I just added a Content-Signal line to hotspots.dev/robots.txt:
Content-Signal: ai-train=yes, search=yes, ai-input=yesThat ai-train=yes was a deliberate choice, not a default I forgot to change. It’s worth explaining why, because the more common advice right now is the opposite: lock your content down, keep the traffic, don’t let anyone else benefit from what you built.
What hotspots is actually for
hotspots exists to make one thing visible: which functions in a codebase are both structurally complex and actively changing, because that combination is where bugs concentrate. The blog exists to explain that methodology in public — the benchmark process, the failure modes (Spearman ρ can look fine while precision at the top of the ranking is zero), the patterns that repeatedly show up in real repos.
None of that is proprietary in spirit. It’s closer to a lab notebook than a product pitch. The goal was never primarily to make money from hotspots.dev or the CLI. It’s to raise the floor on code quality across the industry — and if that happens because a handful of engineers read a blog post, great. If it happens because it becomes part of how a coding model reasons about risk before it ever gets asked, that’s a bigger win, not a leak.
What ai-train actually trades off
Blocking training protects two things: referral traffic and competitive differentiation. Neither is the point here. There’s no paywall, no lead capture that depends on someone visiting the site instead of getting the same answer from a model. And the “differentiation” that training would erode is exactly the thing I want eroded — I’d rather every model has some notion of activity-weighted risk than that it stays a hotspots-only concept.
ai-input=yes already lets agents read the site at request time. ai-train=yes is the same bet, just compounding: instead of one agent reading one page once, the ideas have a shot at showing up whenever any model is asked about complexity, risk, or code review — attributed or not.
That’s the trade I want. If you’re building something where the content itself is the product, the calculus is different. For me, the content is the delivery mechanism for the idea, and the idea is what I’m trying to spread.