1. Create a project
A project tracks one brand. Give it the brand name, the brand's website domains (used to spot when an AI cites your own site), a category, the competitors to compare against, and a target market.
The market sets the language and region the measurements run in, independently of your interface language. Prompts and reports keep the market they were created under even if the project switches markets later.
2. Add prompts
Prompts are the questions your customers ask AI assistants — "what's the best X for Y?". Reports run these questions and score how your brand comes up in the answers.
Write your own, or generate a researched set with AI from the project page (uses 1 credit). Aim for the questions a buyer would actually ask, not questions that already name your brand.
3. Run a report
Pick any combination of Claude, GPT, Gemini, and Perplexity, and optionally up to five samples per prompt and provider — repeat samples smooth out the run-to-run randomness of AI answers.
A report's credits are weighted by provider for each prompt and sample, and the projected cost is shown before you run. Progress updates live, and failed provider calls are not billed.
4. Read the results
Every answer is scored: whether your brand was mentioned, recommended, the top pick, cited, or framed negatively, and which competitors appeared. The report rolls those up into rates, competitor standings, prompt-level coverage, and the sources the answers cited.
Understanding your metrics explains each rate, why numbers move between runs, and when a change counts as significant.
5. Work the opportunities
After each scored report, the weak spots — prompts you're missing, competitors out-recommending you, negative framing — are consolidated into one ranked action list.
Ask for an AI summary across all of them (1 credit).
6. Put it on a schedule
Schedule daily or weekly runs from the project's reports page. Two email toggles sit alongside the schedule: a digest to members when each run finishes, and change alerts when a run's visibility shifts significantly — alerts cover manual runs too.
If a scheduled run can't start — no prompts for the project's market, or not enough credits — the reports page and a heads-up email tell you what's blocking it.