How much of HN is AI depends entirely on what you count: stories that mention AI, stories actually written by AI, or just whatever's sitting on the front page right now. I ran the numbers myself using Hacker News's own search API, then checked them against two independent trackers that have been watching this for months.
Short answer: Stories with "AI" literally in the title made up about 9–15% of new Hacker News submissions in my August 2026 check, depending on whether you count everything or only stories that gained traction. Broaden the definition to anything AI-related and one long-running tracker puts daily front-page coverage closer to 50%. A separate detector flags roughly 19% of front-page stories as AI-written, not just AI-about.
I test AI tools for a living, which means I spend an unreasonable amount of time on Hacker News watching the same three companies dominate the front page. So when I sat down to actually measure it instead of just complaining about it, I expected a single clean number. I didn't get one. Every method I tried — a strict keyword count, a traction-filtered count, a live front-page snapshot, and cross-referencing two other people's trackers — gave a different answer, and the gap between the lowest and highest is enormous. That gap is the real story here, not any single percentage.
What "how much of HN is AI" actually asks
There are at least three separate questions hiding inside this one phrase, and mixing them up is where most casual estimates go wrong. First: how many submitted stories mention AI at all, regardless of whether anyone reads them? Second: how many stories that actually make the front page are about AI? Third, and completely different from the first two: how many stories are themselves written by AI, as opposed to being about the topic? A post about a new coding model is AI-related. A blog post secretly drafted by ChatGPT and submitted as a personal essay is AI-generated. My guide to spotting AI writing covers the second problem in more depth, because the tells are different from just "the headline says AI."
How I checked it myself
1. Pull a baseline from HN's search API
Hacker News runs its search on Algolia, and the API is public at hn.algolia.com/api/v1/search. In my testing, querying tags=story with a date range and hitsPerPage=0 returns an exact hit count without downloading a single story, which makes it fast to run several queries in a row. For the 30 days ending August 26, 2026, HN logged 29,912 new story submissions.
2. Filter to stories with "AI" in the title
Adding restrictSearchableAttributes=title&query=AI narrows that same window to titles containing the word "AI" as its own token (Algolia's short-query matching doesn't catch it buried inside words like "OpenAI," so this undercounts slightly). That query returned 4,385 hits — 14.7% of all submissions in the period.
3. Narrow to stories that actually got traction
Most submissions never get a single upvote past the "new" queue, so counting all of them overstates how much AI content anyone actually sees. Adding points>50 to the same 30-day window drops the total to 2,006 stories, of which 188 mention AI in the title — 9.4%. That's the number I trust most for "how much of what people actually read is AI," because it filters out noise nobody saw.
4. Snapshot the live front page
The tags=front_page filter returns whatever is on HN's actual front page right now — always exactly 30 items. When I ran it while writing this, 2 of the 30 titles used "AI" as a standalone word, and a third referenced OpenAI by name. That's 7–10% depending on how strictly you count it, and it'll be a different 7–10% an hour from now — front-page snapshots move fast and shouldn't be treated as a stable trend on their own.
5. Cross-check against independent trackers
My keyword counts only catch stories that literally say "AI." They miss posts about GPT-5, Claude, LLMs, or specific model releases that never use the word. Two other trackers close that gap using different methods, and I cited both below instead of re-deriving their numbers myself.
Example prompts you can copy
If you want to run a version of this check yourself without writing raw API queries by hand, these prompts get an AI assistant to do the query-building for you:
- Build the query: "Write a URL for the Hacker News Algolia search API that counts stories with 'AI' in the title over the last 30 days, using hitsPerPage=0 to get just the count."
- Compare methods: "I have four percentages measuring 'AI content on Hacker News' from different methods — all submissions, high-point submissions, front page snapshot, and a third-party AI-text detector. Explain in plain language why these numbers would legitimately differ this much."
- Spot AI-generated text: "Here's a Hacker News comment or linked article [paste text]. What specific phrasing patterns suggest this was written or heavily edited by an AI model?"
How much of HN is AI, by measurement
| Method | Window | AI share | Source |
|---|---|---|---|
| All new submissions, "AI" in title | 30 days (Jul 27–Aug 26, 2026) | 14.7% (4,385 / 29,912) | My own count via HN Algolia API |
| Submissions with 50+ points, "AI" in title | 30 days (Jul 27–Aug 26, 2026) | 9.4% (188 / 2,006) | My own count via HN Algolia API |
| Live front page, "AI" in title | Snapshot, Aug 26, 2026 | 6.7–10% (2–3 / 30) | My own count via HN Algolia API |
| Daily top-5 stories, broadly AI-related or AI-generated | Manual review, Feb–Jun 2026 | ~40% (Feb) rising to ~50–60% (Jun) | lcamtuf’s manual tracking |
| Front-page stories flagged AI-written by a text detector | August 2026 (2,028 stories, 1,214 domains) | 19% (391 / 2,028) | Salah Adawi’s HN AI Detector |
Common mistakes to avoid
The biggest mistake, in my testing, is quoting one of these numbers without saying which method produced it — "half of HN is AI" and "under 10% of HN is AI" are both defensible depending on what you're measuring, and citing either alone without context is misleading. Second is treating a front-page snapshot as a trend; I watched the count shift within an hour just from normal ranking churn, so one screenshot proves nothing on its own. Third is confusing "about AI" with "written by AI" — Salah Adawi’s detector is measuring authorship, not topic, and the two numbers aren't interchangeable. Fourth, AI-text detectors themselves aren't perfect; false positives happen, and I've seen genuinely human writers get flagged as AI on other platforms, so treat any single detector's percentage as an estimate, not a verdict. Fifth, keyword searches like mine miss anything that doesn't literally say "AI" — a post titled "GPT-6 benchmarks leaked" won't show up in a title search for "AI" at all, which is exactly why the broader trackers above land so much higher than my strict count.
Tools that make this easier
You don't need to run raw API queries to get a feel for how AI-saturated your own reading feed is. If you're trying to figure out whether a specific piece of writing was AI-assisted, start with my guide to spotting AI writing, which covers the tells beyond just "does it mention AI." If the bigger question behind this is whether the AI industry's attention is proportional to its actual output, my AI bubble breakdown checks the spending and revenue side of that same question. And regulators are starting to force the labeling issue directly — the EU's rules on labeling AI-generated content would apply to a meaningful share of what these trackers are flagging, if it were published in the EU. For picking which AI tools are actually worth your time once you've accepted how much of the conversation is about them, my AI tool ratings and AI tool reviews hubs are where I keep the hands-on verdicts, and my free AI tools list covers ones worth trying before you pay for anything.
My take
None of these numbers are wrong, and that's the point. HN's front page is genuinely more AI-saturated than it was two years ago — even my strictest count, limited to submissions that actually got traction, put AI at nearly 1 in 10 stories, and the broader trackers that catch model names and adjacent topics land far higher. What surprised me wasn't the saturation itself; it was how much the answer moves depending on a methodology choice that most people never state out loud. If someone tells you a single percentage for "how much of HN is AI" with no window, no threshold, and no definition of AI-related versus AI-generated, treat it as a vibe, not a measurement.
Frequently Asked Questions
How much of Hacker News is literally about AI?
Using a strict title-keyword count, about 9–15% of stories in a 30-day window mention "AI" directly, depending on whether you count every submission or only ones that gained traction. Broader trackers that also catch model names and adjacent topics put daily top-story coverage closer to 50%.
Is Hacker News content actually written by AI?
Some of it, yes. Salah Adawi's HN AI Detector, which runs the Pangram text-detection model against front-page stories, flagged 19% of August 2026's front-page content as AI-generated writing — a separate question from whether the story is about AI as a topic.
Why do different "how much of HN is AI" estimates disagree so much?
Because they're measuring different things: all submissions versus only popular ones, a live snapshot versus a monthly average, keyword matching versus AI-authorship detection. Each method is internally consistent; they just aren't answering the same question.
Is Hacker News still worth reading if this much of it is AI?
Yes, based on my own daily use — even the highest AI-saturation estimates still leave roughly half the front page covering everything else, from security research to hardware builds to pure curiosity posts, and the site's ranking system hasn't changed.