How to sell your first AI visibility audit, the part the skill packs skip

The skills are installed, the report looks good, and the first email to a prospect is still unsent. I have heard that exact sentence from three people this year, and I said it myself before that. The audit is the easy half. What stops people is a list of questions that have nothing to do with GEO: who do I write to, what do I say, what do I charge, what exactly do I hand over, what happens on the call.

This post answers those five, in the order you hit them. I run an SEO agency and a set of my own sites, so the prices and the script are the ones I use, not a theory.

The fact that opens the email

Cold email about "AI visibility" dies in the inbox because the words mean nothing to a clinic owner. A fact does not die. The one I use is built on a check anyone can repeat on their phone.

Open ChatGPT. Ask for the best [category] in [city]. Read the names. If the prospect is missing and a competitor is present, you have the email:

I asked ChatGPT for the best physiotherapy clinics in Leeds. It named three of your competitors. It did not name you. I checked why; it takes two minutes to explain.

One fact, one checkable claim, one offer to explain. No deck. The reply rate on that line is not something I will put a number on, because my sample is small and your city is not mine. What I will say is that the replies you get are from people who already ran the check themselves, which is the only kind of lead worth a call.

Twenty of those in week one is the volume target. Each needs its own fact, which means twenty checks. That is the actual work of prospecting, and it is why most people never send the first one.

What you are selling, with the number written down

Price is where the sale is lost, usually on the call, usually in the pause after "how much". The fix is boring: decide before the first email and write it in a package.

The 2026 GEO agency guide at LLM Pulse lists a one-time AI visibility audit at $500 to $2,000 for a small business, covering a baseline, a competitive benchmark and a gap analysis. Xpand Digital's pricing page sits higher, at $1,500 to $5,000 for a one-time audit, and a consultancy I read puts the common band at $500 to $3,000. All three are vendor numbers, so treat them as the market's shape rather than a survey. The shape is wide enough that you have room.

Three packages are enough:

Adjust to your market. What matters is that the number exists before the call, so when the pause comes you read it off the page instead of inventing it. The two allowed discounts are a first-client rate and a case-study rate, both named, both time-boxed. Anything else is you negotiating against yourself.

The 20-minute conversation, and the written version

The call is twenty minutes because the prospect already ran the check. You are not convincing them that the gap exists. You are agreeing on what to do about it.

Minutes one to five: their business, in their words, and the one answer they wish AI gave about them. Minutes five to ten: the gap, on screen, from your check. Minutes ten to fifteen: the Diagnostic, what it contains, what it does not, the number. Minutes fifteen to twenty: the next step, which is a payment link and a date.

Some prospects never take calls. For them the same conversation runs in two emails: the gap with a screenshot, then the package with the number and a payment link. Written sells slower and closes cleaner, because every claim is on the record.

The objection you will hear most is "we do fine on Google". The answer is that the check was not Google. It was the assistant their customers ask when they do not want ten blue links, and the assistant is already answering without them.

Delivery in 48 hours, which means a fixed sequence

A 48-hour promise is only honest with a routine. Mine is a command sequence, a prompt that ranks the fixes, and a gate that stops anything unverified reaching the client.

The sequence starts with access, because the cheapest finding is the one that ends the audit: the AI cannot read the site. The crawler documentation is public and specific. OpenAI lists its bots and what each one does. Perplexity documents PerplexityBot and Perplexity-User. Google's crawler list covers Google-Extended, the token that controls AI training and Gemini grounding. A robots.txt that blocks two of those is a page-one finding, and I check it with the free crawler checker before anything else. The five-minute access audit walks through the rest.

Then the rest of the checklist: entity clarity, citability of the key pages, schema that validates, the brand's presence in answers across four assistants. The AI Search Visibility Toolkit runs that as eleven skills in Claude Code, and the ranking prompt turns the findings into ten fixes ordered by effort and impact.

The QA gate is a short list. Every finding reproduced on the live site, not inferred. Every crawler claim matched to the actual robots.txt. Every number in the report traceable to a check the client could repeat. No guarantees about rankings or mentions anywhere in the PDF. It takes twenty minutes and it is the difference between a report and an opinion.

From one audit to a retainer

The retainer conversation starts with one question on the walkthrough: "Who is going to do these ten things?" Most owners have nobody. That is the retainer, and the 90-day plan is the ten fixes in order, then monitoring, then the next ten.

I packaged all of this because the skills were never the hard part. AI SEO Agency OS is the 41-skill toolkit plus the operating layer: the seven-day schedule, the three packages, the outreach scripts, the call minute by minute, the delivery SOP, the six templates, and a Pro edition where you send me your first paid audit and get written notes back within three business days. The free Found by AI workbook is the client-side version of the same five questions, which makes a decent leave-behind after the call.

Send the first email this week. The fact is two minutes to find, and the rest is a schedule.