SerpApi Google News → APITube migration kit

Everything you need to move off SerpApi's engine=google_news and onto the APITube News API: a parameter map, a drop-in shim for Node.js and Python, and an honest list of what does not carry over.

Every mapping here was executed against the live APITube API before being written down. Where something does not work, it says so.

Read this first: you are replacing a scraper with an index

SerpApi does not have a news database. It runs a Google News search on request and parses the page. That single fact explains almost every difference below:

	SerpApi google_news	APITube
What you get back	Title, link, source name, date, thumbnail	Full body and body_html, plus everything below
Article text	None — you follow the link and fetch it yourself	Included in the response
Pagination	None. One request, one page of results	page + per_page, up to 250 per request
Filtering	Google's q operators (site:, when:, -, OR)	60+ structured parameters
Enrichment	None	Entities, sentiment (title/body/overall/per-entity), IPTC categories, topics, industries, publisher bias and rank, readability
Result identity	Opaque tokens (story_token, topic_token)	Stable integer ids
Determinism	Google's ranking, which changes between calls	An index you query with explicit filters

If your code follows every link and fetches the article itself, the migration deletes that entire stage.

The three things that break silently

1. There is no pagination to port. SerpApi's Google News endpoint has no num, start or page parameter — you get one page and that is the query. Code written against it usually loops over queries rather than pages. On APITube the same result set is per_page=250 plus page=1,2,3…, and the count comes from a separate endpoint. This is a gain, but it changes the shape of the loop.

2. Tokens do not translate. topic_token, publication_token, section_token, story_token and kgmid are opaque Google identifiers. There is no lookup table, and nothing on APITube accepts them. Every token-based query has to be re-expressed as a filter — the shim tells you which one.

3. NOW-30m means thirty MONTHS. APITube's relative date syntax reads m as months, not minutes. Verified: published_at.start=NOW-30m returns 28 509 of the 29 744 title=tesla articles — effectively no filter. Minutes are not expressible; use NOW-1h as the floor.

NOW-1h   →  2 articles      (hours)
NOW-24h  →  80 articles
NOW-1d   →  80 articles     (days — matches 24h, as it should)
NOW-7d   →  1 611 articles
NOW-1MONTH → 4 632 articles
NOW-30m  →  28 509 articles ← thirty months, not thirty minutes
NOW-1y   →  27 795 articles

What is in here

Path	What it gives you
reference/parameter-mapping.md	Every google_news parameter → its APITube equivalent
reference/query-operators.md	Google's q operators (site:, when:, OR, -, quotes) translated
reference/response-mapping.md	news_results field by field, and what APITube adds
reference/limitations.md	What does not carry over, plus live API quirks
shim/node/	SerpApiNewsShim — accepts google_news parameters, returns the SerpApi response shape
shim/python/	Same shim for Python
examples/	Before/after request pairs
tools/ai-migration-prompt.md	System prompt for Claude/ChatGPT that converts your queries

Quick start

# SerpApi
curl "https://serpapi.com/search?engine=google_news&q=tesla&gl=us&hl=en&api_key=YOUR_KEY"

# APITube
curl "https://api.apitube.io/v1/news/everything?title=tesla&source.country.code=us&language.code=en&per_page=250" \
  -H "X-API-Key: YOUR_API_KEY"

Note what changed: gl and hl are Google interface settings — the country and language of the search you are simulating. source.country.code and language.code are properties of the article. They are close enough to be useful and different enough to check.

Or use the shim

import { SerpApiNewsShim } from './shim/node/serpapi-news-shim.js';

const client = new SerpApiNewsShim({ apiKey: process.env.APITUBE_API_KEY });

const response = await client.search({
    engine: 'google_news',
    q: 'tesla site:theguardian.com when:7d',
    gl: 'us',
    hl: 'en'
});

for (const item of response.news_results) {
    console.log(item.position, item.title, item.source.name);
}

Same parameters in, same news_results shape out, APITube underneath — and a warning for every conversion that loses something.

site: works, but check your domain first

site: maps to source.domain, and the parameter itself is exact. Two things to verify before you rely on it.

Coverage is not uniform. A domain that is indexed but has no articles returns 0 results with a 200 — no error to catch. On 27 July 2026, theguardian.com, apnews.com, techcrunch.com and bbc.co.uk all returned four-figure counts, while bloomberg.com, nytimes.com and forbes.com returned double or single digits, and reuters.com and cnn.com returned zero. A domain missing from the source index entirely returns 400 ER0214 — www.reuters.com does, while reuters.com does not.

Counts for a single domain are not currently stable. Repeating the identical source.domain count request returns alternating values — theguardian.com alternated between 3 658 and 4 831 across six consecutive calls, bbc.co.uk between 909 and 741 — and the source object attached to the returned articles is sometimes a different publisher's record (the articles themselves are correct; the source.domain and source.id fields on them are not). Filters that do not resolve a domain, such as title=, are stable across repeats.

Until that settles, treat a domain count as an order of magnitude rather than a number, and do not key anything on source.id from a domain-filtered query.

# stable
curl "https://api.apitube.io/v1/news/count?api_key=YOUR_API_KEY&title=tesla"

# alternates between two values on repeat
curl "https://api.apitube.io/v1/news/count?api_key=YOUR_API_KEY&source.domain=theguardian.com"

License

MIT. Use it, fork it, ship it.