We only estimate visits from buyer-intent prompts, where someone asks ChatGPT to recommend a product or service (like “what’s the best keyword tool?”), and only across the commercial categories we track. We exclude branded searches (people already asking for Hal.science by name) and informational or navigational prompts. So this is Hal.science’s share of non-branded ChatGPT recommendations in the categories we track, not its total ChatGPT traffic.
Each topic written the way a person would ask it. Estimated visits = this site's share of that prompt's demand.
ChatGPT named Hal.science here in only some runs (below our 2-of-5 confidence bar), so these earn no estimated traffic, but they show where it occasionally surfaces this site.
One of the 5 runs for What are the best websites to read research papers for free?, verbatim.
How often ChatGPT places this site at each position across its prompt topics.
Sites that come up for the same prompts as Hal.science, ranked by how many topics they share.
ChatGPT doesn't publish prompt data. We measure real demand for each commercial topic from Google search volume, scale it to ChatGPT's user base (about 1 ChatGPT user for every 8 Google searchers), and run each topic through ChatGPT 5 times to record which sites it recommends and in what order. A click-through-by-rank curve splits each topic's demand across the recommended sites. Every figure is a modeled estimate, not measured analytics. Full methodology →