We found 10 mistakes in 100 AI tool listings

Jason T. · 2026-10-05

On its first full night, 2026-10-03, New Site Radar processed 4,000 newly registered domains with AI-related names and listed 396 of them. The next morning we hand-checked a random sample: 100 listed sites and 100 left-out sites. 10 of the 100 listed sites were wrong, so about 90% of what we listed was right. We wanted at least 85%. The mistakes were not random, and they are worth knowing if you pick new AI tools for a directory or a newsletter.

Two warning patterns

Most of the wrong listings were one of two kinds:

"AI trading platform" sites built from the same template. Random made-up names, registered within a few days of each other, the same headline in several languages.
Sites using another company's name in their title.

What we changed the same day

Trading-platform templates are now treated as unsafe, and a title that starts with someone else's brand is rejected. Re-running the new rules over everything already listed removed 26 sites.

Why most domains were left out

Every night New Site Radar checks newly registered domains with AI-related names and decides which are real, working AI products for the free Just launched list. The checks run in order: does the domain respond, is there a real homepage, is it an AI product, and is it safe to list. An automated reader looks at the homepage for the AI-product question; the rules for parked pages, redirects and brand names are plain code. On the first night, the most common reasons a domain was left out were:

995 did not respond at all.
801 showed an empty page.
230 blocked automated visits in robots.txt.

Others were parked or for-sale pages, "coming soon" pages, default hosting pages, website-builder templates, redirects to another site, or sites that are not AI tools. A newly registered name and a working product are different things.

The left-out half of the sample

Among the 100 left-out sites we found no clear case of a real product wrongly rejected; most were portfolios, casinos, tutorials or empty pages. Two were unclear from the stored reason alone and will be opened by hand in the next check.

What the next check will tell us

We will repeat the sample check weekly with the same method and report whether the new rules wrongly reject any real product. The full method, including what we will not list, is on the methodology page. If you run a new AI tool that is missing from the list, check it with the new site checker.

Free RSS: rising sample, 20-site monthly sample; just launched, AI tool sites registered in last 90 days; no traffic numbers.