Rixe measures ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Microsoft Copilot, Claude, Grok, DeepSeek, Mistral and Qwen. Your plan measures three engines at once; Leader measures all eleven. The plan sets how many are measured, never which ones exist.
Coverage is deliberately broad because the engines do not behave alike. A chat assistant answers conversationally and names a handful of brands in prose. Google AI Overviews sits above a results page and leans harder on sources it can link. Perplexity foregrounds its citations. DeepSeek and Qwen read a largely different body of sources, and Copilot inherits Bing's index rather than Google's. The same prompt can produce a different shortlist in each, and a brand that is strong in one is frequently invisible in another.
That is also why the limit counts prompts, not the checks behind them. Your plan measures three engines and Leader measures all eleven; a prompt counts once against your allowance however many engines read it, so there is no incentive to narrow your coverage to save money.
The practical consequence is that market and engine are the same decision. If your buyers live inside Microsoft 365 all day, Copilot is where the recommendation actually lands; if they lean on Google, AI Overviews and AI Mode are.
Why the coverage is deliberately broad
The engines do not behave alike, and a brand that is strong in one can be absent from another on the same question in the same week. A chat assistant answers conversationally and names a handful of brands in prose; a results-page surface renders a summary above links and often cites more sources; a research-style engine lists citations as it goes.
Measuring one of them and calling it your AI visibility is the most common mistake in this field. It is the equivalent of measuring one country and reporting a global figure — comfortable if the one you picked happens to like you, and useless the moment a buyer opens a different app.
What actually differs between them
Three differences change what can be measured at all. Whether the engine retrieves from the live web or answers from training data alone; whether it discloses the sources it used; and whether it shows the searches it ran before answering. An engine that discloses nothing can still be measured for mentions, but its citations cannot be read.
Ad slots are a fourth difference and a narrower one. Only a surface that still renders a results page around the generated answer carries a countable ad, which is why ad load can be reported for AI Overviews and not for an assistant answering through a provider endpoint.
How many you should actually measure
Every engine you add multiplies every prompt you track, so this is a budget decision before it is a strategy one. Most teams start with the two their buyers actually use, watch for a month, and add a third when a column starts to matter — which the per-engine grid shows before anyone has to argue about it.
One rule is worth keeping whatever you choose: nothing is estimated for an engine you did not run. A blended index that quietly fills gaps with a model of the missing engine is easier to sell and impossible to audit; a day where an engine refused is marked incomplete instead of counted as a zero.
