How we measureAI citations.
When a buyer asks an AI assistant which tool to use, the answer names a few companies and links a few sources. This page explains how we count that, so you can check our numbers instead of trusting them.
01 / The short version
Same questions,same settings, twice.
A citer shows their sources. Every number in a GEO Sprint comes with the answers behind it.
- Engines
- ChatGPT, Perplexity, Claude and Google AI Overviews
- Prompts
- 50, agreed with you in writing and frozen on day 7
- Runs
- 3 per prompt on each engine, in fresh sessions
- Metric
- Citation share: the share of answers that name you or cite your domain
- Timing
- Baseline in days 1 to 7, the same measurement again at day 60
- Target
- +20 percentage points, or a full refund under the written terms
02 / The engines
Where we ask,and how we set it up.
Which engines
Four engines named in the terms
ChatGPT, Perplexity, Claude and Google AI Overviews. Gemini sits behind much of Google's AI search; we say in the measurement sheet whether a panel includes it.
Session setup
Clean sessions
Fresh conversations or threads for every run, with web search on where the engine offers it, and no memory or saved instructions. Google is checked in a clean browser profile.
Fixed settings
Region, language, device
We fix the country, language and device type before the baseline and write them on the measurement sheet. They do not change until the day-60 readout. We also record the model or mode each engine shows on screen for every run.
03 / The prompt panel
50 questions your buyersactually ask.
We draft the panel, you approve it, and it is frozen on day 7. After that neither side can change it, which is what makes the day-60 comparison fair.
Recommendation
Best-of questions
"Best [category] for [type of company]." The question a buyer asks when they want a shortlist.
Problem
Job-to-be-done questions
"How do I [job your product does]." A good answer would name tools.
Constraint
Fit questions
"[Category] that works with [tool]." Buyers narrow the list by what they already use.
Comparison
Alternatives questions
"[Competitor] alternatives." Useful, but kept to a minority of the panel.
Prompts are phrased the way a buyer would type them, and only a small share name your brand. A panel made only of questions that contain your name would measure nothing.
04 / Runs per prompt
Never one answer.Three runs, saved.
Why repeat
Ask an AI the same question twice and you can get different companies and different links. So we never judge a prompt on a single answer.
How a run is collected
In the GEO Sprint each prompt runs 3 times on each engine, in fresh sessions, inside the measurement window. A person collects runs from a written checklist. Every answer is saved as a screenshot or share link, so any row in the sheet can be checked. The free AI Visibility Check is lighter; its page says exactly how.
05 / What counts
Named is not the sameas cited.
We record both for every answer, so you can see whether AI talks about you, links to you, or both.
Named
The brand appears in the answer
The answer mentions your brand, including spelling variants we agree with you in advance.
Cited
A page on your domain is a source
The answer links to, or lists as a source, a page on your domain.
A run counts if the brand is named or the domain is cited. Ads and sponsored placements do not count. If Google shows no AI Overview for a run, that run counts as not cited, and we report how often an Overview appeared at all.
06 / Citation share
One formula,in the open.
Citation share = answers that name or cite you ÷ all answers × 100
With 50 prompts, 4 engines and 3 runs each, a full measurement is 600 answers. We break the result down by engine and by prompt type, and list the competitors and third-party sources the engines name most often. That list shapes the work in the middle of the Sprint.
07 / Baseline and day 60
Measure, change,measure again.
Days 1 to 7
Baseline
The full panel is run and sent to you as a dated export with every answer saved. This is the number the day-60 result is compared with.
Days 8 to 45
The work
Page rewrites, structured data, fixes to how engines read your site, and a plan for third-party mentions. You publish what we deliver.
Day 60
Re-measure
Same prompts, engines, settings and counting rules. You get both measurements side by side.
08 / What +20 points means
Percentage points,on the frozen panel.
Points, not percent
The target is 20 percentage points of citation share on the frozen panel, not a 20 percent increase. It is measured against your own baseline, not against a competitor or an industry average.
Illustration, not a client result
A worked example
From a baseline of 12%, the target is 32% or more at day 60. A move from 12% to 14.4% is a 20 percent increase but only 2.4 points, so it would not meet the target.
If the lift is under 20 points and the conditions in the terms were met, the Sprint fee is refunded in full.
09 / Eligibility and your part
What has to be truefor the guarantee to apply.
This is a summary. The written terms are the ones that bind us both.
Read the full GEO termsEligibility
Room to move
- Your baseline citation share is 40% or lower. Above that, there is not enough room for a fair test, and we tell you before work continues.
- The panel, engines and settings were frozen on day 7 and not changed.
Your obligations
What we need from you
- Publish at least 80% of the agreed deliverables by day 45, so engines have time to see them.
- Keep your site available and indexable by search and AI crawlers during the Sprint.
10 / Limits
What this methodcannot tell you.
Answers vary
Answers differ by run, by model version and by location. Three runs reduce the noise; they do not remove it. A screenshot from your own phone may not match our sheet.
Tools can differ
Results fetched through an API can differ from what the consumer app shows, so numbers from a tool and numbers from the app are not interchangeable. Where a tracking tool is used, its export is kept alongside the saved answers.
A sample, not everything
50 prompts are a sample of how buyers ask, not every question. Engines also change during 60 days, and a model update can move results for reasons neither of us controls. We note it when we see it.
Citation is not revenue
We measure whether AI names you. We do not promise traffic, leads or sales.
Clean sessions
Signed-in users with long histories may see different answers from the ones we record.
11 / On ourselves first
We run this methodon citerlabs.com first.
We have not published a baseline or a study yet. When the measurement is done we will publish the results here, whatever they show.
Until then, the method above is the whole commitment: the panel, the settings and the counting rules are written down before anyone looks at a result. Want to see the method applied to your category? The free AI Visibility Check is a lighter version of it. For the terms of the paid program, see the GEO Sprint page, and for plain-language definitions of the words used here, the AI search glossary.
12 / FAQ
Questions about the method.
Free, no call needed
See who AI namesin your category today.
Tell us your domain and category. We run 10 buyer questions and send a short report with the raw answers attached.
Get a free AI Visibility CheckOr write to [email protected]