DEV Community

Cover image for Your AI news feed is missing half the story: one search, four Google News editions, 765 headlines
Fast Data
Fast Data

Posted on

Your AI news feed is missing half the story: one search, four Google News editions, 765 headlines

#ai #news #datascience #python

If you monitor a topic through Google News, whether in an RSS reader, an n8n workflow or a script, you are almost certainly reading one country's edition. Usually that is the US one, because it is the default.

I wanted to know how much that matters, so I ran the same searches in four English-language editions and compared the results. The short answer: a lot more than I expected.

Disclosure: I built the Apify actor used to collect this data (Google News Scraper). Everything below can also be reproduced with the free Google News RSS feed and a few lines of Python; I show how at the end.

The setup

  • Queries: AI, AI startup, artificial intelligence, AI model
  • Editions: US, UK, India, Australia (all English)
  • Time range: last 7 days (15 to 22 October 2026)
  • Pulled: 22 October 2026, around 08:00 UTC
  • Matching: two results count as the same headline when the title is identical

Each edition returned between 365 and 376 unique headlines. Across all four editions there were 765 different headlines.

Result: only 15% of headlines are shared by everyone

Same AI search in 4 Google News editions

Edition Unique headlines Only in this edition
US 370 95
UK 365 105
India 376 118
Australia 370 82
All four 765 total 118 in all four (15%)

So any single edition showed you less than half (about 48%) of what the four editions surfaced together. Between 22% (Australia) and 31% (India) of each feed appeared in no other edition.

One caveat: matching on identical titles undercounts overlap a bit. When two outlets cover the same story with different headlines, that counts as two items here. The point still stands: each edition ranks and selects different sources, not just different wording.

What the local-only headlines look like

The exclusive headlines are not noise. They come from each country's own business and public broadcasters:

  • US-only: Reuters (7), WSJ (5), Axios (4), The New York Times (4). Examples: a Philadelphia-area software firm buying its third AI startup for $40 million; a WSJ piece on AI agents causing headaches for online sellers.
  • UK-only: BBC (8), Financial Times (8), The Guardian (7), Tech.eu (4). Lots of European funding rounds you won't see in the US feed, such as small seed rounds for Nordic and EU startups.
  • India-only: NDTV (8), Moneycontrol (5), Financial Times (5), The Economic Times (5). Indian startup funding (for example an AI hiring startup raising Rs 39.3 crore) and Indian policy coverage.
  • Australia-only: AFR (11), The Conversation (8), The Guardian (7), ABC (7). Bank job cuts linked to AI, AI training on Australian music, a police warning about AI and youth crime.

If you work in sales, investing, policy or recruiting, these are exactly the stories you want, and a US-only feed hides them.

Even the "big company" picture shifts

I also counted how many headlines in each edition named the big AI companies (whole-word match on the name):

Edition Most-mentioned companies
US OpenAI 20, Meta 15, DeepSeek 14, Anthropic 11, Apple 11
UK Nvidia 13, Anthropic 13, OpenAI 11, Meta 11, DeepSeek 10
India OpenAI 17, Google 13, Nvidia 13, Meta 13, Amazon 11
Australia OpenAI 16, Meta 12, Anthropic 11, DeepSeek 10, Apple 9

OpenAI leads in three editions, but in the UK feed Nvidia and Anthropic come out ahead. In India, Google and Amazon rank much higher than elsewhere. If you report share of voice from one edition, your ranking depends on which country you happened to pick.

The fix: one URL parameter

Google News RSS URLs take three edition parameters. This is the India edition of a search:

https://news.google.com/rss/search?q=AI+startup+when:7d&hl=en-IN&gl=IN&ceid=IN:en
Enter fullscreen mode Exit fullscreen mode
  • gl is the country (US, GB, IN, AU, DE, ...)
  • hl is the interface language (en-IN, en-GB, de, ...)
  • ceid is country:language (IN:en)
  • when:7d inside the query limits results to the last 7 days

Here is a minimal script that reproduces the overlap count with the free feed (standard library plus feedparser):

import feedparser
from urllib.parse import quote_plus

EDITIONS = {"US": "en-US", "GB": "en-GB", "IN": "en-IN", "AU": "en-AU"}
QUERIES = ["AI", "AI startup", "artificial intelligence", "AI model"]

titles = {}
for gl, hl in EDITIONS.items():
    seen = set()
    for q in QUERIES:
        url = (f"https://news.google.com/rss/search?q={quote_plus(q + ' when:7d')}"
               f"&hl={hl}&gl={gl}&ceid={gl}:en")
        seen |= {e.title for e in feedparser.parse(url).entries}
    titles[gl] = seen

union = set().union(*titles.values())
common = set.intersection(*titles.values())
print(len(union), "headlines,", len(common), "in all editions")
Enter fullscreen mode Exit fullscreen mode

Each RSS query returns up to about 100 items, so for broad topics you will want more specific queries rather than one big one.

When the free feed is enough, and when it isn't

The free RSS feed is enough if you only need titles, sources and dates for a handful of queries and you are happy to run the script yourself.

It gets annoying when you need the real article URL (the RSS links are Google redirect URLs, not the publisher's), the full article text for an LLM summary, or many queries across many editions on a schedule. That is the part my actor automates: it takes a list of queries, an edition (18 countries and languages) and a time range, resolves the original URLs and can optionally extract the full text. It costs $2.50 per 1,000 articles, plus $2.50 per 1,000 full texts if you turn that on. For this analysis I left full text off, which is the cheap mode.

If you already use n8n, I published a free workflow that sends a daily competitor news digest to Slack: data-recipes / daily news alert. To cover more countries, run the same node once per edition and merge the results.

Takeaways

  1. Google News is not one feed. Each country edition picks different sources, and here only 15% of headlines appeared in all four.
  2. Monitoring from one edition shows you under half the coverage and skews your share-of-voice numbers.
  3. Add gl, hl and ceid to your RSS URLs and loop over the markets you care about. It is a two-minute change.

If you run this for other topics or languages, I'd like to see whether the overlap is similar. My guess is that it is lower for non-English editions.

Top comments (0)