I run a small AI tools directory as a side project. Since late 2025 I've reviewed somewhere north of a thousand AI products, and listed only about a hundred of them. This post is the checklist I wish I'd had on day one — it might be useful whether you're picking tools for your team or building any kind of curated list.
Why "list everything" fails
Most AI directories auto-scrape: every landing page with "AI" in the title gets a row in the database. That maximizes page count for SEO, but it pushes the actual filtering work onto the visitor — which is the problem a directory is supposed to solve in the first place.
So the only real value a curated list can add is saying no. Here's what my "no" looks like.
The checklist
1. Task first, demo second
The first question is never "is this impressive?" but "what task does someone finish with this?" A surprising number of AI products demo beautifully and map to no recurring task. If I can't write one sentence in the form "use this when you need to ___", it doesn't go in.
2. The free tier has to be real
I check: can you complete the core task at least a few times without paying? "Free trial that dies in 7 days" and "free tier that watermarks everything into uselessness" both count as bait. I list paid tools too — but the pricing has to be stated honestly on the listing, not discovered at checkout.
3. Two honest cons, minimum
This is my favorite filter, and the most work. For every tool I force myself to write at least two genuine downsides (slow output, weak non-English support, aggressive upsells, no export, whatever is actually true). If I can't — it means I haven't used it enough to list it. The cons section is also the part readers trust most, which says something about directories in general.
4. Is anyone home?
Changelog or release notes from the last ~3 months, a responsive support channel, a team page that isn't stock photos. AI tools die fast and quietly; a directory full of dead links is worse than no directory.
5. Open-source alternative check
For each category I try to also know the self-hostable option. Sometimes the right recommendation for a privacy-sensitive task is "don't use a SaaS at all" — a directory that can't say that is just an ad wall. (This is why I ended up maintaining a separate open-source section with 80+ projects.)
What gets rejected most
In practice, the three most common rejection reasons have been:
- Wrapper with no margin of value — a thin UI over a foundation model API, priced higher than the API itself.
- Landing page only — waitlist products. Nothing against them, but a directory entry for something you can't use helps nobody.
- Identity churn — product renamed/repositioned twice in six months. Usually predicts a shutdown.
What I'd tell anyone building a curated anything
- Decide your rejection reasons before your acceptance reasons.
- Write the cons first. If the cons are hard to write, you don't know the product yet.
- Date everything. Curation rots; visible "last reviewed" dates keep you honest.
Disclosure: I'm the maintainer of SoFarBot, the AI tools directory described above — free to browse, no signup, in English and Chinese. The checklist works without it, but if you want to see it applied, that's where it lives. Happy to answer questions about the curation process in the comments.
Top comments (0)