How to Run an AI Visibility Audit (Full Protocol, Free)
This is the whole protocol. It takes about 45 minutes per business once you have run it twice, and it produces a document you can charge $500–$1,500 for. Nothing here is withheld — the paid workbook automates the scoring, but the method below is complete on its own.
Step 0 — Pick the right subject
The audit works on any local business, but it only sells to businesses where a single additional job is worth real money. Roofing, HVAC, legal, dental, med spa, remodeling, commercial cleaning contracts. Average job value above $2,000 is the practical cutoff. Below that, the retainer maths never closes.
Step 1 — Control the session
This is the step people skip, and skipping it invalidates everything downstream.
- New browser profile, or a private window with no extensions.
- Logged out of every AI account you can be logged out of.
- Location set explicitly in each tool rather than inferred.
- No prior conversation in the thread — start clean for each prompt.
Generative engines personalize. If you are logged into an account that has been researching one business all week, the answers will flatter that business. Your audit then measures your own browsing history.
Step 2 — Choose the prompt set
Six intent categories, ten prompts each. The categories matter because they surface different failure modes:
- Emergency — "who can come out tonight for X in {city}". Tests presence in urgency answers, where hours data and review recency dominate.
- Comparison — "{Competitor} vs alternatives in {city}". Tests whether the business exists in the engine's model of the competitive set at all.
- Price — "what should X cost in {city}". Tests quotable content; engines pull from whoever published a real number.
- Trust — "licensed and insured X near me". Tests corroboration and credential data.
- Specific-need — "X that takes {insurance}" / "X for {niche situation}". Tests service specificity.
- Local-general — "best X in {city}". The headline prompt, and the least diagnostic — everyone wants it, it moves last.
Use the same set every month. A changing prompt set produces a trend line that means nothing.
Step 3 — Run it
Four engines × sixty prompts × three runs is 720 data points, which sounds enormous and is not: you are pasting prompts and scanning for names. Batch by engine, not by prompt — switching tools is the expensive part.
For every run, record:
- Named? Yes or no. The business, exactly.
- Named instead. Every other business the answer mentions, in order.
- Cited source. Whatever the engine credits — a directory, a review site, a local publication, the competitor's own site.
Field three is where the fix list comes from. If nine of ten answers in a market cite the same regional directory, you now know exactly which listing to fix first.
Step 4 — Score it
Nine factors, weighted. A workable weighting for local service businesses:
| Factor | Weight |
|---|---|
| Entity consistency | 20 |
| Third-party corroboration | 15 |
| Review corpus depth and recency | 15 |
| Structured data coverage | 12 |
| Quotable on-site content | 12 |
| Service and location specificity | 10 |
| Aggregator dependency (negative) | 8 |
| Freshness signals | 5 |
| Answer-surface presence | 3 |
Score each 0–5 against defined criteria, multiply by weight, total out of 100. The absolute number matters less than the fact that it is reproducible: the same auditor scoring the same business twice should land within a few points.
Write the scoring criteria down before you start. Undocumented judgement is not a methodology, and a client who senses improvisation stops paying.
Step 5 — Calculate revenue at risk
Four inputs:
- Monthly search volume for the category in the market
- Estimated share of that demand resolving inside an AI answer (be conservative — a defensible range today is 15–35% depending on demographic)
- The business's historical close rate on inbound leads
- Average job value
Multiply. Present it as a range, label every input as an estimate, and show the arithmetic. An owner who can see the maths can argue with one input instead of dismissing the whole number.
Step 6 — Write the document
Five pages, in this order:
- The finding. One sentence: across 60 prompts and 4 engines, {Business} was named in N of 180 answers.
- Who is being named instead. The competitor frequency table. This page does most of the work.
- Why. The scored nine factors with the two or three that are dragging the total.
- What it is costing. Revenue at risk, with the inputs shown.
- What would change it. The 30-day fix sequence, in dependency order, with an honest note on what is fast and what is not.
Deliver it as a PDF with their name on the cover. Not a call, not a Google Doc, not a Loom. A document gets forwarded to the partner who controls the budget; a call does not.
Step 7 — Price it
Charge $500–$1,500 for the audit. Free audits get skimmed and filed. Paid audits get read, argued with, and forwarded — all three of which are what you want.
Then the fix is $2,500–$6,000, or month one of a $1,500–$3,000/month retainer where the monthly deliverable is the re-run plus the trend.
What the paid version adds
Nothing here is held back. The workbook adds the weighted scoring with conditional formatting, the prompt log structure, the entity matrix, and the revenue model as live formulas — so a 45-minute audit stays 45 minutes instead of becoming an afternoon of spreadsheet building. That is the Audit Kit, $27.
The full System adds the 30-day fix protocol, five vertical playbooks with real deal sizes, the pricing and contract language, and the monthly client report template. $97.