# GEO for performance and budget owners: what it measures and how to judge the results > GEO measures mentions inside AI answers, not clicks. For performance and budget owners: the metrics, what is counted, the rules for comparison and the limits. Published: 2026-11-07 For performance and budget owners, what GEO measures is whether a brand is named in the answer an AI engine gives to a buyer's question, how it is named, and which sources the answer cites; the result reads, for each engine, as "named in x of y measured questions". GeoMono is a platform that measures how visible brands are in AI answers; this guide sets out the metrics, the rules for comparing periods and the limits of the measurement for the people who have to judge GEO spending. The short answer: you cannot judge GEO by clicks, conversions or revenue, because the event it measures often produces no click. What you can judge is what your mention rate does over time, on the same questions and the same engines. ## What GEO measures Your brand can appear in an AI answer in three distinct ways, and each should be measured separately: - **Mention:** is your brand's name in the text of the answer? - **Citation:** does the engine show your site as a source for the answer? - **How you are named:** as the leading option, as one of the alternatives, or with a caveat? The three are independent. Your site can be cited while your name never appears in the answer; your name can appear in an answer that cites nothing. We explain the difference in detail in [brand mentions versus citations](/en/blog/marka-anilmasi-ve-kaynak-gosterilmesi-farki). ## The main metric: a rate, and what it counts GEO's main metric is a rate: of the questions measured, in how many were you named? When you judge that rate, look at what it was calculated over. **Out of how many?** The rate is calculated over the questions measured. "40%" on its own says nothing; "20 of 50 questions" does. With few questions, a few answers changing moves the rate a lot. **Where are the failures?** If an engine timed out or returned an error, that answer must not be counted as "not mentioned". If it is, the rate falls unfairly. A good report states separately how many questions were planned and how many were actually measured. **Is each engine separate?** ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews read different sources and can recommend different brands. Folding the five into one number hides which engine changed. For a budget decision, engine-by-engine results tell you more. **Whose questions are they?** The rate only means something relative to the questions asked. They should not contain the brand's name, and they should be the questions your real buyers ask. Measuring consumer questions for a company that sells to enterprises can produce a high but meaningless rate. ## How to compare two periods For budget owners the key question is "did anything change after we spent?". Answering it honestly takes three rules. ### 1. Same questions, same rules Two measurements can be compared only if they used the same question set and the same counting rules. If the question set changed, the difference between the two rates may come from the questions rather than the work. If you want to add questions, keep asking the old ones. ### 2. The engine's own variation AI engines do not give the same answer to the same question every time. [Research published by SparkToro](https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/) in January 2026 shows this for brand recommendations, and the April 2026 paper ["Don't Measure Once"](https://arxiv.org/abs/2604.07585) argues that visibility should not be read from a single measurement. So a small difference between two periods can appear even if nothing was done. To read a difference as a change, it has to be larger than asking the same questions again could produce. ### 3. The model version The models behind the engines change from time to time, and when they do, the answers to the same questions can change. If an engine's version changed between two periods, part of the difference may come from the engine rather than your work. The report should say which version was measured. {{cta}} ## What not to judge GEO by **Clicks.** An AI answer often produces no click. In a measurement in the United States, fewer than a third of Google searches in early 2026 ended in a click ([SparkToro, Similarweb data](https://sparktoro.com/blog/in-2026-less-than-one-third-of-google-searches-still-send-a-click/)). That study does not cover Türkiye, but it shows the direction: a reader can decide without ever reaching your site. A mention your analytics cannot see is not a mention that did not happen. **Search rankings.** A page ranking high on Google does not mean it will be cited in the AI answer to the same question. A [2025 Ahrefs analysis](https://ahrefs.com/blog/ai-search-overlap/) found that only a small share of the links AI assistants cite rank in Google's top 10 for the same question. **Revenue, directly.** There is no reliable way today to tie a mention in an AI answer to a particular sale. A report that claims to do so is stating something it did not measure. If you want to see the effect on revenue, read the GEO measurement beside your own sales and brand data, and do not present either as proof of the other. **A single score.** A score that folds several engines and metrics into one number hides which part changed. When the score rises, you do not know what rose. ## What to look for in a GEO report When you assess a GEO report, look for the answers to these questions: 1. Which questions were asked, and are they the same as last period? 2. For each engine, how many questions were planned, how many were measured, and in how many were you named? 3. Were failed answers written down separately? 4. If a change is claimed, is it larger than the measurement's own variation? 5. In which questions, and from which sources, were your competitors named? 6. Which model versions were measured? If the report does not answer these, ask before you base a budget decision on its numbers. ## How GeoMono reports it GeoMono's registered measurement report gives no single composite score and does not rank the engines. Questions are registered before the measurement in a written, versioned declaration for the brand, and are written without the brand's name. Every question is asked of each of the five engines in the same way. For each engine we write "named in x of y questions", with the state of that result beside it: measured, measured but not named, could not be measured, or not yet measured. A failed answer is left out of the count and retried. Two measurements are compared only with the same question set and the same rules. A difference is written as "up" or "down" only when it is larger than asking the same questions again could produce; otherwise the report gives both numbers and the range the difference could lie in, with no direction word. When an engine's model version changes, a new measurement series starts and is never compared with the old one as a single series. The full rules are on the [How we measure](/en/yontem) page. ## Frequently asked questions ### How do we calculate the return on GEO? There is no reliable way today to tie a mention in an AI answer to a particular sale. What you can judge is what your mention rate does over time on the same questions and engines. Read it beside your own sales and brand data. ### Our mention rate rose 5 points in a month. Is that a success? First check that both measurements used the same questions, and that the difference is larger than the measurement's own variation. With few questions, a 5-point difference can appear even if nothing was done. ### Our analytics show very little traffic from AI. Do we need GEO? An AI answer often produces no click. Low traffic does not show that you are absent from the answers; to see whether you are named, you have to measure the answers themselves. ### How many questions are enough? There is no single right number. The fewer the questions, the more each answer moves the rate, and the harder it is to say whether a difference between two periods is real. Before the count, check that the questions reflect your real buyers. ## Sources - [In 2026, Less than One Third of Google Searches Still Send a Click](https://sparktoro.com/blog/in-2026-less-than-one-third-of-google-searches-still-send-a-click/), SparkToro (Similarweb data), 9 June 2026 - [AIs are highly inconsistent when recommending brands or products](https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/), SparkToro, 28 January 2026 - [Don't Measure Once: Measuring Visibility in AI Search (GEO)](https://arxiv.org/abs/2604.07585), arXiv (Schulte, Bleeker, Kaufmann), 8 April 2026 - [Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt](https://ahrefs.com/blog/ai-search-overlap/), Ahrefs, 11 August 2025 - How we measure: the GeoMono method, 2026-11-07: published method; weights, thresholds and figures are read from the code that computes the score (https://geomono.com/en/yontem)