twice. This is what separates an audit from a vibe.
From map to measurement
Seven dimensions, measured — or just opinions
The seven dimensions tell you what to look at. An audit is measuring them — and without rigour, "measuring" collapses into an impression you could have had for free. The whole value of an audit is that it produces a baseline you can stand behind and measure change against later. That only holds if it's done in a way you could repeat and a way that doesn't quietly flatter you.
The machinery · what a company can change
The outcome · what the engines say
What makes it trustworthy
Four things, every time
- Reproducible. The same questions, the same conditions, run again — so movement later is real, not a different test. A number you can't re-run tells you nothing.
- Clean-room. Signed out, memory off, fresh session — because logged in, the assistant flatters you with your own history. You want the answer a stranger gets.
- Per-engine. Each assistant audited separately, because they don't agree — and the differences are findings, not noise.
- Human-verified. A person stands behind each call — checking facts against reality — rather than trusting an automated grade. An audit nobody verified is a guess with a number on it.
The traps it has to dodge
Drift, the mirror, and the single sample
Three things sink a careless audit. Drift: engines answer the same question differently on different days, so one reading isn't a measurement. The mirror: checking while logged in shows a flattering phantom. The single sample: one engine, one query, one day tells you almost nothing — the pattern across engines and a real spread of buyer questions is the finding. A good audit is built to dodge all three; a quick check usually walks into them.
What you end with
A baseline, not a verdict
Done properly, an audit hands you a scored, dated, reproducible picture of where you stand across the seven dimensions and the major engines — not a single grand number, but a baseline. That baseline is the thing the rest of the method acts on: you can't correct what you haven't located, and you can't prove a fix worked without a measurement to compare it to.
You can run a rough version yourself — see audit your AI visibility in an afternoon. Next: turning what the audit finds into fixes.
Next — Week 11: Correct & Engineer — turning findings into fixes
New here? Start with the overview → · By Makefield · written in the open