Follow the evidence.
Start here.
Inspect sampled AI answers, citation sources, and observations from your existing account inside ChatGPT.
Developer preview. Public directory installation is not available yet. These instructions are for a test deployment. They do not establish that the production endpoint is reachable, approved, or published.
Website analytics measure setup and installation-link clicks as interest only, not verified installations or completed checks. No account IDs, prompts, or evidence are included in this click event.
1. Connect a deployed preview
You need ChatGPT developer access and an HTTPS Rank.ai preview endpoint supplied by the deployment owner. Account linking also requires configured OAuth.
- Get the verified MCP endpoint and enabled capabilities from your deployment owner. The intended production route is
https://api.rank.ai/plugins/rank/mcp; this guide does not establish that it is deployed. - Use the current OpenAI server guide to connect a remote MCP server in ChatGPT developer mode. Use your verified preview endpoint.
- Confirm the connection and advertised visibility tools, then enable Rank.ai in a new conversation. Sidebar or panel availability depends on the client; inline results are the fallback.
- Run a public check before testing private account access. For connection failures, ask the deployment owner to verify HTTPS reachability, transport, and authentication discovery.
Use OAuth for account linking. Never paste API keys, passwords, or access tokens into chat. The broader Rank.ai MCP integration has a separate setup and tool surface.
2. Ask one buyer question
Replace the example domain with your own:
Check example.com for this buyer question: “What should I look for in accounting software for a small business?” Show the sampled answers and citation evidence.
The public check requires no Rank.ai account, email, or purchase. It may reuse an eligible cached observation. Availability and operating limits apply; some providers may finish while others are pending or fail.
Inspect mentions, linked citations, exact model version, search method, observation time, and cache status. Open the source page before treating an excerpt as support for a recommendation.
Public results are temporary, with a current one-hour cache lifetime. These anonymous results are separate from the opt-in evidence snapshots available to connected accounts.
3. Connect existing account evidence
Ask to connect your Rank.ai account and follow OAuth authorization. Select an authorized site. Available reads depend on your account’s enabled capabilities.
- “Show citation evidence for my tracked prompts.”
- “Which observed competitors appear where my brand is absent?”
- “What changed in my tracked prompts this week?”
Comparisons require matching prompts, providers, methods, and enough history. Otherwise, expect an explanation of insufficient data.
Review before acting
Use selected sources to decide what to improve. Connected accounts can also inspect existing organic and local search evidence when available.
Confirm tracking and private drafts
Ask for a tracking proposal for your authorized site. Review the exact prompts, daily or weekly cadence, account allowance, and credit cost in the workspace before confirming. Asking for a proposal does not start tracking.
For content, ask for a draft proposal, review its title and maximum credit cost, then confirm generation in the workspace. The result is a private Rank.ai draft for review. Generation does not publish it; publishing requires a separate action outside this plugin.
Share selected evidence deliberately
Select observations to propose an evidence share. Review the exact redacted snapshot before confirming a public link. Anyone with that link can read the selected snapshot. Revoke an owned link when you no longer want it accessible; revocation cannot recall copies already made.
Write actions require an authorized account, the appropriate capability, and available allowance. The plugin does not run ads or initiate purchases. If the workspace confirmation control is unavailable, do not assume that a conversational approval executed the action.
What a check measures
Rank.ai makes model API calls through OpenRouter with Rank.ai search tools. These are sampled API answers, not recordings of consumer ChatGPT, Claude, or Gemini sessions. Consumer answers can differ with context, personalization, time, and product behavior.
- Mention: the sampled answer names the brand or domain.
- Citation: the sampled answer links a source. A mention and citation are different observations.
- Context: inspect the exact model version, search method, prompt, observation time, and cache status.
- Coverage: inspect successful and attempted sample counts. Failed checks are disclosed and excluded from the success denominator. Missing data is unknown, not zero.
- Comparison: use like-for-like observations. Changes to the model or search method can break comparability.
A single question cannot measure overall visibility. An absent citation does not establish its cause. Recommendations are hypotheses grounded in available evidence, with no guarantee of citations, ranking, traffic, or consumer-answer accuracy.
When something is missing
- Connection failed: confirm developer access and your deployment’s verified endpoint.
- A provider failed: inspect other results and the failure label; failure is not brand absence.
- Result expired: request a new check. Admission and budget limits still apply.
- Site missing: confirm the connected account and authorized sites.
- No trend: check for enough comparable historical samples.
Contact the team with the visible error and approximate time. Remove credentials and private content. Read our privacy policy and terms.
Technical references: MCP server guide, account linking, and submission and publication.