What's on the site
Three things:
- Tool pages. What a tool does, its plans and prices, pros and cons, and an AI Score from 1 to 10. Entries that have been through a refresh also carry the date their pricing was last checked.
- Comparison pages. Same-category "X vs Y" pages built from the tool data, so a search for two products lands on a page that puts them side by side.
- The blog. News, opinion, guides, comparisons, and step-by-step tutorials written against vendor documentation.
The AI Score is an editorial judgment, not a measurement. It weighs capability, price-for-what-you-get, usability, and how fast the product is shipping, all read off public information. Use it to sort, then read the cons and the pricing before you decide anything.
How a page here gets made
The pipeline runs twice a day on a small server and does this:
- A research script asks Grok, with live web and X search, what actually happened in AI in the last day and which tools the directory is missing. It gets handed the full list of what we already cover so it stops proposing duplicates.
- A drafting script sends the topic, the site's writing rules, and the patterns of recently published posts to Grok, which writes the draft.
- The draft goes through a quality gate. That gate is a program, not a person, and it rejects drafts on named defects: invented hands-on testing, links to pages that don't exist, near-duplicate coverage, banned stock phrasing, missing or oversized metadata. The full list is on the How we verify page.
- What passes gets published, the site rebuilds, and the change is committed to a git repository.
- A human reads what shipped.
Step 5 is the one that matters most. The gate only catches failures somebody wrote a rule for. When something bad gets through anyway, the fix is a new rule in the gate, not a promise to be more careful next time. The rules have been tightened repeatedly, and the reasons are recorded in the repository's history.
What the writing rules forbid
- No first-person testing. Nobody here spent a weekend with the tool. "I tested", "in my testing", "we benchmarked both" and their variants are rejected by the gate before publication.
- No invented numbers. Latency, cost per run, accuracy, benchmark scores. Either the number comes from a named source in the same paragraph or it doesn't appear.
- No invented observations. No "the refactor left an extra re-render". That would be fiction dressed as experience.
- Opinion is allowed and it is labeled. When you read "My read:" or "The honest take:", that is a judgment call, and the rest of the piece is meant to be checkable fact.
- Sources for anything non-obvious. Prices, dates, quotes, and benchmark figures should trace back to something you can open.
The instruction the writer works from ends the rule list this way: if you can't write a section without breaking these, cut the section.
Who runs it
One person, working alone, publishing under the name The AI Dude. That name is on every byline on the site. There is no editorial team, no roster of contributors, and no invented reviewer personas.
The operator stays pseudonymous and does not put a résumé on this page. So judge the work by whether it cites its sources, gets the version numbers right, and admits what it doesn't know, rather than by a bio you'd have no way to check either.
Contact goes to @theaidood on X, which is read by that same person, or through the contact page.
How the site makes money
Some links out to tools are affiliate links. If you sign up through one, the site may earn a commission, and it costs you nothing extra. Those links carry rel="sponsored" in the markup, so they are machine-readable as paid links, not just disclosed in small print. One program is live as of August 2026 (ElevenLabs); the other tools link straight to the vendor and are not marked sponsored.
Coverage and scores are not for sale. There is no paid placement, no sponsored post, and no review written in exchange for anything. An affiliate link can only be added by the operator running a command by hand; the pipeline cannot create one, and the model that writes tool entries never sees that field at all. A vendor paying nothing and a vendor paying a commission get scored the same way, because the score is computed from the same public information either way.
The site runs no third-party ad network and no analytics beyond its own server logs. Details are in the privacy policy.
Corrections
Prices change, models get renamed, and links rot. When something here is wrong, it gets fixed and the page is rebuilt. If the fix changes what a piece actually claimed, the correction is stated on the page rather than swapped in quietly.
The fastest way to report an error is a message to @theaidood. Tell us the page and what's wrong with it. For the checks a piece has to pass before it ships, see How we verify.