Two ways, and neither needs a card up front. Request the free AI Visibility Report and we run your categories by hand — no account at all. Or start Starter, which opens with a 7-day free trial you can cancel inside the app.

Six: ChatGPT, Google Gemini, Google AI Overviews, Grok, Perplexity and Claude. ChatGPT, Gemini, Google AI Overviews and Grok are on by default. Claude and Perplexity are switched on per brand, so you can spend your prompts where your buyers actually are.

Four structural limits — brands, categories, monitoring prompts and tracked competitors — plus a monthly credit balance that engine runs and any AI generation draw on. Starter is 1 brand / 3 categories / 75 prompts / 5 competitors and 25,000 credits. Growth is 2 / 10 / 200 / 10 and 60,000 credits. Business is 3 / 25 / 500 / 20 and 150,000 credits. A run costs 70 credits per prompt per engine, so you can see exactly what a cadence will spend before you set it. Nothing is metered by page views, SKUs or seats.

For AI visibility monitoring, none — give us your URL and we discover your brand, categories and competitors from it. For the web analytics you add one script tag, or install @brandkarma/tracker from npm if you are on React or Next.js. The content API is a plain authenticated GET.

No. Monitoring reads public AI answers and needs no access to your site at all. Analytics is one beacon, which is a plain image request and therefore platform-agnostic. The content API is something you call, not something that rewrites your pages.

Not for analytics. The tracker writes no cookies, no localStorage and no sessionStorage, so it stores nothing on the visitor's device and adds no consent banner. Sessions and unique visitors are estimated server-side from a hash of user agent and IP address; the salt rotates daily and is never written to disk, and IP addresses are discarded after a coarse geographic lookup. Your own site may still set strictly necessary cookies for things like a CDN or bot protection — those are exempt from consent, and none of them come from us. Traffic is still personal-data processing under GDPR, so confirm your lawful basis with counsel.

You can open any score and read the exact prompt, the engine and the raw answer behind it. Every run is stamped with the model set that produced it, and the timeline flags the date that set changed, so you are never shown our instrument drift as your progress.

The next scheduled run is skipped rather than half-completed, and it picks up again when your monthly allowance refreshes with your billing period. Nothing is charged beyond your plan without you asking for it. If you are consistently short, the fix is usually fewer prompts or a longer cadence rather than a bigger plan — a run costs 70 credits per prompt per engine, so the arithmetic is yours to do.