How does Brandtrace decide which prompts to track?
You enter your website address. Brandtrace scans the site and suggests the questions a buyer would ask before they know your name. You can accept the suggestions, edit them and add your own prompts.
Write prompts without your own brand name. Brandtrace flags a prompt that names your brand as branded, because the name pulls the answer away from how a buyer discovers you.
Which AI models does it ask, and how often?
Each tracked prompt runs on a schedule and the full answer is stored. Entry plans cover ChatGPT. Pro and above cover all major models, including Claude, Gemini and Grok.
A prompt can also be asked as a stated audience, called a persona. The question stays the same and only the asker changes, which makes it the cleanest comparison in the product.
The prompt limit of a plan counts each tracked prompt once, however many models it runs on. Each run on a model also uses credits from the plan's monthly allowance, and extra credits can be added.
What does it measure in an answer?
For every tracked brand it records whether the brand was mentioned, where it first appeared, how the answer frames it, and its share of voice against competitors. Share of voice is the share of brand mentions that went to you across the completed runs.
Only completed runs count in the denominator. A run that failed at the provider is shown as failed and never lowers a score, so a provider outage does not read as a loss of visibility.
Which sources do AI answers cite?
Citation analytics list the domains and pages that answers cite. For your own pages it shows whether the citation was retrieved from the web or generated from training data, because only the first kind can be influenced by changing a page.
Which AI crawlers visit my site?
AI crawler analytics show which AI crawlers reach your site, which pages they read and whether your robots.txt blocks them. The free AI readiness audit at /audit checks seven crawlability and citation signals and returns a score from 0 to 100, with no signup.
What should I do after reading the numbers?
Brandtrace lists recommended actions and the source gaps where competitors are cited and you are not. Alerts tell you when your share of voice drops or a competitor gains ground, and a weekly summary covers what moved.
Can my own code or an AI agent read the data?
Yes. The customer API returns projects, prompts and brand, domain and URL reports with an API key. The MCP server lets Claude, Cursor, VS Code Copilot and other MCP clients read the same data with OAuth sign-in, and it is available on Pro and above.
A complete implementation guide for AI coding agents is served at /llms-full.txt.
A key can be read-only, and the MCP server asks for the write scope separately. A client that connects with read access can report on the data and cannot change a project.
Is it built for agencies?
Agencies track several brands in separate projects and share reports with clients. Plans differ in how many projects, seats and prompts they allow, and /pricing lists the current limits and prices.