How to measure AI visibility (GEO) in ChatGPT, Gemini and Perplexity? Method, KPIs, tools and limits

How to measure AI visibility (GEO) in ChatGPT, Gemini and Perplexity? Method, KPIs, tools and limits
‍

Measuring your visibility in ChatGPT, Gemini or Perplexity has become a recurring question for marketing leadership. SEO reflexes transfer poorly: a generative engine does not return a list of results, it writes an answer that can change depending on the question, the session or the moment. This article presents a measurement method, the limits of each platform, how to link this visibility to conversions and the tools that let you track it. We take a position here: paid traffic from ChatGPT ads and visibility in answers are measured separately.
‍

1.What is GEO and why does it change visibility measurement?
‍

GEO (Generative Engine Optimization) covers the actions that increase the probability that a brand is cited in the answers of generative AI engines: ChatGPT, Gemini, Perplexity, Copilot, but also Google’s AI Overviews and AI Mode. The term was formalized in a research paper published in 2023 by researchers from Princeton and the Allen Institute for AI. To compose an answer, these engines rely on your site, but also on directories, press, reviews or comparisons published by third parties.
‍

In SEO, you track a position on a query. In GEO, the same question asked twice can produce two different answers, with different sources. Visibility is therefore measured as a frequency of presence across a set of questions. This is also why GEO is first a strategy topic, as we explained in GEO: the right answer to the wrong question.
‍

GEO vs SEO vs AEO: what differences for measurement?

The three disciplines share a common base (structured content, authority, sound technical setup), but they are not measured the same way:

• SEO: positions, impressions, clicks and click-through rate on a query.
• AEO (Answer Engine Optimization): presence or absence in the featured snippet and in position zero.
• GEO: brand mention, presence of a link, position in the answer and tone, measured across a panel of prompts.

In GEO, the engine picks up a piece of information and chooses, or not, to cite its source. The authority of the source, the clarity of the information and the consistency with which your brand is described from one site to another then matter more. A page that ranks well on Google can therefore be absent from generative answers. Our article on GEO myths versus SEO reality goes back over the classic signals that remain decisive.
‍

2.How to measure AI visibility (GEO) step by step?
‍

A reliable measurement starts with a simple, reproducible protocol. Each check is done in private browsing, without a logged-in account or memory enabled, with neutral wording. A team member who queries ChatGPT from their usual account gets an answer influenced by their past exchanges, which does not reflect a prospect’s experience.
‍

Tracking AI citations relies on a common grid: each answer is scored with it, which makes results comparable from one period to the next. The following three steps detail the panel, the scoring and the segmentation. They extend the GEO audit approach presented elsewhere.
‍

Build a prompt taxonomy by intent

In GEO, the prompt panel plays the role the keyword list plays in SEO. It takes up the questions your buyers actually ask, split into six categories:

• Discovery: the user is looking for a solution without knowing the players.
• Comparison: they compare several options.
• Validation: they check an option already identified (references, reviews).
• Brand: they directly query your name or your offers.
• Local: their query includes a geographic area.
• Transactional: they want to get in touch or request a quote.

Brand prompts logically give good results, since the engine cites the brand it is given. We recommend keeping them in the minority, because new prospect acquisition plays out on discovery prompts. Search listening helps you phrase prompts close to your buyers’ language. We recommend starting with a panel of 30 to 50 prompts.
‍

Set up a scoring protocol and measure volatility

The GEO scoring grid ranks each result from strongest to weakest:

1. Citation with a link to your domain.
2. Mention without a link.
3. Secondary source: your content feeds the answer without the brand being named.
4. Absence.

Position and tone are scored separately, to see clearly what is progressing. Tone is ranked on three levels: positive (explicit recommendation), neutral (simple citation in a list) and unfavorable (reservation or negative comparison).

Each prompt is also run five times per platform in order to calculate a stability rate: the number of runs in which the brand appears, divided by the total. With five runs, the rate moves in steps of 20 points, which is enough to distinguish a regular presence (4 or 5 runs out of 5) from an intermittent one (1 or 2 out of 5).
‍

Segment visibility by entity and by competitor

Results are also read by type of entity: the brand, its products or offers, its category, its experts and its executives. A brand can, for example, be well cited while its offers remain absent, which directs the work toward product pages. Laura Vol (Les Sherpas) addresses these issues in her interview on GEO and scaling marketing teams.

AI share of voice is then calculated by dividing the sum of your scores by that of all tracked brands, yours and your competitors’. To remain comparable, this calculation requires keeping the same panel and the same grid at each wave. We also identify the answers where a competitor is proposed as an alternative to your brand: these prompts become priorities in the content plan.
‍

3.What are the measurement limits by AI platform?

Answers vary depending on wording, account and session, and models are updated regularly. A GEO measurement is therefore read as a trend across several waves. Each platform also has its own particularities.
‍

ChatGPT’s specific limits for tracking visibility

ChatGPT produces two types of answers: cited answers, backed by a web search with visible sources, and synthetic answers, drawn from the model’s knowledge without a source. Only the former let you verify where the information comes from. Inaccurate mentions are logged separately, so they can be corrected.
‍

Some isolated indicators should also be interpreted with caution. A citation observed only once remains an isolated case, and a referral from chatgpt.com in GA4 says nothing about mentions without a click. Proprietary tool scores, for their part, are hard to compare from one tool to another.
‍

Perplexity, Gemini and AI Overviews: what each source lets you measure
‍

Perplexity displays its sources in every answer, which makes Perplexity visibility easier to measure. You can also see which third-party pages feed the answers on your category. The citation rate with a link is the most useful KPI here.
‍

To track Gemini visibility, you need to distinguish the Gemini app from answers generated in Google Search. The Gemini app is measured with the prompt panel, like the other assistants. For answers generated in Google Search, Google states in its documentation on AI features that AI Overviews and AI Mode data are included in the Search Console performance report, under the “Web” search type. AI Overviews data in Search Console therefore cannot be isolated: it blends with that of classic web search.
‍

4.Why AI data remains hard to compare
‍

Platforms do not communicate prompt volumes, which prevents weighting a panel by real demand. Each engine also uses its own sources, and the tools on the market do not all use the same method. The most reliable approach remains comparing each platform with itself over time, with the same grid.
‍

5.How do you attribute conversions from AI visibility?
‍

Part of the influence of generative answers happens without a click: a buyer reads an answer, remembers a name, then searches for it on Google. In a last-click model, these conversions show up as branded search or direct traffic. Many CMOs facing generative AI and the new search are working on this topic.
‍

Set up an attribution model suited to AI-assisted journeys
‍

AI attribution works best by combining four sources:

• GA4: to track ChatGPT traffic and that of other assistants in GA4, create a custom “Generative AI” channel group built from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com.
• Landing pages: they show which content is actually cited.
• Declarative attribution: a “ChatGPT or AI assistant” option in the “How did you hear about us?” field captures journeys without a click.
• CRM: the declared source, attached to each opportunity, lets you track the value generated.

The quality of this model depends on the quality of data collection. Server-side tracking limits data loss, and the four questions on customer data help you check that your CRM is ready.
‍

6.Measure the impact of mentions in ChatGPT, Gemini and Perplexity on sales
‍

Commercial impact is read by bringing several signals together from one wave to the next: panel score, branded searches in Search Console, direct traffic, AI referrals, declarations in forms and CRM opportunities. None is sufficient alone, but their joint evolution gives a solid basis for decisions. We detail this link between visibility and revenue in how GEO turns visibility into demand and revenue.
‍

7.Our position: do not add up paid traffic and visibility in answers
‍

Ads in ChatGPT (ChatGPT Ads) are a very useful paid lever to generate qualified traffic and grow your sales. However, a paid campaign and organic visibility in answers are not measured in the same way.
‍

Paid traffic, tagged with UTMs, shows up precisely in GA4. The natural presence of your brand in AI answers, on the other hand, is assessed through your prompt panel. A spike in sponsored traffic does not reflect your organic awareness in the model, and vice versa. To manage your ROI effectively and understand the real value of each lever, we recommend tracking these two channels in separate reports.
‍

8.Which thresholds and tools for scaling GEO?
‍

The volume of answers to analyze equals the number of prompts multiplied by the number of platforms, runs and markets tracked. It grows quickly with each new country or product line. With 50 prompts, 3 platforms and 5 runs, one wave represents 750 answers to score. We recommend moving to a tool from 50 tracked prompts, or as soon as the panel covers several countries: beyond that, manual coding with a consistent grid no longer holds.
‍

The choice of an AI visibility tool can rely on these criteria:

• Coverage: ChatGPT, Gemini, Perplexity, and ideally AI Overviews.
• Scoring detail: distinction between link, mention and secondary source, with position and tone.
• Multiple runs: to calculate a stability rate.
• Data export: to cross-reference results with GA4 and the CRM.
• Cost per tracked prompt.
• Production support: the ability to move from measurement to content creation.

9.Which tools to measure your AI visibility (GEO)?
‍

An AI visibility tool is most often limited to measurement. The one we recommend first also covers content improvement.
‍

GetMint: measure, then act
‍

GetMint tracks a brand’s visibility in ChatGPT, Gemini, Perplexity and Google AI Overviews, among about ten engines. The tool shows where and how often the brand is cited, calculates its share of voice against competitors and identifies the sources used by the models. Its Content Gaps module shows the topics on which your competitors are cited and you are not.
‍

Its strength is that it does not stop at the finding. Its Content Studio lets you write briefs and produce content built around the prompts to work on, and an AI agent ranks opportunities by expected impact. Diagnosis, content plan and writing sit in a single interface. It is the tool we use at Spaag for our prompt panels and our citation audits.
‍

The other tools on the market

• Semrush AI Visibility Toolkit: tracking of mentions, share of voice and sentiment, with prompt research. Convenient if you already use Semrush.
• Ahrefs Brand Radar: tracking of mentions, citations and AI share of voice against competitors, within the Ahrefs environment.
• Peec AI: tracking across several models of your choice, with a Looker Studio connector for your dashboards.
• Otterly AI: tracking in AI Overviews, ChatGPT, Perplexity and Copilot, with GEO content audits.

These tools complement Google Search Console and GA4, both free, to link visibility to traffic and conversions. This is the approach we deploy in our SEO and GEO engagements.

‍

FAQ: frequently asked questions on GEO visibility measurement
‍

How can I manually check my presence in ChatGPT, Gemini and Perplexity without skewing the results?
Work in private browsing, without a logged-in account or memory enabled, with neutral wording. Run each prompt five times and note the date of each test.
‍

At what volume of countries or product lines should we move to a dedicated tool?
From 50 tracked prompts, or as soon as the panel covers several countries. Manual coding with a consistent grid becomes hard to sustain.
‍

Which tool should I choose to measure my AI visibility (GEO)?
If your teams already use Semrush or Ahrefs, their AI modules let you get started without changing environment. To link measurement to content production, GetMint brings the two together.
‍

How often should I measure my AI visibility (GEO)?
Once a month, on the same panel and with the same grid. This pace lets you follow the evolution without adding to the workload. Choose a fixed date, aligned with your monthly reporting. Add a one-off measurement after a content overhaul or the release of a new model.

‍