GEO audit

Visibility audit

The baseline of your AI visibility: how often you are mentioned, cited and recommended, against named competitors. And what is wrong.

Request an audit

A reply within one working day, from the person who measures.

Measured in
  • ChatGPT
  • Perplexity
  • Google AI Overviews
  • Copilot
  • Gemini
Share of answersSample data
Question:

Which suppliers automate packaging lines in Germany?

Answer: For automated packaging lines in Germany, Hollweg Antriebstechnik and Marenholt Systeme are the names most often given. For smaller runs and special machinery, Rieckert & Sohn is also an option.

Your companynot named

Your company

7 %

Rank 4 of 5

  • Hollweg Antriebstechnik38 %
  • Marenholt Systeme29 %
  • Rieckert & Sohn19 %
  • Your company7 %
  • Tessmar Industrie4 %

Sample data with invented names. Shows the format, not a client result.

First place on Google is the ticket, not the seat.

AI engines do not simply summarise the top ten. Which sources they prefer in your category is an empirical question. And being mentioned, cited and recommended are three different things.

How GEO differs from SEO

How it runs

The proposal names the timeline once we know your category.

  1. Collect the questions

    A conversation with you and your sales team. Questions naming your brand go into a separate group.

    OutputPrompt set, frozen

  2. Set competitors

    Your list plus the names the engines volunteer. Rewritten per language, not translated.

    OutputCompetitor and language plan

  3. Measure

    Several runs across five engines, clean sessions, documented region. Every answer kept verbatim.

    OutputRaw observations

  4. Analyse

    Share per engine, split into mentions, citations, recommendations. Errors, sources, spread.

    OutputFindings with spread

  5. Prioritise

    A sorted list with effort, impact and evidence. Reviewed with you.

    OutputReport and action list

Two readings. What really happened in between.

The bar shows the sample's spread. If the movement lies within it, we measured no change, and the report says so. The third row is the case nobody else shows: up, but within the spread.

Prompt set "packaging automation supplier choice" · 40 runs per reading · 5 enginesSample data
Prompt set "packaging automation supplier choice" · 40 runs per reading · 5 engines
BrandBaselineFollow-upSpreadVerdict
Your company7 %19 %±4+12, measured change
Hollweg Antriebstechnik38 %31 %±5-7, measured change
Marenholt Systeme29 %33 %±6+4, within the spread
Rieckert & Sohn19 %18 %±3-1, within the spread

Sample data with invented names. Shows the format, not a client result.

What the report contains

Share of answers

Per engine and overall, against competitors, with sample size and spread.

Mentions, citations, recommendations

Three separate figures, because they have three different causes.

Errors

Every false statement with wording, date and frequency.

Sources

Which websites the engines cite in your category, and whether yours is among them.

Technical findings

Crawler access, readability without JavaScript, the first characters of every page.

Raw data

Date, prompt-set version, runs, engines, wording. Every claim is traceable.

Questions about the audit

Why not a single snapshot?

Because it proves nothing. Only comparing two readings shows whether a figure is stable and whether an action worked.

How many questions are in the set?

As many as your buying situations yield. The number is in the proposal. What matters more is separating questions with and without your name.

Do you analyse competitors too?

Yes, in the same set with the same figures. A technical review of other companies' websites is not included.

Which language do you measure in?

The language your buyers use. German questions surface different brands and sources than English ones. For further markets, questions are rewritten, not translated.

What can you not measure?

We have no view into training data: we see that an engine prefers a source, not why. A published question set stands in for what individual customers actually asked. There is no cleanly isolated revenue contribution, because many AI visits arrive without a referrer.

Start with the baseline.

Tell us your category and one question your customers ask.

A reply within one working day, from the person who measures.