cURL AI Visibility API

You can't curl ChatGPT - the UI is JavaScript-rendered, login-walled and bot-protected. You can curl AgentGEO: one POST returns the raw answer each of the six engines gives - ChatGPT, Perplexity, Google AI Overviews, Copilot, Gemini - with citations and sources as JSON that pipes straight into jq. An AI answer becomes something a shell script can work with.

Call the AI Visibility API with cURL

One POST with your ag_live_ key as a bearer token. -d carries the query and the engines you want in surfaces; the answer, citations and sources come back as JSON. Pipe to jq to pretty-print - the full response shape is in the docs.

curl -s https://api.agentgeo.org/v1/fetches \
  -H "Authorization: Bearer ag_live_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{"query": "best running shoes for flat feet", "surfaces": ["chatgpt"]}' \
  | jq

The full response - every field addressable by a jq path:

{
  "id": "run_84c1f2ab91d3",
  "query": "best running shoes",
  "surfaces": ["chatgpt"],
  "status": "completed",
  "recordsDelivered": 1,
  "creditsCharged": 1,
  "answers": [
    {
      "surfaceKey": "chatgpt",
      "answerText": "For most runners, top picks include ...",
      "sources": [
        { "title": "Best Running Shoes", "url": "https://example.com/guide", "position": 1 }
      ]
    }
  ]
}

From there, extraction is jq. Wrap the call in a two-line shell function and the one-liners stay one-liners:

# A tiny helper so the one-liners stay one-liners
ag() {
  curl -s https://api.agentgeo.org/v1/fetches \
    -H "Authorization: Bearer ag_live_your_key_here" \
    -H "Content-Type: application/json" \
    -d "$1"
}

# Just the answer text
ag '{"query": "best running shoes for flat feet", "surfaces": ["chatgpt"]}' \
  | jq -r '.answers[0].answerText'

# Every URL the answer cited
ag '{"query": "best running shoes for flat feet", "surfaces": ["chatgpt"]}' \
  | jq -r '.answers[].sources[].url'

# Two engines, one call - per-engine citation lists, ready to diff
ag '{"query": "best running shoes for flat feet", "surfaces": ["chatgpt", "perplexity"]}' \
  | jq -r '.answers[] | .surfaceKey, (.sources[] | "  " + .url)'

And because it's just curl, it schedules. Append each run as a timestamped JSON line and you have a minimal visibility monitor that runs unchanged in cron or CI:

# Append one timestamped JSON line per run - cron- and CI-friendly
curl -s https://api.agentgeo.org/v1/fetches \
  -H "Authorization: Bearer ag_live_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{"query": "best running shoes for flat feet", "surfaces": ["chatgpt", "perplexity"]}' \
  | jq -c '. + {fetchedAt: (now | todate)}' >> visibility.jsonl

# crontab -e - track the answer every morning at 07:00
# 0 7 * * * /usr/local/bin/track-visibility.sh

What the cURL API gives you

AgentGEO is a thin answer-access layer, not a dashboard you log into. The API hands your scripts the raw material; whatever you pipe, diff, alert on or archive is yours.

• Six engines, one contract - ChatGPT, Perplexity, Google AI Overviews, Google AI Mode, Copilot and Gemini answer through the same endpoint; switching engines is editing one JSON array in your -d payload.
• Citations at a stable jq path - every cited source lives at .answers[].sources[] with title, url and position, so extraction is a one-liner, not a parser.
• Managed-scraper engine, maintained for you - a plain HTTPS POST that behaves like every other API you script: no browser, no cookies, no bot wall between you and the data.
• Provider metadata - surfaceKey and providerFields on every response keep a long-running JSONL log auditable as the models underneath change.
• MCP connection, too - the same data over MCP via agentgeo-mcp on npm, for the day the consumer is Claude Code or Cursor instead of a script.
• Usage-based billing - run allowances by tier with a spend cap, so a forgotten cron loop can't surprise you. Never per-seat.

Common problems with DIY answer-scraping from the shell

The engines' UIs are apps, not documents - which is why pure shell against them goes nowhere, and why every DIY route ends with a headless browser: the opposite of what you reached for curl to get.

• curl gets an empty shell of the page. ChatGPT and Perplexity render answers client-side; fetch the page from a script and the body is a JavaScript bootstrap with no answer text anywhere in it.
• No session from a script. The UIs sit behind login and rotating session tokens - nothing a shell script can legitimately hold on to, so requests never reach an answer at all.
• Bot walls flag curl instantly. Cloudflare-class protection exists precisely to stop non-browser clients, and curl's TLS and header fingerprint is the first thing it filters.
• Parsing rendered HTML in shell is hopeless. grep and sed against a hydrated React DOM is write-only code - and it dies on the next frontend deploy anyway.
• The workaround erases the point. Each dead end pushes you toward bolting on headless Chrome - and once a browser farm sits behind your 'one-liner', the simplicity that made curl attractive is gone.

MCP: when the consumer is an agent, not a script

If the thing consuming answers is an AI agent instead of a shell script - Claude Code, Cursor, Codex or any MCP client - the same data is available over MCP via the agentgeo-mcp npm package. The install is, fittingly, one command:

claude mcp add agentgeo -- npx -y agentgeo-mcp --key ag_live_your_key_here

Who builds on this

Anyone who wants answers without writing an application: a quick audit before a client call, a CI job that fails when your brand drops out of an answer, a cron loop building citation history in flat files. It's also the fastest way to evaluate the API before committing real code - start with the free audit, graduate to a key. And if you were about to script against ChatGPT's UI instead, read that comparison first.