Skip to content
Free · AI visibility report

The AI visibility report.

A full audit in two parts, organic search and AI answers, benchmarked against three to five competitors you name. Findings are numbered and ordered by what they cost to fix, every figure carries the tool and date that produced it, and the report states what we couldn’t see from outside.

Get your free auditSee it on a real client

A short form, then we run the audit and walk you through it on a call. The $1,500 is the value we put on the work in it; nobody has ever been charged it.

Vignette. A name being marked inside a generated answer. Illustration, not measurement: no figure appears in the loop.
Price
$0
the audit is free, and the report is yours whatever you decide afterwards
Caldrin · published offer · 2026-08-15
Questions
50+
written for your category, then fixed, so later readings can be compared with the first
Caldrin · stated scope, minimum per account · 2026-08-15
Engines
3
ChatGPT, Gemini and Google AI Overviews, each asked repeatedly. Perplexity and Claude are named as not covered
Caldrin · stated scope · 2026-08-15
Competitors
3–5
named by you, benchmarked by name rather than anonymised
Caldrin · report scope · 2026-08-15
Why it costs nothing

Free, because the report is the argument.

Of the six providers we reviewed in August 2026, five either charged for the first assessment or routed it to a sales call. The sixth is self-serve software with no assessment to gate. We give ours away. The fastest way to show you something is measured properly is to measure it.

There is a limit, and it’s capacity. A real audit takes most of a working day, so a short qualifying form comes first. If we’re not the right fit, you hear that at the form stage rather than after.

Usually
Gated
Paid up front, or held behind a sales call before you see anything.
Five of six providers reviewed, August 2026 · access, not price
Caldrin
Free
The same full audit, both parts, benchmarked against three to five competitors you name.
You keep the report and the raw measurements either way
Part one · organic search

Where you actually stand in Google.

Rankings for the searches that pay, checked town by town where your business is local. Then the index itself. Pages built to catch searches routinely sit outside it, earning nothing, so we check yours one URL at a time. Every dead page gets listed.

This half usually carries the cheapest fixes in the whole report. A page Google can’t see costs nothing to argue about and everything to leave alone.

Vignette. A position climbing an index across recaptures. Illustration only: the firms in it are invented, and no figure appears inside the loop.
Part two · AI answers

A fixed question set, run across ChatGPT, Gemini and Google AI Overviews.

We write the questions your buyers actually ask, fix the set, and run it repeatedly per engine at the same point in the month. The set never moves. That is what makes a second reading comparable.

  • A question set written for your category, then fixed, so a second reading can be compared with the first
  • No brand names inside the questions, ever
  • Fresh, logged-out sessions, with no account history carried in
  • Every question asked repeatedly per engine, so you can see how much the answers move on their own
  • Perplexity and Claude can't be measured within their rules, so they're listed as not covered rather than quietly left out
Vignette. One question fanning out across the engine set. The engine roster and question count live in the copy beside it, never inside the loop.
Then the count

Every answer scored, against competitors you name.

Competitors are named in the report rather than anonymised into a benchmark. If a rival is winning the category you will read their name, which is the only version of this that is useful to you.

The same count separates the two failures a single score hides. AI never pulls you up at all, which means the sources it trusts don’t vouch for you yet. Or it pulls you up and picks someone else, which means your pages don’t give it a usable answer. They need opposite work. The report tells you which one you have.

Vignette. Question outcomes sorting into cited and not cited. A proportion emerges without printing a number; a measured share is always set in type, sourced and dated.
Deliverablenumbered findings, plus the raw measurements

What the report contains.

  • Where you actually rank, for the searches that pay, checked town by town where your business is local rather than averaged into one number.
  • Whether Google holds your pages at all. Pages built to catch searches routinely sit outside the index earning nothing. Checked one URL at a time, with every dead page listed.
  • How often AI names you, across ChatGPT, Gemini and Google AI Overviews, against the competitors you nominate.
  • The specific questions where a competitor is named and you are not. The actual questions, with the actual answers, dated.
  • Which of the two absence problems is yours. AI never pulls you up at all, which is an authority problem, or pulls you up and picks someone else, which is a content problem. They need opposite work, and most reports never separate them.
  • Whether machines know who you are: if your name, services and service area resolve to one business everywhere, and who you're being confused with if not.
  • Numbered findings, ordered by what they cost to fix, with the reasoning attached to each one, an appendix tracing each number back to the tool that produced it, and a section on what we couldn't see from outside.

You also get the raw measurements: questions, dates, engine, run counts. If you take the report to another agency, everything in it is checkable by them.

What a finding looks like

We show the answer, not a score.

A score tells you where you sit on someone else’s scale. The answer tells you what a buyer actually reads, which competitor gets named instead of you, and the exact wording you have to get into.

Every finding in the report is built this way: the engine, the question as asked, the capture date, and the answer stored unmodified so you can read it rather than take our word for it.

ChatGPT2026-07-31
For independent clinics in the Chicago area, people often mention Northside Reproductive Health alongside two larger networks.
prompt
“who are the best independent fertility clinics in Chicago”
captured
2026-07-31
Illustrative capture. The brand, the question, the engine and the date are placeholder; a real report reproduces the stored answer verbatim, with its capture date.
Scopesix pillars, each read separately

The six pillars.

They are kept apart on purpose. Fusing signals that have opposite fixes into one composite score is the thing that makes most visibility reports unusable.

How often AI names you
How often each named firm appears in AI answers to your category's questions. Measured with our own tracking, which is the only source we permit for that number.
Whether machines can read your site
Whether Google and the AI crawlers can reach and understand your pages at all: access, rendering, structure. Cheapest pillar to fix, and the one most often missed.
Earned and paid media
The coverage and brand-search signals that feed authority. Paid ads don't put you in an answer directly, though the brand search they generate is a real input.
Reputation and consensus
The third-party sources a model draws from: comparison pages, forums, reviews, industry media. Slowest pillar to move, and the likeliest explanation for a stubborn absence.
Conversion readiness
Whether attention that arrives can become revenue. A mention that lands on a page which doesn't convert is a vanity result.
Competitive benchmark
The five pillars above set beside the competitors you nominate, so every finding is relative rather than abstract.
What this cannot tell you

What it is not.

It is not a ChatGPT screenshot with commentary. That version is quick to produce, and it reads that way.

Nothing in it is a guarantee, either. Generative engines are probabilistic: ask the same question twice and you can get different answers. Any report claiming a fixed ranking has either measured once or is not telling you how many times it measured. Ours states the run count on every figure.

And nobody arrives at the call to find a predetermined package waiting. Sometimes the finding is that your category is not being answered by AI in a way that matters yet, and that your money belongs somewhere else this quarter. That is a cheaper outcome for us than selling work that will not pay back.

instrument
Caul
what was measured
Where each named firm ranks and whether its pages are held by Google, plus whether it appears in AI answers to a fixed question set built for the category, per engine.
how
Rankings checked per town and per service where the business is local. Every question run multiple times per engine in fresh sessions, answers stored unmodified, with the run count printed beside every figure.
over what window
A stated capture window, published on every figure in the report.
what this cannot tell you
Being named is not being recommended, and it is not traffic or revenue. Answers vary by session and by region. ChatGPT, Gemini and Google AI Overviews are covered; Perplexity and Claude are named as not covered rather than quietly left out.
What happens next

A short form, then five working days.

The form asks for your site, three to five competitors you actually lose to, and enough about the business to write questions your buyers would ask. No analytics access is required, though it sharpens the conversion pillar considerably if you share it.

The audit takes about five working days once the inputs are in. It is delivered on a call rather than by email, because the findings need explaining and the conversation is usually where the useful part happens. You keep everything either way.

Some readers arrive here already holding a report, because we build one before we ever write to a company we think we can help. If that’s you, skip the form: reply to the email and we’ll walk you through what we found.

Get your free auditIf you do foundation work

Free, no card. The form takes about two minutes, then we run the audit and walk you through it on a call.