Fix pack + verification
$2,500, one time. One brand, one category.
5 business days for the fix pack; re-measurement 2 weeks after you confirm the changes are live.
What you get
- ~144 sampled answers across three enginesThe same category questions asked many times on each engine, because one answer is a random draw. A brand can read as “not mentioned” and “mentioned” minutes apart.
- Every rate with a 95% confidence intervalYou get the range, not a single flattering number. Where the data cannot support a claim, the report says so instead of guessing.
- No ranking, and the evidence for whyWe do not report your average position. Independent research repeating the same prompts 2,961 times found fewer than 1 in 1,000 reruns returned brands in the same order. How often you are named survives that; where you placed does not, so we report the first and refuse the second.
- The sources shaping the answersWhich pages the engines actually cited when composing answers about your category — and whether any of them are yours. This is usually where the leverage is.
- Up to three competitors, measured identicallySame questions, same sample, same intervals. So the comparison means something.
- A technical audit of your siteWhether AI crawlers can reach and read you: robots rules, server-rendered content, llms.txt, sitemap and structured data.
- A prioritised fix pack for your developerEvery issue the audit finds, written as a concrete change with a copy-paste-ready snippet, grounded in your actual page markup rather than generic advice. Ranked by impact, and labelled by who can do it — a shop setting, a theme edit, or a developer task.
- We do not touch your siteYour developer implements. We do not take write access to a live storefront, because a bad theme edit breaks checkout and no audit fee covers that. You get the exact changes; they ship them.
- A re-measurement on the identical questionsOnce you confirm the changes are live, we run the same prompt set again — same fingerprint, because a trend across different questions confounds a wording change with a real one.
- The attributable effect, with a controlTest and control prompts are compared so category-wide drift is not credited to the work. If the movement is not distinguishable from noise, the report says exactly that — including when that is not the answer you were hoping for.
What it will not do
It will not promise to raise your score. This is a measurement, not an optimisation retainer. It tells you where you stand and which sources decide your category.
It cannot prove a change caused an effect from one audit. Attribution needs a before and an after with a control. A single report is a baseline.
It measures three engines, not ten. OpenAI, Anthropic, and Anthropic with live web search. If a specific engine matters to you, ask before ordering rather than after.
Some results will read “provisional”. When a rate sits mid-range, the interval stays wide until the sample grows. The report says how many answers it would take rather than pretending to precision it lacks.
Order
Questions before you buy? hello@perceply.com