How the audit works

The whole method is published here so the result can be checked rather than taken on trust. There is nothing proprietary about it. You could run it yourself, and some people do.

The four questions

The same four prompts are used for every business, filled in with that business name, city and main service. Using a fixed set is what makes one audit comparable to another and to a later re-run.

Tell me about [business name]Filled in: Tell me about [your med spa]Whether the assistant knows the business exists at all.
[business name] in [city]Filled in: [your med spa] in HinsdaleWhether it can place the business in the right town without being told.
best [service] in [city]Filled in: best med spa for lip filler in NapervilleWhether the business is named when nobody asked for it by name.
Who should I hire for [service] in [city]?Filled in: Who should I hire for laser hair removal in Oak Brook?The hiring question, which is the one with money attached.

The first two test whether the assistant knows the business exists. The second two test the harder and more commercially useful question: when someone describes a need instead of a name, does the business get recommended? The examples are med spa work in Chicagoland because that is the category we know best. For any other trade the slots just change, so a plumber gets best emergency plumber in Berwyn and the same grid.

The fixed prompt set4 prompts / 4 engines

The same four questions, every time

Filled in with your business name, your city and your main service. A fixed set is what makes one audit comparable to the next.

01Tell me about [business name]
02[business name] in [city]
03best [service] in [city]
04Who should I hire for [service] in [city]?

Run by hand. Recorded with the date asked.

The four assistants

ChatGPT with web search, Perplexity, Google AI Overviews and Microsoft Copilot. Four questions across four assistants gives sixteen results.

01ChatGPT
with web search enabled
02Perplexity
the default answer model
03Google AI Overviews
where one fires for the query
04Microsoft Copilot
which also feeds Bing answers

Claude and Gemini are not part of the free audit. That is a scope decision, not a judgement about them.

What each result can say

Cited: The assistant named the business and linked or referenced a source for it.
Mentioned without citation: The assistant named the business but pointed to no source. This is weaker than a citation and is reported separately rather than counted as the same thing.
Not found: The assistant answered the question and did not name the business. The report records which businesses it named instead.
No AI answer: The assistant declined, or no AI Overview fired for that query. This is not evidence of invisibility and is never counted as a miss. It is reported as what it is: no answer to read.

What the report contains

Two things, and they are different instruments. Keeping them apart is the point, so neither gets read as the other.

1. The sixteen cell grid

Four assistants down, four prompts across. Every cell carries one of the four states above, the date the query was run, and what was actually said. For every miss, the report names the businesses that came up instead of yours. There is a count of how many of the sixteen named you. There is no number beyond that count, and the grid is never summarised into a grade.

2. The public signal readiness checklist

Seven buckets, each marked pass or fail. Unlike the grid, none of this depends on what an assistant felt like saying on a Tuesday: these are things about your business that anyone can observe on the public web, which is also why they are the things you can change.

Identity: Name, address, phone and the one line description of the business agree everywhere we can see them. Disagreement here is the most common single cause of a business being skipped.; Pass or fail
Site accessibility: The pages an assistant would need can actually be fetched and read, and AI crawlers are not blocked in robots.txt.; Pass or fail
Services and location: The site says in plain text what is offered and where, rather than leaving it to a booking widget, an image or a phone call.; Pass or fail
Structured data: Schema on the site identifies the business, its services and its location in a form a machine can parse without guessing.; Pass or fail
Local profiles: The Google Business Profile and the other profiles that matter for the category exist, are claimed, and are current.; Pass or fail
Proof and reputation: Reviews, third party mentions and coverage that an assistant can find and point at when it decides who to name.; Pass or fail
Expertise content: Pages that answer the questions customers actually ask, written so that a single passage can be lifted and quoted.; Pass or fail

This is not a score. Seven pass or fail marks do not get added up, weighted or converted into a number out of 100, and the checklist is never called a visibility score. A number would have to claim a calibration that does not exist yet. If that ever changes, the inputs, the weights, the things deliberately left out and the limitations will be published on this page first.

How long it takes, and what happens when we are busy

Under normal load the report is emailed within 2 business days. One person opens each assistant, asks each question and reads each answer, and that work does not compress, so the promise moves with the queue instead of staying pretty and becoming false.

Normal load. Two business days, Monday to Friday.
Busy week. Once the week passes the first capacity tier, the site and the confirmation page say five business days instead, before you give up an email address rather than after.
Full week. Past the second tier, new requests go on a waitlist and are told so on the spot, usually clearing the following week.

No request is rejected and no request is quietly dropped. The only thing that changes is the date we are willing to commit to.

What the audit cannot tell you

Assistants are not deterministic. The same question can return a different answer an hour later, to a different person, in a different place. A result is a dated observation of what one assistant said once, not a ranking and not a guarantee. Anyone presenting this kind of check as a stable position is overstating it.

It also cannot tell you how many customers an assistant sent you. No assistant publishes that, and anyone claiming a precise figure for it is guessing.

It cannot get you added to anything. There is no submission route into an AI assistant, from us or from anyone. What the audit does is show what the assistants can currently see and say, which is the part that can be worked on.

If you want the plan written out: the paid audit

The free check tells you where you stand. Some people read it and want the whole fix sequenced for them. That is the AI Visibility Architecture Audit, $1,500 to $3,500 depending on how many locations and services are in scope.

It is a written product, not a retainer and not a subscription. What gets delivered:

an entity map of every place the business is described online and where those descriptions disagree
an information architecture for the site, meaning what pages should exist and what each one is responsible for answering
a schema plan
a list of the source gaps, meaning the pages and profiles that would have to exist before an assistant could cite anything
a prompt test baseline, so a later run can be compared to this one
a 90 day sequence in the order the work should happen

It is offered only to people who ask for it. The form asks whether you want options if we find gaps, and a no is taken at face value. Requesting the free check does not put you on a list for anything.

Request the free audit. A person runs it and emails the report within 2 business days. What we will and will not claim about the results is set out on what we claim and what we do not.