Mainstream news isn’t looking.
Same AI stories, two different news worlds. Every morning we ask Tavily the same question — “what’s been published about artificial intelligence in the last 24 hours?” — once for each mainstream news outlet, and once for each tech publication, and we count what comes back. The gap between the two is the number below.
For every AI story mainstream news ran today, tech publications ran 0.7×.
Ratio this week: 0.8× — tech to mainstream.
Capability, safety, policy. These are the three topics that track whether AI is actually getting more capable, whether anyone is worried enough, and whether anyone’s in a position to act. Volume elsewhere (scandal, business, jobs) is noise by comparison.
The leading indicator. Model releases, benchmarks, breakthroughs.
Alignment, red-teaming, researcher warnings, existential risk.
Regulation, legislation, courts, government action.
Wire services break a story; mainstream outlets syndicate it. Each shows up once in our volume count even though it’s the same underlying piece. We cluster near-identical titles (Jaccard on normalised title tokens) so the real signal shows: how many distinct stories each side is actually running.
- Other
- Business
- Capability
- Policy
- Scandal
- Safety
- Quirky
- Business
- Other
- Capability
- Safety
- Scandal
- Jobs
- Quirky
Mainstream news today · 22 Oct
- BBC News10
- Reuters9
- Washington Post9
- Associated Press8
- The Guardian7
- New York Times7
- Al Jazeera6
- CNN5
- Fox News3
- The Telegraph3
- The Times1
- Sky News0
Tech publications today · 22 Oct
- Gizmodo10
- The Next Web9
- The Verge8
- TechCrunch7
- Axios6
- Ars Technica4
- VentureBeat3
- Wired1
- Engadget1
- Futurism0
Exactly what we’re counting.
Once a day, after the daily agent run, we fire one Tavily News query per outlet with a filter pinning it to that outlet’s domain and a 24-hour window. The query is the single phrase below, kept broad enough to catch the AI beat across publication styles and narrow enough to filter out unrelated stories.
Per-outlet “story count” = the number of results Tavily returns. The ceiling is 15, so a publisher running 20 AI stories in a single day shows as 15. Good enough to compare the shape and the ratio; not precise to the last story.
On top of the volume count we classify each sample headline into one of 8topic buckets via a single DeepSeek call per day. That’s what drives the “what are they covering?” section above — the real thesis-relevant signal. Volume alone gives mainstream credit for scandal and celebrity coverage; the topic mix reveals what’s actually being reported.
Outlets we query — mainstream
- BBC News
- The Guardian
- Reuters
- Associated Press
- CNN
- Fox News
- New York Times
- Washington Post
- Al Jazeera
- Sky News
- The Telegraph
- The Times
Outlets we query — tech
Caveats
- Tavily indexes news, not every article every outlet publishes; a story that’s paywalled or syndication-blocked might not land in their index.
- The query is the literal phrase “artificial intelligence” — it’ll catch most AI coverage but miss stories that use only “LLM” / “machine learning” / “GenAI” without the spelled-out form.
- Counts are capped at 15 per outlet per day. A publisher genuinely running 40 AI stories in a day shows as 15. This under-counts the tech side of the gap — the real ratio is wider than the chart suggests, not narrower.
- Counting stories isn’t the same as counting prominence. A front-page splash and a buried sub-section item both score 1 here. A future layer could measure salience instead of volume.