NORTHFORM methodology: how we measure visibility in AI
We measure visibility in AI by putting the same set of real questions from your field to ChatGPT, Gemini, Perplexity and Google's AI-generated summaries (AI Overviews) every month. We record when your brand is mentioned and when it is cited as a source, then calculate your AI presence rate, your citation rate and your coverage by question type. We follow the trend, not a single answer. All our GEO (Generative Engine Optimization) work, meaning optimization for AI assistants, rests on these measurements.
- Models tested
- ChatGPT with web search, Gemini, Perplexity, Google AI Overviews and AI Mode
- Frequency
- Monthly, with the same questions
- Metrics
- AI presence rate, citation rate, coverage, competing sources
- Tool
- GEO Run, our own automation in Google Sheets
The question set
Everything starts from the questions your customers ask, not from keywords.
Every question in the set is written exactly the way a real customer would phrase it to an AI assistant. The set stays the same from one month to the next, so the results can be compared.
We build the set from concrete sources: the questions you get by phone or email, searches in Google Search Console, your reviews and your competitors' reviews, and forums and groups in your field. We then group the questions by intent:
- Recommendation. "Which dental clinic in Iași would you recommend for an implant?"
- Comparison. "What's the difference between a new combi oven and a refurbished second-hand one?"
- Price. "How much is a monthly membership at a gym with a sauna in Iași?"
- Local. "Where can I get physical therapy after knee surgery, near the city center?"
For the free GEO audit we use 10 questions; for a full GEO audit, between 30 and 50. You get the final list in the report, so you know exactly what we tested.
Which models we test, and how
We work under the same conditions every month, so that any change in the report comes from your website and sources, not from the way we asked.
The models
- ChatGPT, with web search turned on
- Gemini
- Perplexity
- Google AI Overviews and AI Mode, where available for the question and language in question
If a model does not generate an answer to a particular question, we record that separately. We do not count it as the brand being absent.
Testing conditions
- One test run per month, with exactly the same questions
- Logged-out sessions or clean sessions with no history, where possible
- The same location setting and the same language in every run
- For each question we save the full answer and the sources cited
We ask the important questions several times, to see how stable the answer is.
Mention or citation
Two different things, which we record separately.
| Type | What it means | Example |
|---|---|---|
| Mention | Your brand name appears in the text of the answer. | "A good option in Iași is Clinic X." |
| Citation | A URL from your domain appears in the list of sources or as a link in the answer. | The page clinicax.ro/implant appears as source number 2. |
| External citation | The cited source is another website that talks about you. | A local news article about Clinic X. |
A mention shows that the model knows you. A citation shows that it uses your website as a trusted source. Ideally you have both, but we report them separately because they improve through different kinds of work.
The metrics and their formulas
All of them are calculated on the same question set, separately for each model and then combined.
AI presence rate
Shows in how many of the questions in the set your brand appears, either as a mention or as a citation.
AI presence rate = questions where the brand appears ÷ total questions × 100
Example: if you appear in 12 of 40 questions, the rate is 12 ÷ 40 × 100 = 30%.
Citation rate
Shows what share of all the citations in the answers goes to you.
Citation rate = the brand's citations ÷ all citations in the answers × 100
We report it in two clearly labeled versions. The source-based version counts the URLs from your domain out of all the sources cited. The mention-based version counts how many times your brand is named out of all the brands named in the answers.
Coverage by question type
Shows which types of questions you appear for and which you are missing from. We calculate the AI presence rate separately for each of the four groups: recommendation, comparison, price, local.
This way you can see, for example, that you show up well for price questions but are completely missing from recommendations.
Sources and competitors
Who appears instead of you and which websites the models cite most often: competitors, directories, the press, forums.
A competitor's AI presence rate = questions where the competitor appears ÷ total questions × 100, calculated the same way as yours.
The figures in the examples only illustrate the calculation. You will see the real values in your own report, based on your own questions.
The limits of measurement
We state them openly, so you can read the report correctly.
- Answers vary. The same question can get different answers from one run to the next, even on the same day.
- Personalization. A user's history, account and location can change the answer. We test under conditions that are as neutral as possible, but every customer sees their own version.
- Models change. Providers update their models and search sources without notice. A sudden swing may come from there, not from your website.
- Availability differs. AI Overviews and AI Mode do not appear for every question or in every language.
That is why we follow the trend over several months rather than a single run, and repeat the important questions. We do not guarantee citations: the answers are generated by other companies. What we can do is measure honestly and work on the signals that matter.
What the report includes
A short monthly report that you can read in ten minutes and check line by line.
- Summary. AI presence rate, citation rate and coverage by question type, compared with the previous month.
- By model. The same metrics, separately for ChatGPT, Gemini, Perplexity and Google.
- By question. Every question, with the saved answer, the sources cited and whether or not you appeared.
- Competitors and sources. Who appears instead of you and which websites are cited most often.
- Recommendations. Three to five concrete actions for the following month.
We run the automated tests with GEO Run, our automation in Google Sheets, which sends the questions through the models' APIs and saves the answers. Google AI Overviews and AI Mode, as well as spot checks in the public interfaces, we do by hand. The monthly report is part of monthly GEO monitoring. If you would rather run the tests yourself, you can find the automation among our resources.
Frequently asked questions
The terms are also explained in our GEO glossary.
What is the difference between the AI presence rate and the citation rate?
The AI presence rate shows how many of the questions in the set your brand appears in. The citation rate shows what share of all the citations in the answers goes to you. You can appear in many answers and still have a low citation rate if the model also cites many competitors.
Why do you ask the same questions every month?
So the results can be compared. If we changed the questions, we would no longer know whether an increase came from the work on your website or from easier questions. We can add new questions, but we report them separately.
Can I check the results myself?
Yes. The report includes the list of questions, the models and the testing conditions. You can ask the same questions and compare the results. Keep in mind that answers vary slightly from one run to the next and depending on the account used.
Do you guarantee that my business will be cited?
No. The answers are generated by other companies' models, which decide on their own what to cite. We work on the signals that improve your chances: a clear picture of your business for AI assistants, structured data, content that is easy to cite and consistent external sources. Then we measure the effect honestly.
See the methodology applied to your business
The free GEO audit: 10 real questions from your field, tested in AI assistants, with a score and your top 3 priorities, within 48 hours.