How we measure AI visibility
Our whole platform rests on one promise: when people ask AI for a business like yours, we help you be the one it names. So you deserve to know exactly how we measure that — the prompts we ask, the engines we query, and what we do and don't claim. Here it is, in full.
Four honest steps, every time
No proprietary magic — just a repeatable process you could describe to your own team.
1 · We ask real buyer questions
We don't ask an AI “is GrowMyWebsite good?” — that's a leading question. We ask the questions your customers actually ask when they're choosing, like “what are the best options for [your category] in [your city]?” These category, buyer-intent prompts are where discovery really happens.
2 · We ask several engines, the same way
The identical question goes to every AI engine you've connected — Claude, ChatGPT, Perplexity, Gemini, and Google's AI Overviews. We run them at temperature 0 (the most repeatable setting) and instruct each model, in its system prompt, to answer only from what it actually knows and to never invent a brand.
3 · We record what actually happened
For each answer we capture: were you named or recommended, which competitors were named instead, the sentiment toward you, and one concrete change that would improve the odds next time. “Share of voice” is simply how often you appear versus the rivals the engines name.
4 · We track the trend, not a single roll of the dice
AI answers vary between runs and personalize per user, so one response proves little. We re-check on a schedule and across multiple prompts, so what you see is a trend line over time — the honest signal — not a lucky or unlucky one-off.
Which AIs we actually ask
Each engine lights up when you connect it. One you haven't connected is reported as “not connected” — never simulated and counted as real.
Claude
Queried directly via Anthropic's API.
Always on.ChatGPT
Queried via OpenAI's API (GPT-4o mini).
When connected.Perplexity
Queried via Perplexity's API (Sonar).
When connected.Gemini
Queried via Google's Generative Language API.
When connected.Google AI Overviews
The real Google answer panel, read from live search results — not an LLM simulation.
When connected.What we're careful not to overclaim
- We can't see inside any AI company's private ranking system — nobody outside them can, and we won't pretend to.
- Claude, ChatGPT, and Gemini answer from the model's own training knowledge without live web browsing — so those results reflect training-knowledge recall and may differ from a live, browsing assistant. Perplexity (Sonar) and Google AI Overviews do read live web and search results, so we report them as the live, browsing engines they are.
- We don't guarantee you'll be recommended. We measure where you stand and do the work that improves your odds.
- AI answers vary and personalize — so we track trends across repeated checks and many prompts, never a single response.
- An engine you haven't connected is shown as “not connected.” We never fabricate a score or simulate an engine and pass it off as real.
This is the same standard we hold the rest of the platform to: everything delivers what it advertises, and we'd rather show you an honest number than a flattering one.
AI visibility, measured — your questions
Do you query the real ChatGPT, Perplexity, and Gemini — or simulate them?+
Real ones, through each provider's official API. ChatGPT (OpenAI), Perplexity (Sonar), and Gemini (Google) are queried directly when you connect them; Google AI Overviews is read from the actual Google answer panel, not an LLM guess. Claude is always available. An engine you haven't connected is shown as “not connected” — it is never quietly simulated and counted as if it were real.
Branded vs. category queries — what's the difference, and which do you track?+
If you ask an AI about your business by name, it will almost always say something nice — that's a branded query, and it tells you little. The queries that decide whether you get discovered are category queries: “best [category] near me,” “top tools for X.” We focus on category queries because that's the moment a customer is choosing and you either get named or you don't.
How is “share of voice” calculated?+
For each buyer prompt, the engine returns the set of businesses it would recommend. Your share of voice is how often you appear in those sets relative to the competitors named alongside you, across all the prompts and engines we track for you. Rising share of voice means AI is naming you more often, more competitors are losing ground to you, or both.
Why might two runs give slightly different answers?+
Because that's how AI assistants genuinely behave — their answers are probabilistic and change as the underlying models update and as they personalize. We don't hide that. We run at the most stable setting available and average across repeated checks and multiple prompts so you're looking at a real trend, not noise.
What do you deliberately NOT claim?+
We don't claim to see inside any AI company's private ranking systems, we don't guarantee you'll be recommended, and we never fabricate a score. AI visibility reflects what the models say from their general knowledge at the time we ask. What we can do is measure it honestly, show you where you stand versus competitors, and give you the specific content, schema, and technical work that improves your odds.
See where you stand — honestly
Run a free check on your own business and see exactly which AI engines name you today, and which name your competitors instead.