ChatGPT can connect to the same Mirror Panel MCP server Claude uses, so it can draft and launch studies, list focus groups, and read results for your account — the full tool set, with nothing set up separately.
Every tool Claude gets is available to ChatGPT: the endpoint catalog, focus-group creation with its required fields spelled out, results and persona evidence, brand settings, and the low-level request fallback. Result-producing tools also return the same inline Mirror Panel result panels, which ChatGPT renders through the MCP Apps standard when it can and falls back to text otherwise.
Each row shows the concept's own wording, trimmed to the part that differs from the others. A ranking panel reports first-choice share — how many panelists picked each concept first — because that is the question an A/B or ranked-choice study answers. It is not the same as the average of the rating scales: two concepts can average the same while one takes two thirds of the first choices. The winner badge appears only when one concept is genuinely ahead, so a dead heat shows as a dead heat.
A 50-persona A/B test read from a chat: each row shows the concept's first-choice share and the raw count behind it.
The connection is tied to your account and honours your existing brand permissions. Disconnect it from ChatGPT's connector settings at any time; Mirror Panel keeps no ChatGPT-specific credential.
The connector works from any ChatGPT conversation. If you would rather ChatGPT help you inside the Mirror Panel page itself, see Use Mirror with a browser agent, which covers the in-page WebMCP tools and their review boundary.