How to Measure AI Visibility: Metrics That Show Where the Brand Wins and Loses in AI Responses
By Pedro Sonda, dooop
Measuring AI Visibility means calculating, from many AI-generated responses, how often a brand appears, how prominently, whether it is cited as a source, if it is described correctly, which sources support that presence, and what effect this has on the business. The measurement is done over a defined set of questions, repeated multiple times on each platform, because AI responses change with each execution.
A single index ("our AI visibility is 63") is easy to report but hard to use. It does not indicate whether the problem is not being remembered, being remembered only at the end of the response, or being described incorrectly. The following metrics separate these situations.
Updated in October 2026.
Why Measuring AI Is Different from Measuring SEO
In traditional search, a page’s position for a keyword is relatively stable and can be checked. In AI responses, two things change.
Responses vary. In a study with 600 volunteers and 2,961 runs on ChatGPT, Claude, and Google AI, SparkToro concluded there is less than a 1 in 100 chance that two responses will have the same list of brands, and about 1 in 1,000 chance they will have the same order. In contrast, the frequency with which each brand appears was much more stable (SparkToro, 2026).
Questions vary. In the same study, when 142 people freely wrote a question with the same intent, the average semantic similarity between texts was 0.081, meaning almost no overlap in wording. Even so, the strongest brands appeared in 55% to 77% of the responses (SparkToro, 2026).
Implication: the unit of measurement is not the exact question, but the intent, tested in various formulations. And the result is not a position, but a frequency, calculated over many runs.
AI Visibility Metrics
Each metric answers a different question. Together, they show where the brand loses ground.
| Metric | How to Calculate | Question It Answers |
|---|---|---|
| Presence Rate | Responses mentioning the brand ÷ total responses | Are we remembered? |
| Share of Voice | Mentions of the brand ÷ mentions of all brands in the category | What share of the conversation do we occupy versus competitors? |
| Prominence | Average position of the mention in the response or portion of text dedicated to the brand | Do we appear as the main option or as a footnote? |
| Citation Rate | Responses citing a brand page as a source ÷ total responses | Is our content used as a reference? |
| Fidelity | Responses with correct description of the brand ÷ responses mentioning the brand | Does the AI tell the truth about us? |
| Source Participation | Most cited domains in category responses | What sources shape the AI’s opinion in our market? |
| Effect | Sessions, leads, and conversions from AI platforms | Does visibility generate results? |
The prominence metric has academic support: the study that formalized GEO measured visibility by the amount of response text associated with a source, weighted more heavily for what appears earlier, a metric called Position-Adjusted Word Count (Aggarwal et al., KDD 2024). The logic is the same: appearing early and with more text is worth more than appearing at the end, in a single line.
Why separate presence and citation: a brand can be mentioned without its website being cited (the AI knows it from other sources), or the website can be cited as a source without the brand being recommended (the content is useful, but the company is not remembered as an option). These are different problems with different solutions.
How to Build the Measurement System
1. Define intents, not keywords
Start with a set of canonical intents: the needs and decisions that matter to the business. Example: "hire AI governance consulting" is an intent; "best AI governance consulting Brazil" is one way to express it. Organize intents by family: brand identity, category, problem, solution, comparison, and hiring.
2. Create variations for each intent
Write 5 to 10 formulations per intent: executive and colloquial language, with and without region, with and without sector. This makes measurement robust to the real diversity of questions.
3. Run multiple times, per platform
SparkToro recommends running each prompt 60 to 100 times to obtain a meaningful visibility measurement and warns that tools showing "AI position" lack statistical basis (SparkToro, 2026). Measure each platform separately: ChatGPT, Gemini, Google AI Overviews and AI Mode, Perplexity, and Claude retrieve sources differently.
4. Control the environment
Use accounts without history or personalization, record whether web search was active, and keep the same set of questions between cycles. Without this, a change in numbers may just be a change in environment.
5. Record observations, not just "yes or no"
For each response: did the brand appear, in what position, how was it described, which competitors appeared, which sources were cited. All metrics in the table come from these observations.
6. Report with margin of error
Every rate calculated from a sample has uncertainty. A presence rate of 30% measured in 100 responses has a margin of about 9 percentage points up or down (95% confidence interval). If the rate goes from 30% to 34% in 100 responses, that is not proof of improvement. Comparing cycles without considering the margin is the most common way to draw wrong conclusions about GEO.
dooop Framework: From Intent to Diagnosis
dooop organizes measurement in a fixed sequence:
Canonical Intent → Variations → AI Runs → Observations → Metrics → Diagnosis
The central point is the last step. Metrics are read by intent family, not summed into a single index. A hypothetical example of reading:
| Intent Family | Presence Rate | Interpretation |
|---|---|---|
| Identity (questions with brand name) | 94% | The AI knows who the company is |
| Authority (questions about the topic) | 76% | The company’s content is used to explain the subject |
| Category | 58% | Association with the category is average |
| Hiring | 42% | The company is little remembered when someone is hiring |
| Comparison | 31% | Competitors dominate comparisons |
Illustrative values. In this case, an average index would hide the main diagnosis: the company has authority but lacks commercial consideration. The right intervention would be choice content and third-party evidence, not more informational articles.
Continuous Brand Presence Monitoring
After the baseline, monitoring repeats the same set of intents in regular cycles. Three precautions:
- Automate execution. The recommended volume makes manual processes unfeasible. Market tools exist; before contracting, check if they publish methodology, how many runs per question they perform, and if they show margin of error.
- Track effect with data platforms provide. ChatGPT includes
utm_source=chatgpt.comin referral links (OpenAI — FAQ for publishers). On Google, AI Overviews and AI Mode traffic enters the Search Console Performance report along with web search, without separation (Google Search Central). - Don’t measure clicks only. Users who saw an AI summary on Google clicked traditional results in 8% of visits, versus 15% for those who did not (Pew Research Center, 2025). Part of the influence happens without clicks, and only presence measurement captures it.
What to Do Now
- If you don’t measure anything yet: start with 10 to 20 intents, five variations each, on two platforms. This already creates a baseline.
- If you use a tool showing "position": replace the main indicator with presence rate and ask the provider for methodology.
- If you report a single index: break down the number by intent family and platform before the next results meeting.
- If you will invest in content or public relations: measure beforehand, with sufficient volume, to prove the effect later.
To build the baseline with intents, run volume, and reading by family, the path is a dooop AI visibility diagnosis.
Frequently Asked Questions
What is the most important AI Visibility metric? It depends on the business question. For B2B sales, presence rate in hiring and comparison questions is usually the most revealing, as it shows if the company enters the set of alternatives.
How often to measure? Monthly is a common rhythm to track trends, plus extra measurements after relevant changes. The important thing is to keep the same set of questions and method between cycles.
Can AI Visibility be measured only with Search Console and Google Analytics? No. These tools measure clicks and visits, not presence in responses. They complement effect measurement but do not replace presence measurement.
Also Read
- What is Generative Engine Optimization? Origin, mechanism, and what research really shows
- How to know if your company appears on ChatGPT: a test that produces data, not impressions
- How to measure a company’s presence in AI responses: start with the intent map
- How to monitor competitors in AI responses: discover who occupies your place, where, and why
About dooop
dooop is a consultancy based in Caxias do Sul (RS) that prepares leaders and organizations to transform artificial intelligence into business capability, with strategy, governance, and training. This article is part of dooop’s series on GEO and AI Decision Infrastructure.