How Dowse builds listings
Dowse is a discovery layer for AI: it lists tools by the job they do, so you can start from what you need done. Each listing is built from the tool's own website, written with AI assistance, checked by a separate step, and dated.
1. We read the tool's own pages
Our crawler, DowseBot, fetches each tool's homepage and pricing page. It follows every site's robots.txt, keeps a polite pace, and never works around a site that blocks it. When a site can't be read, the tool isn't listed until it can be.
2. AI extracts facts, each with a quote
An AI model reads those pages and pulls out facts: what the tool does, who it's for, its plans and prices, features and platforms. Every fact must come with a quote from the page. Code then checks that each quote really appears on the page.
3. A separate model checks every fact
A different AI model, which didn't extract the facts, compares each one with the pages and rejects anything overstated, missing a condition (such as a plan, region or billing period) or not stated at all. Rejected facts are dropped, not softened. Prices, features and platforms never come from a model's memory.
4. Text is written from checked facts only
Summaries, pros and cons, alternatives and the reasons on ranked lists are written from the checked facts, not from the vendor's marketing copy, and then checked again. We also check that no sentence copies the vendor's wording.
5. Every listing is scored
A listing scores out of 100 on completeness, how much of it could be verified, how useful it is, whether the product is real and working, and how recently it was checked. Weak or incomplete listings aren't published, and parked, broken or copycat sites are rejected.
6. Everything is dated
Each page shows when it was last verified. Pricing is shown with the date we checked it. Prices change, so every listing links to the tool's own site.
What we don't do
- We don't claim to have tested tools hands-on.
- We don't publish ratings or reviews.
- We don't copy other directories' listings.