Visibility
The share of answers that mention your brand, per model and per prompt. Your name, aliases and domain are matched exactly first.
AI visibility
Autoranko asks AI models the questions your buyers ask and checks whether your brand comes up — and says exactly what that measurement is, and what it isn’t.
Default mode: model knowledge. These are answers from GPT-6 and Claude 5.5 models through an API, without web search — not what ChatGPT.com or Claude.ai show a particular person.
What we measure
Every chart and number carries the mode it was measured in, and trends only compare runs of the same mode.
The share of answers that mention your brand, per model and per prompt. Your name, aliases and domain are matched exactly first.
When you’re mentioned, where you appear among the brands an answer names. First place is the recommendation people read first.
Your mentions as a share of all brand mentions across the answers: how much of the conversation is about you rather than your competitors.
How often repeated samples of the same prompt agree. AI answers vary from run to run; stability tells you how far to trust a single result.
Two modes
AI assistants answer from two sources: what the model learned in training, and what it finds on the web right now. We measure them separately and never mix them.
We send your prompt to the model through its API with web search off. The answer reflects what the model learned about your market — slow to change, and shaped by what was published about you before its training.
When a model provider returns real web-search traces, a run can record the searches the model actually made and the sources it cited. We only call them “observed” when the provider gives us those traces.
The ChatGPT, Claude, Gemini and Perplexity apps add personalization, memory, location, their own browsing and model versions that change over time. Our numbers are a consistent, repeatable measurement of the models — a strong indicator, not a screenshot of anyone’s screen.
AI visibility · example
Model knowledge| Buyer prompt | GPT-6 | Claude 5.5 |
|---|---|---|
| Who installs heat pumps near me? | Mentioned | Mentioned |
| Heat pump or furnace for an old house? | Mentioned | Not mentioned |
| How much does heat pump servicing cost? | Not mentioned | Not mentioned |
Never mentioned for a prompt? Turn it into an article.
Engines
We only track engines we can actually query, and we name the exact model families.
Queried through our AI gateway. The closest model-level view of what ChatGPT knows about your market.
Queried through the same gateway. Claude powers Anthropic’s own assistant and many products built on it.
Not tracked by default. They appear only once a live-search provider for them is configured and tested — and we won’t publish a tracker page for an engine until it is.
If a model family is unavailable during a run, the run is marked partial and its missing answers stay missing — we never fill them with another model’s answer under its name.
Method
Published so you can judge the numbers yourself.
We draft buyer questions in your market’s language from your site — without your brand name in them — and you can edit, add or import your own.
Each prompt is sent exactly as written, as the only message. No system prompt names you, and nothing asks the model to list companies or links.
Where your plan allows, each prompt runs more than once per model, because answers vary. That’s what stability measures.
Your brand is matched first by exact name, aliases and domain. A separate, cheaper model call then lists the other companies an answer names.
A prompt where you’re never mentioned links straight to a new article on that topic, so measurement turns into content.
The answer request, verbatim
model: one GPT-6 or Claude 5.5 model, via its API
messages:
- role: user
content: "<the tracked prompt, exactly as written>"
system prompt: none
web search: off (mode: model knowledge)The prompt goes out unchanged, in the language you wrote it in. Extraction runs as a second, separate call, so it can’t influence the answer being measured.
AI gateways have bad days. A run that loses a model family is marked partial, counts only completed answers and shows which ones failed.
Plans and limits
Prompts, model families, samples and cadence, straight from our plan limits.
| Plan | Tracked prompts | Model families | Samples per prompt | Automatic re-runs | Extra runs a month |
|---|---|---|---|---|---|
| Free | 5 | 1 family | 1 | Monthly | 1 |
| Starter | 20 | up to 2 families | 2 | Monthly | 2 |
| Pro | 50 | up to 2 families | 2 | Weekly | 4 |
Model families available today: OpenAI GPT-6 and Anthropic Claude 5.5. A plan’s family count is a cap, not a promise of more engines.
Before AI models can learn about you or cite you, crawlers have to reach your pages and understand them. Our free tools check that part — access for AI crawlers, structured data, answer-first content — and they say plainly that a high readiness score doesn’t mean you’re mentioned.
Straight answers about what the numbers mean.
Start free and track 5 buyer prompts. No card needed.