DEV Community

Louis Li
Louis Li

Posted on Originally published at neonstack.net

I submitted 18 sites to 235 AI directories. The data is worse than I expected.

I spent a night submitting 18 sites to every AI and startup directory I could find, then checked what came back. 2,700 attempts, 81 acceptances, 8 directories that ever said yes.

I published the whole table, including the directories that ignored me, because the "submit to 200 directories" advice is repeated everywhere and almost nobody shows the outcome column.

What I actually did

  • Probed 235 directories in headless Chrome, not with a plain HTTP fetch. Half of these sites render their form after JavaScript runs; a static GET returns a shell with a search box and no form.
  • Recorded whether each one had a real form: two or more visible fillable fields, at least one matching URL, name, email or description, no login redirect, no CAPTCHA.
  • Submitted from 18 different sites, 2,700 attempts total, and counted an attempt as accepted only when the response contained a receipt: a thank-you message, a pending-review notice, or a verification email prompt. HTTP 200 on its own counts as unknown, never as success.

That last rule matters more than it sounds. A client-side form that swallows your POST still returns 200.

The numbers

Measure Value
Directories probed 235
Accept a fully automated submission 46 (19.6%)
Single-input or field-less pages 146 (62.1%)
Require an account before the form 34
Behind a CAPTCHA 4
Submission attempts 2,700
Accepted by the directory 81 (3.0%)
Directories that ever accepted anything 8 (3.4%)

3.4% of the pool absorbed 100% of the acceptances. The rest took the form, returned 200, and produced nothing.

Three failure modes, in order of size

1. One-field pages (62.1% of the pool). A single input, usually a newsletter box. The real form is hidden behind a "Submit your tool" button, or it does not exist because the site aggregates products it found itself.

2. Accounts (34 directories). These redirect to a signup page first. Most now use magic links or OAuth, so scripted registration is not just hard, it is impossible: the login needs a click in an inbox, not a password.

3. Receipts that do not exist. Plenty of SPA directories accept the POST and swap the DOM without navigating. No success URL, no status code a script can read. From the outside, a stored submission and a dropped one look identical.

What the accepting directories had in common

Four of them accounted for most acceptances: they are small, they review manually, and their submission form is server-rendered. Those are also the only ones whose listing could plausibly rank for anything, because a curated page with 300 entries can beat an uncurated page with 30,000.

The large aggregators were the least useful: an account wall, a paid tier, and a page that ranks for its own brand and nothing else.

The uncomfortable part

81 acceptances is not 81 backlinks. Every submission was still in review when I checked, and none had gone live. Some fraction of those promises will simply never be honoured, and there is no way to tell in advance which.

So the honest summary of a night of work: a spreadsheet, a scraper I can reuse, and a very clear picture of which 8 directories are worth a second visit.

If you want the data

The full table, with field counts, login walls, CAPTCHA flags and the rejection reason for every row, is published under CC BY 4.0:

If you add your own results to the acceptance column, the sample stops being one person's.

If you have run a similar sweep, I would like to see your numbers. Mine are probably not representative: 18 sites, one night, one operator.

Top comments (3)

Collapse
 
li_yi_dcec96bf2647dec1b7e profile image
Mindutil •

The rule "HTTP 200 counts as unknown, never as success" is the real finding here, and it cuts in both directions. I run a directory of ~4,300 AI tool listings, and the mirror image of your 3% acceptance rate is that only 614 of my rows carry a vendor claim (14.2%) - so roughly 86% of what a directory publishes never went through a submission path at all, it was scraped in. Your 81 receipts and my 4,300 rows are one population seen from opposite ends. Two rules I took from your outcome column: count an attempt only when the site returns a receipt, and count a listing as evidence only when someone accountable touched it. The number that would actually mean something on my side is outbound clicks - mine sums to 4, which is exactly why my category leaderboards need to be labelled as a borrowed popularity counter instead of as readership.

Collapse
 
shashankrajurkar profile image
Shashank Rajurkar •

I like the distinction between acceptance and an actual live listing. I'd take the measurement one step further: which directories generate relevant visits, qualified enquiries or trials? For early-stage founders, the real value of a distribution channel is what happens after the listing, not how many submission receipts it produces.

Collapse
 
alexcodebytes profile image
Oleksandr •

"http 200 on its own counts as unknown, never as success" is the most important sentence here. i've built scrapers that hit a client-side form, got a 200 ok, and thought everything was fine, only to realize the js had silently swallowed the payload. treating a receipt as the only proof of success is such a solid methodology. and honestly, 3.4% is way better than i expected for these "submit to 200 directories" schemes lmao.