To run a check, open Checks, choose New check, pick the assistants and samples per prompt, and start it; the form shows the credit estimate (prompts × assistants × samples), which is reserved when the check starts. Owners, admins and members can run checks.
What you choose
- Assistants: ChatGPT, Claude and Perplexity are ticked by default; Gemini and Google AI Overviews can be added. Each shows its evidence scope and the model it answers with, and says so if that model can't be used for checks right now (see Choosing models).
- Samples per prompt: 3 (the default) or 1.
- Brand name: the name the answers are read for. It starts as the workspace name.
- Competitors: up to 5. Suggest competitors asks Claude (Anthropic), with up to 2 web searches, for up to 5 real businesses buyers compare you with, using the brand name you typed and your homepage when your website is set in Settings. Each suggestion shows a short reason; nothing goes in until you choose Add, and Add all fills only the empty slots, never one you typed. It uses no credits and can be run 5 times an hour per workspace, so check each name before adding it.
How is the estimate worked out?
1 credit = 1 sampled answer from one assistant, including its analysis. The estimate is:
prompts × assistants × samples
For example, 10 prompts × 3 assistants × 3 samples = 90 credits. The form updates the estimate as you change your choices and tells you if it is more than the balance.
Reserve, then settle
The estimate is reserved when the check starts. When it finishes, only the usable answers are charged; any answer that didn't come back is refunded. If no answer comes back, the check shows Failed: no usable answers and everything is refunded; see A check says no answers.
The Checks list shows both numbers: Estimate (reserved) and Used (charged).
Reading a check's answers
Open a check from the Checks list. Its page shows the status, progress, credits reserved and used, what was asked, and every answer, grouped by prompt.
- Each answer card names the assistant, its evidence scope, the sample number and the model that answered, with the answer's full text.
- Sources lists the pages the answer cited, as links. A usable answer that cited nothing says No sources surfaced.
- An answer marked No answer or Not asked says in plain words what happened and what you can do, and wasn't charged. Most of these are temporary: running the check again usually fills them in.
Results, sources and exports
A finished check also shows its Report (visibility, average naming order, answer shares, gaps, each assistant's numbers, coverage, Google AI Overviews presence, the comparison with an earlier check, and every cited website), with times in the workspace's time zone. From the report, Exports downloads a CSV with one row per answer, and Print / save PDF opens a printable report that your browser's print dialog can save as a PDF.
If some answers didn't come back, the top of the report says how many it has (for example, "We got 3 of 4 answers"), that you weren't charged for the missing ones, and that the figures use the answers it has. Such a check isn't used for comparisons or trends; running it again usually fills the gaps. Your first check with a setup has nothing to compare against yet, and a forecast needs a few checks with the same setup, so those parts fill in as you run more checks. Perplexity answers are shown in the app only, so their text and wording are left out of the CSV and the printed report; their numbers and citations are kept. Checks run before results were recorded show their answers but no results or exports. Owners and admins can also download all of a workspace's prompts, checks and results as a JSON file; see Exporting a workspace.