Visibility in AI answers, measured monthly
What do AI models recommend when buyers ask about your category?
25 buying prompts, 4 models, 2 languages — every month. You get a score from 0 to 100, your share of all recommendations, and the list of questions no model can answer about you.
- 25 prompts in 5 blocks
- 4 models with live web search
- Monthly, automatic
In 62% of buying prompts the company does not appear at all.
The strongest competitor sits at 80%.
Questions no model could answer about the company.
Google rankings no longer tell you whether you get recommended
Buyers do not open a results page any more, they ask a model: "Which CRM fits a trade business with 20 people?" What comes back as a shortlist decides your pipeline — and nobody measures it.
That is where the Monitor starts. Every month it asks the same 25 questions your buyers would ask, across every relevant model, and records who gets recommended, what is claimed about you, and where a gap opens up.
Process
Four minutes of setup, then it runs monthly
- 01
Create a case
Domain, category, ideal customer, 3–5 competitors, 2–3 use cases. Four minutes. Plus a fact sheet with your real prices — without that ground truth there is no way to judge whether a model tells the truth about you.
- 02
Measure the baseline
We generate 25 buying prompts in five blocks: category shortlist, facts about you, use case, buying criteria, head-to-head against your competitors. Every prompt runs against every model in every language — with live web search, the way your buyer sees it.
- 03
Have it scored
Every answer is judged against your fact sheet: visibility (0–2), accuracy (0–2), sentiment (−1 to +1), the dominating competitor, and the open gaps. That yields a score from 0 to 100, per model and overall.
- 04
Repeat every month
The run repeats automatically. You see the trend, the delta to last month, and get a report by email. If the score drops by more than 10 points or a new competitor shows up in the recommendations, you get an alert.
In the dashboard
What you get every month
Score 0–100
One value per model and one overall, built from visibility, accuracy and sentiment. Trend chart across every month measured.
Share of answer
The share of comparison questions where you are named — next to every competitor. "Craftnote is recommended in 80% of comparisons, you in 40%."
Answer matrix
Every single answer verbatim, sorted by prompt and model, with its score and reasoning. Filterable by block and language.
Gaps
What the models could not answer about you — deduplicated and ranked. That is your FAQ backlog for the next few weeks.
Alerts
Email when the score drops by more than 10 points or a competitor newly enters the recommendations. Pro plan.
API instead of click paths
Everything the dashboard does, the API does too. Bearer tokens, an OpenAPI spec, one endpoint per job — built for agents that ask on their own.
The models we measure
Measured through OpenRouter with live web search — the mode your buyer actually uses in a chat window.
- OpenAI GPT-5.6 Solfrom Trial
- Anthropic Claude Sonnet 4.6from Trial
- Perplexity Sonar Profrom Starter
- Google Gemini 3.1 Profrom Starter
Pricing
One case costs less than an hour of agency time
Trial
Start here- 14 days free, no credit card
- 1 case, 25 prompts, baseline run
- 2 models (ChatGPT, Claude), English
- Score, share of answer, gaps
Starter
Most chosen- 1 case, monthly measurement run
- 4 models: ChatGPT, Claude, Perplexity, Gemini
- English + German
- Monthly report by email
Pro
- 3 cases (e.g. several products or markets)
- 4 models, English + German
- Alerts on a score drop over 10 or a new competitor
- Public share links for reports
- Manual re-runs, any time
All prices net per month, cancel any time. No setup fee, no minimum term.
FAQ
Common questions
How is this different from SEO tools?
SEO tools measure rankings in a list of results. We measure answers. There is no position 1 to 10, there is prose in which three vendors appear and the rest does not exist. So we measure mention, accuracy and sentiment — not position.
Why do I need a fact sheet?
Without ground truth there is no way to judge whether an answer is right. Once your prices, plans and audience are on file, every answer can be checked against them — and you see immediately when a model quotes an outdated price. That is why prices and plans are mandatory.
How often do you measure?
Automatically once a month, plus a baseline run right after you create the case. On the Pro plan you can measure again at any time — after a relaunch, for example.
Which models do you query?
On the trial, ChatGPT and Claude. From Starter, Perplexity and Gemini as well, plus a second language. All with live web search, so the answer matches what a buyer sees in a chat.
How are the answers scored?
A separate model without web search judges every answer strictly against your fact sheet and competitor list. It gets four numbers to fill in: visibility 0–2, accuracy 0–2, sentiment −1 to +1, and the dominating competitor. Score = visibility × 25 + accuracy × 15 + (sentiment + 1) × 5, normalised to 0–100.
How does this relate to the AgentReady Check?
The Check measures the technology of a domain: robots.txt for AI crawlers, llms.txt, structured data, server-rendered HTML. The Monitor measures the outcome: what the models actually answer. Every monthly run triggers a fresh Check scan, so both numbers sit side by side — and a drop in visibility gets a cause instead of a shrug.
Is there an API?
Yes. Every dashboard function exists as an endpoint under /api/v1, authenticated with a bearer token. The OpenAPI spec lives at /api/v1/openapi.json, the documentation at /docs/api. An MCP server mirroring exactly these endpoints follows.
Can I cancel?
Any time in the customer portal, effective at the end of the current period. The trial runs for 14 days and needs no credit card.
Find out what the models say about you
14 days free, no credit card. Or start with the free technical scan.