There is no official Google News API

There has not been one for fifteen years. If you came here looking for an API key from Google, that is why you could not find it — and here is what people actually use instead.

What happened to it

• 26 May 2011Google deprecates the News Search API, part of a wider cull of its search APIs.
• 15 January 2016Google announces the final turndown of the deprecated search APIs.
• 15 February 2016They stop responding. Nothing has replaced them since.

Google's own suggestion was the Custom Search API, which searches the web through Google's index but does not expose Google News. It is a different product answering a different question, which is why developers kept looking after being pointed at it.

What we do instead

Google News still publishes a public feed. The problem is not reaching it — it is that the feed alone is not usable for much:

• Its links are opaque redirects, not publisher URLs. You cannot fetch the article from them and you cannot tell reuters.com from a syndicator.
• It carries headlines and nothing else — no article body, so nothing you can hand to a model.
• It is XML with its own quirks, per country and per language edition.

We solve those three things and return JSON. Link resolution uses Google's own decoder rather than guessing from the headline — we tested the guessing approach and it was quietly wrong about one article in five, which is worse than failing.

curl "https://api.hawkcrawl.com/v1/news?q=tesla&from=2024-01-01&to=2024-03-31&content=true" \
  -H "Authorization: Bearer YOUR_KEY"

Measured, not claimed

From 636 articles measured on 18 September 2026, across thirteen topics sampled in the last seven days and again in 2010, 2016 and 2021:

An article from 2010 resolved as reliably as one from this morning. Full text is 77%, ranging from 40% to 100% by subject — the archive being the strong half, since a 2016 article is far less likely to sit behind a wall than one from this week. What we cannot get is paywalls, publishers that refuse machine traffic, and pages that no longer exist — and you are charged for none of them.

Why not just scrape the feed yourself

You can, and for a weekend project you probably should — we wrote the reference we wish had existed, with every endpoint, operator and language code tested. What you will hit, in roughly this order: the link redirects that do not resolve, Google serving datacenter IP addresses a different page than browsers get, the article extraction that returns navigation menus while passing every length check you write, and then the proxy bill for doing all of it at volume.

None of that is unsolvable — it is just a few weeks that are not your product. The documentation describes exactly what we return so you can judge whether those weeks are worth $39 a month to skip.

The other ways people do this

Competitor figures read from their own pricing pages on 16 September 2026.

Start with the documentation

Every parameter, the real response shape, what each call costs, and the failure numbers repeated wherever they matter. A new account starts with 1,000 free credits, and questions reach a person.

Read the docs See pricing