Integration3 min read2026-07-31

AI visibility in n8n

n8n is the natural home for GEO monitoring: a Schedule Trigger, one HTTP Request node against AgentGEO, and whatever routing you already know — Slack, Sheets, email, a database. No scraper to babysit, no vendor dashboard to log into. There is no "AgentGEO node" to install, and this page won't pretend there is. The integration is one HTTP contract — POST https://api.agentgeo.org/v1/fetches — which n8n's built-in nodes speak natively. If you're building an AI Agent workflow instead, the MCP guide for n8n covers the MCP Client Tool path in depth.

Read this page with an AI

One call returns the answers ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode and Copilot actually give — verbatim text plus every cited source as structured sources[] your later nodes can filter, diff or append to a Sheet.

Connect AgentGEO to n8n

Add an HTTP Request node — works on n8n Cloud and self-hosted alike. Method POST, URL https://api.agentgeo.org/v1/fetches, an Authorization: Bearer ag_live_... header, and a JSON body:

HTTP Request node — JSON body
{
  "query": "best ai visibility tools",
  "surfaces": ["chatgpt", "perplexity", "gemini"]
}

Set the node's timeout generously — live surfaces are slow, and the API holds the request up to 180 seconds while scrapers finish. The response is one run envelope with one record per surface: surfaceKey, status, the full answerText, sources[] with title, URL and position, plus latencyMs and providerRecordId. A live ag_live_ key is minted in the console and comes with a plan; an ag_test_ key returns clearly labelled demo records at zero credits, so you can wire the whole workflow before spending anything.

Prefer not to hold a 180-second request in a workflow? Create a schedule (in the console or via POST /v1/schedules) and let AgentGEO fetch server-side on an hourly, daily or weekly cadence — then your n8n workflow just polls GET /v1/runs, a fast read, and routes anything new. Same records, no long-held connection.

What to build once it's connected

  • Weekly visibility digest to Slack. Schedule Trigger → fetch your money queries across six surfaces → format who was cited where → Slack node.
  • Competitor-domain watch. Cron → fetch your category query → filter sources[] for competitor hosts → append a dated row to Google Sheets. A longitudinal record that lives in your Sheet, not a vendor's database.
  • Drop-out alert. Daily fetch → IF ourdomain.com missing from sources[] → email or Slack alert the day you fall out of the answer.
  • Post-publish check. After your CMS publishes, wait a week, fetch the target query, and log whether the new page earned a citation.

More integrations

The same contract works in Zapier, Make and Pipedream — or from plain cURL if you'd rather script it. See all integrations.

See where your brand stands before you automate: Get a free audit → · Endpoint reference →

Get my free audit

FAQ

Direct answers to the questions this page raises.

No, and none is needed: the built-in HTTP Request node speaks the whole contract, and AI Agent workflows can use the MCP Client Tool node with the agentgeo-mcp server. Both return identical records; the MCP path is covered in depth in the n8n MCP guide.

Yes. The HTTP Request path needs no local process, so it works on n8n Cloud and self-hosted identically. Only the stdio MCP path (spawning npx) requires a self-hosted runtime.

Longer than 180 seconds — 200 seconds is the convention. The API holds the request while live surfaces answer; an AI Overview SERP round-trip alone runs 40–90 seconds. If you'd rather not hold connections, use a server-side schedule and poll GET /v1/runs instead.

One delivered record costs one credit; failed records cost zero. A daily three-query, six-surface sweep is at most 18 credits a day, billed by consumption with a spend cap — predictable, and it can't run away.

Keep reading

Where this page leads next.

Run these checks on your own brand

Two ways in. Send a URL and a person runs the fetches for you, free — or connect your agent over MCP, on a plan, and run them yourself. Either way you get the raw answers and their citations, never a score.