What is an AI citation audit?

An AI citation audit measures how often, how accurately, and in what context AI answer engines cite a brand compared with competitors. A rigorous audit uses a large prompt set, runs each prompt multiple times to account for the non-deterministic nature of generative answers, reports variance alongside averages, and identifies the specific source URLs being cited rather than only which brands appear.

Why do you run each prompt multiple times?

Because generative engines are not deterministic. The same prompt produces different retrieval and different citations across runs, and the variance is largest in competitive categories. A single run per prompt produces a number that cannot be reproduced, which makes it useless for tracking change or attributing results. Multiple runs let us report a distribution with stated confidence rather than a figure that will not survive being checked.

How is this different from an AI visibility tracking tool?

Tools are good at ongoing monitoring and we work with whatever you have licensed. What they generally do not do is run enough repetitions to establish variance, identify the specific third-party URLs driving citations in your category, or diagnose whether a gap is an access problem, a retrieval problem, or a usage problem. A tool tells you that you are absent. An audit tells you why and what would change it.

How many prompts do I need for a meaningful audit?

For a single market, 100 prompts run five times each is our minimum for reportable confidence, producing 500 responses per engine. Broad categories or multi-market work needs 250 to 500 prompts. Below roughly 50 prompts the variance swamps the signal and differences between brands stop being distinguishable.

How often should we re-run the audit?

Quarterly for most businesses. AI engines change behaviour frequently enough that annual measurement misses the movement, and monthly re-runs rarely exceed the variance threshold for meaningful change. Re-runs use the identical prompt set so the series is comparable.

How much does an AI citation audit cost?

A standard audit covering 100 prompts across four engines in one market typically costs $2,500 to $5,000. Extended audits with larger prompt sets and full source analysis run $5,000 to $12,000, and multi-market audits $8,000 to $20,000. Quarterly re-runs on an established baseline are considerably cheaper than the initial audit.

Multi-run methodology · stated confidence

AI Citation Audit Services

An AI citation audit measures how often, how accurately, and in what context AI answer engines cite your brand compared with competitors — using a prompt set large enough and run often enough to produce numbers you can act on. We audit ChatGPT, Perplexity, Google AI Overviews, and Claude, and we tell you which sources they are citing instead of you.

Here is the problem with almost every AI visibility number you have been shown. Ask ChatGPT the same question three times and you will frequently get three different sets of citations. Generative answers are not deterministic. A scan that runs each prompt once and reports “you appear in 40% of answers” is reporting a coin flip with a decimal point attached.

If a measurement cannot survive being re-run, it is not a measurement.

Request a scoped citation audit See real output

Scope, prompt set, price, and delivery date within two business days.

Want a quick look first? The free AI visibility report is a 72-hour directional scan, and it is the right starting point for most people.

Try it · how much can you trust the number?

Say your true citation rate is 22%. Set a sample size and see what a measurement could honestly report.

Prompts

Runs per prompt

Honest reportable range n = {{ sampleN }}

                0%true rate 22%100%
              

Two terms: binomial sampling error, which more prompts reduce, plus engine drift between passes, which only repeated runs reduce. Set runs to 1 and the second term cannot be estimated at any prompt count — which is the argument this page is making.

In practice

What the work looks like

Screens from live engagements, with client details removed.

The main view

Methodology

Why one run tells you nothing

Generative engines sample. The same prompt produces different retrieval, different synthesis, and different citations across runs — and the variance is largest exactly where it matters most, in competitive categories where several sources are plausible.

Approach What it produces What it is good for

Practical consequence: if your citation share moves from 22% to 26% between two audits, you need to know whether that exceeds normal run-to-run variance. Without multiple runs it does not, and you have just made a budget decision on noise.

Scope

What we measure

Dimension What we report

The source-level detail is the finding clients act on most. Discovering that 60% of your category’s citations come from four third-party pages you do not appear on is a concrete, addressable problem. Discovering that your citation share is 22% is not.

The actual product

What the output looks like

Anonymised excerpt from a real audit. This is the format every cluster arrives in.

Prompt cluster 04

“best [category] software for [use case]”

34 prompts · 5 runs each · 170 responses per engine

Brand

{{ p.label }}

Confidence note

±figures are standard deviation across 5 runs. Differences below 10 points between brands are not reliably distinguishable at this sample size.

Every cluster arrives with sample sizes, variance, source attribution, and a diagnosis of whether the gap is access, retrieval, or usage.

Fit

When to buy this

✓ {{ f.title }} {{ f.copy }}

Probably not you if you have never checked at all. Run the free AI visibility report first. If it shows you are absent everywhere because a crawler is blocked, that is a configuration fix rather than a measurement problem, and a paid audit would be premature.

Deliverables

What you receive

The actual documents, not a sample deck built for the pitch.

Citation audit

What five engines say about you today

120 prompts · 5 engines · verbatim

01Every answer captured verbatim 02Who is named instead of you 03Sources each answer cites 04Factual errors about your business 05The shortest route into the answer

B Screenshots and transcripts included

Report

Citation readiness

Entity consistency58%

Extractable structure44%

Third-party presence31%

Schema coverage66%

01 · What do you need

02 · How big

03 · Anything else

Estimate

{{ estLow }} to {{ estHigh }}

Get this scoped properly

An estimate, not a quote. Fixed price follows a scoping call, and where requirements are genuinely uncertain we scope a paid discovery first.

Quarterly re-runs are where the value compounds. A baseline tells you where you stand. A tracked series tells you whether anything you did worked, which is the question you will actually be asked.

Inputs

What we need

This audit runs largely from the outside, which is also why it works on competitors with the same rigour it works on you.

✓ {{ a }}

Evidence

Before and after

The same view, before the work and after it.

Before

The conversation was already happening. You simply were not in it.

After

Modelled from the sources these engines already read and trust.

Our stack

Billion Game runs your work through the same instruments the best in-house teams use.

Licences are on us, and every export we pull from them is yours to keep.

✎

Be wary of any agency whose reporting is a tool dashboard with their logo on it. Tools measure; they do not decide what matters, and every one of these will happily generate a hundred findings that change nothing. What you are paying for is the judgement about which three of those hundred are worth your developers’ time.

FAQ

Measurement questions

Including the one about our own free report, answered honestly.

The improvement work itself is answer engine optimisation, and you are under no obligation to buy it from us. The audit is written to be handed to anyone.

A number you can defend, not a dashboard figure.