# Why do AI visibility tools give different results for the same brand? > Two tools can give one brand different AI visibility rates: the questions, engine version, web search, counting rule and failed answers all move the number. Published: 2026-11-07 AI visibility tools give different results because each one asks different questions, reaches the engines through different versions and settings, and decides by its own rule when a brand counts as "mentioned". GeoMono is a platform that measures how visible brands are in AI answers; this guide explains why two tools can report different rates for the same brand, and what to ask before you trust either number. The short version: two different numbers do not mean one tool is broken. Most of the time the two tools are not measuring the same thing. Before you put the numbers side by side, find out what each one counted. ## Five choices that move the number A visibility rate is the product of five decisions. Change any one of them and the rate changes too. ### 1. Which questions were asked? The rate is calculated over the questions asked. One tool may use broad category questions, another purchase-intent questions, a third the questions you wrote yourself. If some of the questions contain the brand's name, the rate is inflated: an engine asked "Is X reliable?" will name X whatever X's visibility is. Language matters as well. An answer to a question asked in Turkish may not name the same brands as an answer to the same question in English. Tools do not always say in their published documentation which language they use for a given market. ### 2. Which engines, and which versions? "Visibility in ChatGPT" is not one thing. The consumer app, the provider's API and different model versions can produce different answers. When the model behind an engine changes, the answers to the same questions change with it. Two tools measuring on different dates with different versions may in effect be measuring two different engines. ### 3. Did the engine search the web? According to [OpenAI's help page](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt), ChatGPT can search the web automatically when it judges that a question would benefit from current information. A call with web search on and a call with it off draw on different sources and can name different brands. Whether a tool turns search on changes its rate directly. ### 4. How was a mention counted? A name appearing in an answer does not prove the brand was mentioned. It may be another company with the same name, a person, or a sentence repeating the question back. A tool that does not filter namesakes inflates the rate; one that does not recognise the different spellings of your name deflates it. Some tools count only mentions, others also count being cited as a source. Those are separate events and should be counted separately. ### 5. What happened to answers that failed? An engine can time out, return an error or cut an answer short. A tool that counts that answer as "brand not mentioned" lowers your rate; a tool that never records it hides how many questions were actually measured. Either way the number looks more precise than it is. ## The same engine does not give the same answer twice Part of the gap between tools comes from the engines themselves. [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 found that AI tools return very different lists of brands when the same question is asked repeatedly. The April 2026 paper ["Don't Measure Once"](https://arxiv.org/abs/2604.07585) makes the warning in its title: visibility should not be read from a single measurement. So two tools asking the same engine the same question on the same day can still get different answers. A rate from a single measurement can move by several points when it is repeated, and a small difference between two tools may sit entirely inside that movement. ## How tools aggregate results Tools smooth this variation in different ways, which is another reason their numbers differ. A few examples from their published methods: - **Ahrefs Brand Radar:** its [methodology page](https://ahrefs.com/blog/brand-radar-methodology/) says ChatGPT, Perplexity, Gemini and Copilot data are refreshed monthly and reported over a 90-day window. - **Profound:** in [a published post](https://www.tryprofound.com/blog/the-half-life-of-an-ai-citation-is-11-days) it describes smoothing a page's daily citation share with a 14-day average to dampen day-to-day noise. - **Semrush AI Visibility Toolkit:** its [knowledge base page](https://www.semrush.com/kb/1607-semrush-ai-visibility-data) describes where the data comes from and its regional databases. Other tools, such as [Peec AI](https://peec.ai/), also track brand visibility in AI search. Each works with its own question pool, its own time window and its own counting rule. The average of a 90-day window and the result of a single week will not match, and both can be right on their own terms. ## Four questions before you read a rate If you are looking at two different numbers, ask these before asking which one is "right": 1. **Which questions?** Can you see the list? Do any contain the brand's name? Are they in the language your buyers use? 2. **Which engines and versions?** Is each engine reported separately, or folded into one number? On what date, and with which model version? 3. **How was a mention counted?** Are namesakes filtered out? Are mentions and citations kept apart? 4. **What happened to failed answers?** How many questions were planned and how many were actually measured? Were failures counted as "not mentioned"? A number that cannot answer these four questions, high or low, is not fit for comparison. {{cta}} ## How GeoMono holds these choices fixed In our own measurement we fix these five choices in writing; every rule is published on the [How we measure](/en/yontem) page. - **Questions are registered first.** Before a measurement, a written and versioned declaration is prepared for the brand, stating how many questions there are and where they come from. Questions are written without the brand's name, the way a buyer would ask. - **Five engines, asked separately.** Every question goes to ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews in the same way. There are no weights between engines, and the results are not folded into one number. - **The engine version is recorded.** We query each engine through its provider's own interface, with the engine's own web search and no added instructions; AI Overviews are read from Google's results page. When an engine's version changes, a new measurement series starts and is never compared with the old one as a single series. - **Every mention is verified.** Answers are recorded one by one; namesake and echo detection means an answer about another company with your name does not count as you. - **Failures are written separately.** An engine that did not answer is not counted as "did not mention you": the answer is left out of the count, recorded and retried. The result for each engine is written as "named in x of y questions". None of this makes GeoMono's number match another tool's. It only makes clear what was measured. ## Which tool's number should you trust? There is no single correct number; there are numbers that say clearly what they measured. When you choose a tool, choose the one that answers the four questions most plainly, not the one that reports the highest rate. Ask the same questions of the same engines at regular intervals with that tool, and compare your number only with its own history. Read a difference between two measurements as a change only when it is larger than asking the same questions again could produce. We explain how to set up the measurement step by step in [How to measure brand visibility in AI](/en/blog/yapay-zeka-marka-gorunurlugu-nasil-olculur), and why the same question brings different brands from engine to engine in [same Turkish question, different brands](/en/blog/ayni-turkce-soruya-farkli-markalar). To see side by side what each tool measures, from its own published pages, the [AI visibility tools comparison](/en/blog/yapay-zeka-gorunurluk-araclari-karsilastirmasi) puts engines, Turkish-language answers, how questions are asked, source tracking and reporting in one table. ## Frequently asked questions ### Two tools give me very different rates. Which one is right? Both may be right by their own rules. First compare which questions each tool asks, which engines and versions it measures, how it counts a mention and what it does with failed answers. If those four differ, the two numbers are not measuring the same thing. ### The same tool shows a high rate one week and a low one the next. Why? Engines do not recommend the same brands every time. A measurement with few questions can move by several points when it is repeated. To read a difference as a change, it has to be larger than repeating the same questions could produce. ### Why is folding all engines into one number misleading? Each engine can have a different number of measurable questions, and the engines read different sources. One number hides a strong result in one engine behind a weak result in another. Reading each engine against its own number of measured questions tells you more. ### Why do failed answers matter? A tool that counts a failed answer as "not mentioned" reports a provider error as a finding about your brand. A report that states how many questions were planned and how many were actually measured makes that error visible. ## Sources - [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 - [Ahrefs Brand Radar Methodology: How we collect and model AI visibility data](https://ahrefs.com/blog/brand-radar-methodology/), Ahrefs, updated 26 February 2026 - [Where does the data in Semrush's AI Visibility Toolkit come from?](https://www.semrush.com/kb/1607-semrush-ai-visibility-data), Semrush Knowledge Base, accessed 7 October 2026 - [The half-life of an AI citation is 11 days](https://www.tryprofound.com/blog/the-half-life-of-an-ai-citation-is-11-days), Profound, 30 September 2026 - [Peec AI: AI Search Platform for Marketing Teams](https://peec.ai/), Peec AI, accessed 7 November 2026 - [Searching the web with ChatGPT](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt), OpenAI Help Center, accessed 7 October 2026 - 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)