Every AI visibility tool gives you a score.

A score is not a task.
Citeworthy finds the third-party pages AI assistants actually pull from in your category, checks whether you’re on them, and hands you the ones to fix — with the way in for each.
Free. No account. One category, real results.
Citation source report 48 runs · 24 questions
● Missing Top 10 field service management software 2026ifs.com 27 / 48editor@…
● Missing 9 best field service management software of 2026zendesk.com 26 / 48contact form
● No route Best Field Service Management Software | Capterracapterra.com 19 / 48
● Buried Best Field Service Management Software | FieldCampfieldcamp.ai 13 / 48position 7
48runs across 24 buying questions
111distinct pages cited, 77 domains
13/14top pages didn't mention the vendor
7absences with a live contact route
From one real run on field service management software. Raw samples retained.
The problem

You already know you're losing AI answers.
Nobody will tell you where.

Ask ChatGPT the same buying question twice and you get a different list of brands. Research from SparkToro puts the odds of two identical runs returning the same brand list at under 1 in 100.
So the dashboards measure the one thing that won’t hold still, and hand you a percentage. Then your client asks what to do on Monday, and the percentage has no answer.
Meanwhile around 85% of AI brand mentions come from domains you don’t own. Not your site. Someone else’s roundup, someone else’s comparison page, someone else’s directory. That’s where the decision actually gets made, and it’s the part nobody is showing you.
What we do instead

We don't measure the answer.
We find where the answer came from.

Individual answers are noise. The pages behind them are not.
In our first live run — 48 samples across 24 buying questions in one category — a single page was cited in 27 of 48 runs. The top twelve pages accounted for the bulk of every answer, while the brands named in those answers churned constantly.
That set of pages is stable enough to work on. So we work on it.
Answers are noise. Sources are a worklist.
How it works

From “we're invisible” to “here's who to email”

1
Sample

We ask the engines the questions your buyers ask — discovery, pricing, comparison, alternatives, evaluation — many times over, and record which pages they retrieve from.

2
Rank the sources

Pages rank by how many independent runs cited them, using the lower bound of the citation rate. One lucky hit in three runs never outranks a page cited twenty times in forty.

3
Get the worklist

Every page checked for your brand. Where you're missing, who's there instead, where you sit if you are there, and the contact route to change it.

What you get

Built to survive the question
“where did that number come from”

No score

Every figure is a count against a stated denominator. Nothing is a percentage with an invisible bottom half.

Noise-corrected ranking

Pages sort by the lower bound of their citation rate, so small-sample flukes can't top the list in front of a client.

Absence measured on the page

Whether we found someone to email is a separate column. The gap is never quietly understated.

Nothing dropped silently

Pages we couldn't read are counted and shown.

Every run retained

Any number in any report replays back to the exact prompt that produced it.

Client-ready output

Self-contained HTML, light and dark, your logo on it.

Who it's for

Built for the people who have to explain it on Monday

Agencies

One workspace per client, white-label output, a worklist you can assign to a junior instead of a chart you have to interpret.

In-house marketers

The answer to “why aren't we showing up in ChatGPT” that survives the follow-up question.

Founders

Find the ten pages that decide your category and go get on them.

[PROOF] Customer logo bar
Leave this row out entirely until you have real logos and written permission. An empty or faked trust bar costs more credibility than no trust bar.
Questions

The ones you're actually asking

Honestly? For most businesses, not as much as the noise suggests. AI referrals are roughly 0.32% of website traffic — Google organic is still about 43%. Anyone telling you AI search has replaced SEO is selling something.

What’s changed is that AI answers increasingly shape the shortlist before the click. If you’re in a considered-purchase category where buyers research first, being absent from the pages that feed those answers costs you before you ever see it in analytics. If you sell impulse purchases to people who don’t research, spend your money elsewhere. We’d rather tell you that now than sell you a subscription you don’t need.

They measure how often you appear in answers. We find the pages those answers come from, and tell you which ones don’t mention you. One produces a chart. The other produces a to-do list. Plenty of people buy both.

They do, and that’s the point. We don’t measure the answers — we measure which pages the engines retrieve from, which is far more stable. And we always show the count and the denominator, so you can judge the strength yourself instead of trusting our arithmetic.

No, and be careful with anyone who does. Ahrefs ran a test where pages given standard “AI optimisation” treatment were cited less than untouched controls. What we can tell you is which pages the engines are actually reading and whether your name is on them. Getting onto those pages is ordinary editorial and PR work — real, doable, and not magic.

[FILL — state exactly what you have wired and nothing more.]

Being straight with you: ChatGPT drives around 75% of AI referral traffic, so the nine-engine claims you see elsewhere matter less than they sound. We’d rather sample fewer engines properly than list nine and sample them all badly.

Partly. AI answers for local queries vary by where the question is asked from, and we don’t yet vary location properly. Until we do, treat local results as indicative. We’d rather say that than take your money for a number we don’t trust.

Find out who AI cites in your category

One input. Real pages. Takes about a minute.