The Trends site loading is not a working API

"Google Trends API not working" is usually two different failures collapsed into one ticket. The website still plots a term. An unofficial library returns 429, an empty widget, or a timeout. Those clients scrape trends.google.com. They are not an API. Google's official Trends API is an application-gated alpha. Trends API is a third path: POST https://api.trendsapi.ai/api with a Bearer key. body is a JSON string and must be parsed a second time. Reproduction belongs on that POST. Docs: the API reference. Caps: pricing. Google notes: the Google Trends guide.

The website 200 does not mean the scrape client works

A screenshot of Explore is not a passing integration test. Unofficial packages send a different User-Agent, skip consent, and reuse a process-wide session. Google answers those with blocks that the HTML UI never shows. Debug the client, not the chart.

Consent, NID, and geo host changes

Scrape write-ups obsess over NID cookies, hl, and rotating trends.google.com versus a country host. Those knobs exist because the transport is HTML. They go stale when Google changes the handshake. pytrends is archived, so the next handshake change will not get a merge. That is a client end-of-life, not a regional outage.

Official alpha versus a key that exists today

The official alpha requires an application. No self-serve key. A team waiting on that form still has no working official client. Trends API keys are issued for this product. Do not file an official-alpha ticket against POST /api. Do not file a POST /api ticket against Explore.

Minimal POST that should return 200

POST https://api.trendsapi.ai/api
Authorization: Bearer <api_key>
Content-Type: application/json

{"mode":"get_growth","source":"google search","keyword":"bitcoin","percent_growth":["12M"]}

statusCode 200 plus a second parse of body is the pass. 401 is the key. 429 is burst or plan. An inner empty or no_data payload is a keyword miss, not "the API is down." The product MCP server at https://api.trendsapi.ai/mcp is the same modes if the failing runtime is an agent.