AI visibility you can verify
Witness asks ChatGPT, Perplexity, Claude, Gemini and Google AI the questions your buyers ask, many times over. You get every number with its margin of error, and the actual answers behind it.
462 of 1,200 answers. Up from 31% last month, beyond the margin of error.
Rank
2nd of 5
Top source
reddit.com
Measured on every major answer engine
Each engine measured on its ownMeasurement
AI answers change every time you ask
Ask the same question twice and an engine can name different brands. A score built from a handful of answers swings by ten points on noise alone. Witness samples enough to know, and shows how sure it is.
From how many answers? Which engine? Is last week's 34 a real change, or noise?
462 of 1,200 answers, 6 engines, 10 runs per prompt. Last month 31%: a real rise. Every dot opens the answer behind it.
How it works
Three steps, no black box
Prompts, captures and metrics are all inspectable. Brand detection is deterministic, so anyone can reproduce a number from the stored answers.
Ask what buyers ask
Witness drafts the questions buyers type into AI assistants from your category, competitors and personas. Prompts never name you, so results aren't self-fulfilling.
Capture real answers, many times
Each prompt runs repeatedly on every engine, from the real app where a capture provider is connected, from the engine's API otherwise. Every answer says which.
Measure with honest ranges
Mention rate, share of voice and citation share, each with its 95% range and sample size. When a change is inside the noise, Witness says so.
See each answer as the engine gave it, with your brand and competitors highlighted, the order they appear in and the sources cited. Then open the same question in the real app and check it yourself.
Ship a change, pick the prompts it should affect and a control set it shouldn't. If the engines shift for everyone, the control catches it. You get a verdict, Worked, No clear effect or Too early, with the range behind it.
Test prompts compared against a control set, so an engine-wide shift doesn't get credited to you.
Single sign-on with domain verification, tenant isolation enforced by the database, roles, two-factor and an audit log. Built in from the start, not bolted on.
SAML and OIDC single sign-on
Okta, Entra ID, Google Workspace, OneLogin. Verify your domain, then require SSO for everyone on it.
Organizations and projects
One organization for your team, a project per brand. Agencies keep each client in a restricted project.
Isolation in the database
Postgres row-level security pins every query to your organization. A faulty query still can't reach another tenant.
Roles and two-factor
Owner, Admin, Member and Viewer. Require two-factor across the organization and set session length.
Audit log
Every sign-in, invite, role change, connector and run is recorded and can't be edited.
Bring your own connectors
Plug in any capture provider or your own engine API keys. Keys are encrypted at rest.
Our promise
What Witness will never do
Promise placement in AI answers. Answers are probabilistic; nobody can guarantee them.
Show a number without the sample behind it.
Call a change real when it's inside the margin of error.
Hide whether an answer came from the real app or an API.