The Expert Scarcity Index for AI Training

Method and definitions

The public job boards of the AI training platforms this site tracks, read every day, and the apply clicks people make on AITraining.jobs. Fields and languages are read from role titles. Every figure is computed by one database query over the window shown, the same query for the live report and for every edition. Each figure states its sample beneath it.

Scarcity index

The geometric mean of the applicant shortfall, the slow-closing ratio and the pay premium, each held between 0.2 and 5 so one extreme reading cannot decide it alone. 1.00 is the average AI training role; above 1, the group is harder to staff.

Applicant shortfall

The apply clicks a group's roles would have drawn at the market's rate per open role, divided by the clicks they drew in the window: one over the demand index. Above 1, fewer applicants per opening than the market.

Slow closing

For every role listed in the last 26 weeks (from 3 Aug 2026 at the earliest; back catalogues left out), the chance it comes down at each age is read from its own platform's Kaplan-Meier curve; summed over the group's roles, that gives the removals expected at its platforms' pace. Slow closing is expected removals over removals seen. Above 1, the group's roles stay listed longer than other roles of the same age on the same platforms.

Pay premium

The median advertised hourly rate of the group's open roles over the median of every open role stating one. Needs at least 8 roles stating a rate.

Median days listed

The Kaplan-Meier median from the listing date to the last sighting, as in the listing-lifetime report. Published beside the index; not part of it, because it depends first on how each platform posts.

Ranked

A group is ranked when its roles would have drawn at least 30 clicks at the market's rate, at least 10 removals were expected, at least 8 of its open roles state an hourly rate and at least 10 are open. Other groups are shown and marked as small samples.

Expert field and language

Named in a role's title, matched as whole words with the vocabularies of the expert fields and languages reports. A role naming two counts for both.

Open role

A role listed by a given day and still in its platform's feed that day.

Kind of work

Each role is classified from its title and description into one of the twelve kinds of AI training work this site tracks. Every kind links to its own page.

Window

Live: the last 28 complete days, in UTC, with listing lifetimes over every role listed in the 26 weeks to the same end (from 3 Aug 2026 at the earliest). Edition: one calendar month, with lifetimes over every role listed in the 26 weeks to its end, followed to its end.

What this data cannot tell you

• Applicant interest is measured among the people who use AITraining.jobs, not across the market, and what is shown first gets clicked more: each visitor sees the roles open in their country first, which lowers the clicks on roles limited to a few countries, as many language roles are.
• A listing coming down is not a hire: a role can close because it filled, because the project ended or because the platform re-posts it. Reading it against the same platforms removes each platform's posting habits, not the reasons a particular role closed.
• Pay is what postings advertise, not what workers earn, and part of a gap between groups is where they hire: many platforms set lower rates in lower-cost countries.
• Generic titles hide the field or language of many roles, so a group's counts are a floor; a role can count for a field, a language and a kind of work at once.
• Platforms that recruit without a public job board are not counted.

How the job feeds are read and roles are classified: site methodology. How fresh, complete and consistent the data is, checked every day: data quality and coverage.

Editions

The live report covers the last 28 complete days and changes every day. Each complete month is also published as an edition, frozen two days after the month ends, so a figure you cite from an edition never moves.

• Live dataLast 28 complete days, recomputed daily
• September 2026 edition (this page)Frozen 11 Oct 2026
• August 2026 editionFrozen 11 Oct 2026

Cite and reuse

Free to quote, chart and republish under CC BY 4.0 with a link to this page. This edition is frozen, so its figures will read the same later.

@misc{aitrainingjobs_expert_scarcity_index_2026_09,
  author = {{AITraining.jobs}},
  title = {The Expert Scarcity Index for AI Training, September 2026 edition},
  year = {2026},
  howpublished = {\url{https://www.aitraining.run/research/expert-scarcity-index/2026-09}},
}

More research

• Languages in Demand for AI TrainingWhich languages are AI labs hiring for, what do they pay and how much competition is there?Read the report
• Expert Fields in Demand for AI TrainingWhich professions and disciplines are AI labs hiring to train models, and what do they pay?Read the report
• What AI Training Roles RequireWhat do AI training roles ask for: degrees, credentials, experience, hours, tests?Read the report
• AI Training Job TitlesWhat is AI training work called, and does the title say anything about the pay?Read the report
• Programming Languages in AI TrainingWhich programming languages are AI labs hiring coders to train models in, and what does each pay?Read the report
• What AI Trainers DoWhat tasks does AI training work involve: writing prompts, rating answers, red-teaming, coding?Read the report
• Real-World Data Collection WorkWho is hiring people to collect real-world data for robots and physical AI, where, and at what pay?Read the report
• The State of AI Training PayWhat does AI training work pay, by kind of work and by platform?Read the report

All 78 research reports