best-of-ai indexes 1368 tools across 143 categories, and 51 of those categories hold exactly three

The index adds up exactly: 143 categories carrying 1368 tools

The visible table of contents carries 143 category entries with a number in parentheses on each, and those numbers sum to exactly 1368, matching the 143 categories and 1368+ tools claimed further up the page. That internal consistency is the most checkable thing in the repository, and it is worth stating plainly because most directories of this kind do not survive the same arithmetic. One detail sits underneath it. The prose claims 1368+ tools, with a plus sign, while the index totals 1368 exactly. The extra tools the plus sign allows for are not in the index, so the two numbers agree on a floor and disagree on whether anything sits above it. No per-tool detail appears in the visible portion, only category totals and anchor links.

Fifty-one of the 143 categories hold exactly three tools each

Read the counts as a distribution rather than a list and the shape becomes clear. Fifty-one categories sit at exactly three entries, and one more, Others, sits at two. That is 52 of 143 categories holding 155 of the 1368 tools, so a little over a third of the taxonomy is a shelf holding three things. The names on those shelves span Architecture, Biotech, Cloud Management, Compliance, Dubbing, Energy, Events, Fitness, Geospatial, Identity, IoT, Localization, Manufacturing, Medical Imaging, Mental Health, Nonprofit, Nutrition, Parenting, Patent Tools, Pets, Privacy, Robotics, Sentiment Analysis, Space Tech, Sports Analytics, Storytelling, Synthetic Data, Tax and Accounting, Telemedicine, Wellness and Web3. Against a stated promise that every tool is reviewed and ranked rather than scraped, a three-item shelf is a curation decision with nothing left to rank.

DevTools holds 115 tools, more than Writing Assistants and Video Editor combined

The head of the distribution is lopsided in a specific way. DevTools carries 115 entries, which is more than Writing Assistants at 43 and Video Editor at 20 added together, and it alone accounts for a little over eight percent of everything indexed. The ten largest categories, ending with Healthcare at 23, hold 414 tools between them, so the top seven percent of the taxonomy carries close to a third of the inventory. After DevTools the curve drops sharply to Code Assistant at 47 and Writing Assistants at 43, and then steps down through AI Agents at 40, Video Generator at 36 and Image Generation at 34. Read as an editorial signal, the ranking of categories by size tracks how easy a category is to fill from press releases rather than how many tools genuinely exist in it.

The repository is the Hugo build, not a list of files

The tree explains what kind of project this is. The root holds hugo.toml alongside content/, data/, layouts/, scripts/ and static/, and the recorded primary language is HTML. So what lives on GitHub is the build for the site at bestofai.io, not the list itself in a form anyone can consume without a site generator. The README points outward for the useful part: bestofai.io is where search, filters and detailed reviews live, and the category links in the contents all resolve to paths under bestofai.io/categories/ rather than to anchors in the file. The one artifact the repository offers on its own is the flat anchor index, which is long enough that the visible portion ends inside the AI for Professionals entry.

Submissions arrive as GitHub issues, and no release records what changed

The Submit a Tool link in the header does not open a pull request. It points at an issue creation URL carrying a submit-tool.md template and a title of Submit Tool, so the intake path is an issue form on a repository whose root also holds CONTRIBUTING.md. That makes the review trail a conversation rather than a diff. Alongside it, the repository has no GitHub releases at all, and the last push is dated 2026-10-02. Nothing visible in the repository records a version, and the update promise is a single phrase, that new tools are added regularly as the AI landscape evolves, with no cadence, no changelog and no dated entry to compare it against.

The LICENSE link in the README has no file behind it at the root

The header carries a link to LICENSE, and a curated list is exactly the kind of content people republish. The top-level entries are .github/, .gitignore, CONTRIBUTING.md, README.md and the site directories, so no LICENSE file appears among them, and the repository states no license terms of its own. The site and the list are therefore in an awkward position: the content is designed to be copied into other people's stacks, and the terms for doing that are not visible on the repository. This is the one question to settle before quoting the category taxonomy or any of the 1368 entries anywhere, and it is a question the site owner can answer faster than any reader can infer it.

Five differentiators are claimed, and only one has a mechanism attached

The What is this section offers five reasons the list is different: quality over quantity with every tool reviewed and ranked rather than scraped, a practical focus on real-world usefulness rather than hype, broad coverage, always updated, and stacks curated by profession. Only the last has something a reader can go and look at, a professionals page on the site. The other four are process claims, and the repository provides no ranking method, no review rubric, no record of when a tool was added or removed, and no stated rule for what earns a place. The claim most worth testing is the first one, because the category counts show how little room for judgement a three-item shelf leaves.

The visible text ends inside the contents, before the first tool name

The contents list opens with a Leaderboard entry and closes, in the visible portion, partway through the AI for Professionals section, and everything after that point is outside what is shown. So a reader working from the repository alone sees 143 category names, 143 counts, and not one tool. That is a reasonable design for a project whose product is the website, but it means the repository is not an alternative front end for the directory. It also explains why the index, and not the tool list, is the thing worth auditing here: the counts are the only claim in the repository that can be checked without leaving it.

Editorial conclusion

Treat best-of-ai as a starting index rather than a ranking. The numbers inside the README add up exactly, which is more than most directories of this kind offer, and the taxonomy is broad enough to be useful for browsing. What it does not offer is any way to check the editorial claim: the visible text gives no ranking rubric, no removal policy, and no update cadence beyond new tools added regularly, and a third of the categories are shelves of three. If you plan to reuse the content, settle the licensing question first, because the README links a LICENSE file that does not appear among the top-level entries. If you plan to contribute, the route is an issue form rather than a pull request, so expect review to happen around the site build rather than in a diff you can read here.

Frequently asked questions

Official sources

1. best-of-ai/best-of-ai on GitHub
2. Issues
3. Project website
4. README