Documentation
Getting Started
A practical guide to setup, sampled AI answers, accuracy checks, and diagnosis evidence.
Quick Start
Get access, choose the questions you want Discover to measure, and review the setup available to your workspace.
Request access
Discover is not open for self-serve sign-up yet. Use Request beta access on the Discover page. Accepted members get a dashboard account.
Until then, the free AI visibility check and hallucination scanner run on a single URL without an account.
Complete workspace setup
Follow the setup instructions provided with your Discover access. If guided setup is enabled for your workspace, confirm the business details, choose customer questions, review any suggested competitors and facts, then check the measurement plan.
Choose questions that reflect customer intent
Choose the templates that fit your business or add your own, within your plan's limits. Use wording and locations your customers actually use. Comparable questions make it easier to read changes over time.
Discover measures the engines included with your plan, such as ChatGPT, Gemini, Perplexity and Claude. The exact set depends on the plan and capabilities enabled for your workspace.
Confirm facts when accuracy checks are available
Review the business facts you want checked, such as hours, locations, prices, and policies. Discover compares claims in collected answers with facts you confirm. A claim without enough information to check remains unverifiable, not wrong.
Review the measurement plan
Check the questions and engines selected for your workspace before finishing setup. Measurement starts only after required setup is complete and enabled for your account. Sampling runs on its schedule, so results take time to collect.
Read visibility measurements
Discover reports sampled answers, estimated rates, ranges, and change verdicts. These measurements do not promise rankings, citations, or traffic.
Open the answers behind a result
Start with the questions and engines in the report. Read collected answers to see what was observed, including whether an answer mentioned, linked to, or recommended your business.
Separate counts from estimates
Collected answers and mentions are counts. Rates are estimates from those samples and include a 95% range. Read the range with the estimate instead of treating one percentage as exact.
Use change verdicts with their limits
A change may be real, possible, within normal variation, or preliminary. Preliminary means there is not enough complete, comparable evidence yet. A question needs at least 10 answers in a period before its rate is shown.
Compare like with like
When reviewing a trend, check that the question wording, engine, location, and collection method are comparable between periods. An engine or measurement change can make a direct comparison preliminary.
Review accuracy and diagnosis
Check claims against facts you confirmed, then use diagnosis evidence to decide what to address. Availability depends on your workspace.
Keep confirmed facts current
Accuracy checks compare claims in answers that mention your business with facts you confirm, such as hours, locations, prices, and policies. Keeping those facts current gives the comparison a reliable reference.
Read each claim in context
Review the answer beside the fact it was checked against. A claim is marked wrong only when it contradicts a confirmed fact. If there is not enough information to check it, the claim remains unverifiable.
Follow diagnosis evidence
Diagnosis connects findings to evidence such as an access check, an answer, or a tracked question. It distinguishes issues you can fix from factors you can influence or those outside your control.
Choose a next step and review its evidence label
Suggested fixes include steps and an evidence label. Tested, Observed, Unproven, Tested, no effect describe how strongly a fix is supported, not a guaranteed outcome. Make changes in the relevant source yourself. Discover does not publish them for you.