Report · measured 1 September 2026

One board advertises 5,626 AI jobs.
I counted 528.

On 1 September 2026 the public Alignerr job board returned 5,626 listings across 47 pages. Those listings carry 516 distinct titles and 528 distinct title-and-pay combinations, and every one of the 5,626 gives the same location, Remote. I counted it again 31 hours later and every one of those figures came back identical.

Before you read on. This site earns a referral commission from two of the platforms in the second table below. Alignerr, the board counted here, is not one of them and pays me nothing. The same count is applied to the platforms that do pay me, on the same terms, in the same table. If you find a figure on this page that does not match what the board publishes, tell me and I will correct it with the correction noted on the page. Full terms are on the disclosure page.

This matters if you are looking for work. A board showing thousands of openings reads as a busy market with room for you. A board showing 528 openings reads as a queue. The difference between those two readings is not a matter of opinion, and you do not need my word for it. The count below comes from one public endpoint the board serves to anyone, and the script that produced it is at the bottom of this page.

What I counted, and how

I read every page of https://www.alignerr.com/api/jobs?limit=120&offset=N until it stopped returning rows. That is 47 requests at 120 listings each. No login, no detail pages, no scraping of anything the board does not hand out on request.

Then I grouped the listings on title plus the pay wording. Country is deliberately left out of that key, because country is the one thing that can legitimately differ between two real postings of the same role. Pay is deliberately left in, because two listings advertising different rates are two different offers and deserve to be counted twice. This is the same key this site uses to decide which duplicate URL to show a search engine, and it was in the code before I wrote this report.

I counted it twice

One reading of a job board is an anecdote. A board can have an odd day, and a ratio taken once says nothing about whether it holds. So the whole count was run again 31 hours later, on 2 September 2026, and the headline figures came back unchanged. Same 5,626 listings, same 516 titles, same 528 title-and-pay combinations, same 327 copies of the largest one, and the same single location value on every row.

One thing did move. The number of distinct pay bands went from 67 to 66, which is the detail that matters most, because it is how I know the second run was a real read and not a cached copy of the first.

The ten most repeated listings

One title, Software Engineer (AI Training), accounts for 327 of the 5,626 listings on its own, all at $20-100/hr. Across the whole board there are 67 distinct pay bands in total.

The location field settles it

Every one of the 5,626 listings reports exactly one location value, and that value is Remote. There is no second value anywhere in the set. A remote role does not have separate openings in twenty cities. When the same remote title appears at the same pay several hundred times, those are not several hundred jobs you can apply to. They are one opening presented several hundred ways.

The same test, run on everyone here

It would be easy to point this at one board and stop. So here is the identical collapse applied to every platform whose feed this site reads, counted from the roles open right now. Two of these platforms pay me a referral commission and they are in the table on the same terms as the rest.

One honest caveat about that table. Alignerr appears in it at close to 1.00x, and that is not a contradiction. This site collapses Alignerr's repeats before storing them, so the database holds the already counted set. The 10.7x figure in this report was measured at the source, which is the only place it is still visible. Every other platform in the table is stored exactly as its feed publishes it.

What to do with this

Stop counting listings. A listing count tells you how a board is built, not how much work is available. Count distinct roles at distinct rates, and treat the same title at the same pay as one thing however many times it appears. That is the only number that changes what you should do next.

It also means the market is smaller and better paid than the headline counts suggest. Fewer real openings, and the ones worth having are the ones that publish a rate you can check. That is what the rest of this site is for.

Run it yourself

The script that produced every number above is served from this site at /board-census.py. Save it and run python board-census.py. It needs nothing but a standard Python install, reads the one public endpoint named above, writes the raw rows next to its own summary, and prints the counts. If your run disagrees with mine, your run is the newer measurement and I would like to hear about it at contact@aipaylist.com.

Counts move. The board could look different tomorrow, and this page states the date on which it did not. The live figures in the second table are recomputed every hour and were last built on 2026-09-09. Everything here is CC BY 4.0, so quote it freely with a link.

All reports · How pay is read · Which links pay me