Signal filter for HN/news aggregators in the AI-content era
Problem
Long-time HN readers report that in the last 2-3 years, driven by AI-generated content and AI-flavored submissions, keeping up with genuinely interesting articles and discussions has become dramatically harder. The volume of low-signal posts crowds out the discussions that made aggregators valuable, and the only coping strategy people admit to is 'simply ignoring interesting stuff more and more'.
Opportunity
A curation/filtering layer for HN (and similar aggregators) that scores and ranks items by human discussion quality rather than raw points — detecting AI-generated submissions, recycling, and engagement farming, and surfacing high-signal threads. Could ship as a daily digest, an alternative front-end, or an API.
Market analysis
The pain is widely echoed, but the existing coping mechanisms are free (hand-curated digests like Hacker Newsletter, minimal alternative front-ends) and the core technical bet — reliably detecting AI-generated submissions — is a known-unsolved problem with high false-positive costs. As a product it fights both consumer free-tolerance and the reality that the platform's own moderation may absorb the fix.
Market · Information-overloaded developers and tech professionals who read HN daily; strong empathy for the problem, unproven willingness to pay for a filter layer over a free site.
Pricing · Consumer digest norms: free with optional $3-5/mo supporter tier (comparable newsletters are free or ad-supported); an API/B2B angle for other aggregators is speculative.
Pros
- + Pain is current, widely shared, and openly discussed on the platform itself.
- + MVP is cheap: HN's public API plus a scoring model, no infrastructure moat needed.
- + Alternative front-end format offers daily-use visibility that digests lack.
Cons
- − AI-text detection is unreliable; false positives destroy trust in a curation tool.
- − Free incumbents: hand-curated Hacker Newsletter and existing minimal readers.
- − Platform risk: HN algorithm/moderation changes could obsolete the layer overnight.
Existing / similar tools
Source
Hacker News (Ask HN)
The trap in this idea is treating AI detection as the feature when it should at most be one weak signal. Academic evaluations of AI-text detectors consistently show unacceptable error rates, and a curation tool that visibly mislabels a human-written post as slop loses the authority it was selling. The defensible version ranks by what is measurable — discussion depth, comment-to-point ratios, commenter quality history, domain diversity across submitters — and never claims to ‘detect AI’ at all. Notably, the strongest competitor here is behavioral: the community’s own taste expressed through front-page curation, which is exactly the signal a filter layer must out-predict to be worth opening every morning.