The category is drowning in made-up metrics. Here is exactly how every number on a Promptleads report is produced, so you can stand behind it in front of a client.
We query each engine through its real interface. Google AI Overview and AI Mode come from the live SERP, exactly what a consumer sees. ChatGPT, Perplexity, Gemini, and Grok come from their official APIs with web search forced on. Never a scraper, never a non-native answer.
The loudest doubt in this category is that API answers differ from real ones. Here is our honest take. For the Google surfaces (AI Overview, AI Mode) we read the live SERP itself, what a consumer actually sees. For the chat assistants we use each vendor's official API with web search forced on, so the answer is grounded in live retrieval, not a stale, ungrounded guess, but it is not byte-identical to one person's logged-in session with its own memory. That is exactly why we run many samples, report a confidence interval, and corroborate with branded-search lift, instead of selling a single screenshot as the truth.
Every prompt is coordinate-pinned across a geo-grid of the metro, so “near me” resolves the way it does for a real local buyer. Block by block, not a national average.
AI answers are non-deterministic. We run each prompt repeatedly and report a distribution with a 95% Wilson confidence interval, never a single snapshot sold as ongoing truth.
We confirm a gain or loss only when the before and after confidence intervals separate. Everything else is holding: it moved, but within measurement noise. That is why we do not alarm you on a single-run blip the way most tools do.
Showing up and being cited are different things. We separate a bare mention from a mention where the AI actually cites your own site, so you know whether your presence is supported or fragile.
Every figure traces to a logged AI answer: the engine, the date, the cited URLs. We never fabricate “AI search volume” or a black-box visibility score. When we do not have the data, we say so.
We show the exact third-party sites the AIs cite, the citation graph, so a fix is a specific, checkable action (get listed on X, earn a mention on Y), not a vibe.
When we report how many visitors AI sent, we read it from the client's own Google Analytics and Search Console. ChatGPT and Gemini often strip the referrer, so those visits undercount. We show them as a floor and corroborate with branded-search lift, never an inflated estimate.