DataBy James Mac

Six months of AI training listings: what grew, and what pay did

From May 14 to November 11, 2026, we compared saved listings from the same five consistently observed sources about every two weeks. Active listing families rose about 15%, with a peak in mid-September. Engineering, generalist and doctor categories grew most among the named role groups, while translation families fell. These are listings, including potential talent pools, rather than confirmed current vacancies. Listed pay for credential and language roles stayed broadly steady.

SignalGigs may earn a referral commission when you apply through links on this site.

Active listing families rose about 15%, then eased from a September peak

On the five consistently observed sources, 407 AI training listing families were active in the May 14 snapshot and 467 on November 11. Comparing the first two checks with the last two also shows a rise of about 15%. The count peaked at 553 on September 16 and has eased since. One large source added more families than the whole net rise; three of the other four ended with fewer.

We count a role family once even when it is posted in many languages or cities: a role family posted in 30 languages is one family here, not 30. That matters. Counting raw postings, translation work looked like it grew, from about 129 postings to 156, but the number of distinct translation families fell from about 74 to 60. The growth came from new families posted in many languages: most of November's translation postings belong to families that were not active in May, while many older single-language families stopped appearing.

Line chart of open AI training jobs on five steadily tracked sources from May 14, 2026 to Nov 11, 2026: 407 at the start, a peak of 553 on Sep 16, and 467 at the end.
Open jobs on the five sources tracked at every check, each job counted once across its language versions. Historical active saved listings; counts can include talent communities. Matching core titles are grouped, not established as identical contracts.
Show the numbers
Open jobs and postings at each check
CheckOpen jobsOpen postings
May 14407852
May 28425869
Jun 10477960
Jun 23475961
Jul 6443888
Jul 224521015
Aug 5427983
Aug 194511043
Aug 315331107
Sep 165531104
Sep 29490980
Oct 13469931
Oct 27486978
Nov 11467904

Engineering, generalist and doctor categories led the named groups

Engineering and skilled-trade jobs grew the most among named, title-classified categories, from about 29 to 42 active per snapshot. Generalist AI-trainer jobs went from about 5 to 16, doctor jobs from about 10 to 18, voice and audio recording jobs from 19 to 26, and software jobs from about 15 to 21.

Translation and linguistics jobs fell from about 74 to 60, and so did business, policy and security expert jobs (about 34 to 23) and data-labeling jobs (about 12 to 9). Translation still has more open jobs than any other named role type on these sources.

Dumbbell chart of open jobs by role type on five steadily tracked sources, average of the first two checks against the last two: Engineers & skilled trades 29 to 42; Generalist AI trainers 5 to 16; Doctors & physicians 10 to 18; Voice & audio recording 19 to 26; Software & coding 15 to 21; Physics, chemistry & biology 17 to 21; Finance & accounting 15 to 17; Lawyers & legal 13 to 15; Health & clinical experts 9 to 10; Data labeling & annotation 12 to 9; Business, policy & security experts 34 to 23; Translators & linguists 74 to 60.
Open jobs by role type: the average of the May 14 and May 28 checks against the average of the October 27 and November 11 checks. Historical active saved listings; counts can include talent communities. Matching core titles are grouped, not established as identical contracts.
Show the numbers
Average open jobs, May vs now
Role typeMayNow
Engineers & skilled trades28.541.5
Generalist AI trainers4.516
Doctors & physicians9.517.5
Voice & audio recording1926
Software & coding14.521
Physics, chemistry & biology16.520.5
Finance & accounting1517
Lawyers & legal1314.5
Health & clinical experts910
Data labeling & annotation11.58.5
Business, policy & security experts33.522.5
Translators & linguists73.560

Pay held steady for credential and language roles

We grouped jobs by what the title asks for: a credential or professional expertise (doctors, lawyers, engineers, finance, software, science) or a language skill. From May to October, the median top listed rate for credential roles stayed between $100 and $120 an hour, and for language-skill roles between $22.25 and $32.

Roles that ask for no specific credential are left out of this chart. On these sources, the rated titles in that group include professional evaluator roles and voice actors, so its median jumped as those batches opened and closed. The chart stops on October 27: on November 11 our scraper started reading rates written into the posting text, which added rates for many language jobs at once.

Two small line charts of the median top listed hourly rate by qualification, May 14, 2026 to Oct 27, 2026: credential or professional expertise between $100 and $120; language skill between $22.25 and $32.
Median top listed hourly rate by the qualification the job title asks for, with the middle half of rates shaded. Historical active saved listings; counts can include talent communities. Matching core titles are grouped, not established as identical contracts.
Show the numbers
Median top listed hourly rate: average of the first two checks and of the last two, and the lowest and highest check
QualificationMayOctoberRange
Credential or professional expertise$113.75$110$100 to $120
Language skill$26.65$26.50$22.25 to $32

Most listings keep their rate

To see whether pay itself moves, we followed the same listings from one check to the next. At each check, between 0.4% and 7.9% of the listings open at both checks changed their listed rate. Between October 27 and November 11 it was 5 of 320, all of them increases. Changing job mix can move the median even when most continuing listings keep their rate; this comparison does not decompose the causes of median changes.

If you are looking for work, check the current source’s rate, availability and eligibility. Most continuing listings in this historical sample kept their rate; that is not a prediction for an individual posting.

Column chart of the share of listings that changed their listed rate between checks about two weeks apart, May 28, 2026 to Nov 11, 2026: between 0.4% and 7.9% at each check; 5 of 320 at the last check.
Share of listings open at two consecutive checks, still pointing at the same posting, whose top listed hourly rate changed. Historical active saved listings; counts can include talent communities. Matching core titles are grouped, not established as identical contracts.
Show the numbers
Listings open at both checks, and how many changed their rate
CheckOpen at bothChangedUpDown
May 281931266
Jun 10172303
Jun 23275431
Jul 62231459
Jul 22241101
Aug 524510100
Aug 1931725916
Aug 31425431
Sep 16352303
Sep 293861019
Oct 13360624
Oct 27300514
Nov 11320550

Frequently asked questions

Did the number of AI training jobs grow in 2026?

On the five sources we tracked at every check, active listing families rose about 15% from May to November 2026: 407 on May 14 and 467 on November 11, with a peak of 553 on September 16. One large source accounted for most of the rise. Counts can include talent communities rather than active openings. Sources added during the period, or whose scraper broke at any point, are left out, because they would distort the trend.

Did AI training pay go up?

Rates for credential and language roles stayed broadly steady in this sample. From May to October, the median top listed rate for credential roles stayed between $100 and $120 an hour, and for language-skill roles between $22.25 and $32. Most listings kept the same rate from one check to the next: 5 of 320 changed between October 27 and November 11.

Which AI training jobs grew the most?

Engineering and skilled-trade jobs, generalist AI-trainer jobs and doctor jobs grew the most among named, title-classified role groups between May and November 2026. Translation, business-expert and data-labeling jobs fell.

How we measured this

Sources: the five job sources whose scrapers worked at every check from May 14 to November 11, 2026. Sources we added later, or whose scraper broke at any point, are left out, so a scraper problem can't look like a trend.

Active listing families: a family counts at a check when the latest successful saved source run observed at least one of its listings. This is historical saved-record evidence, not a fresh vacancy check; talent communities may be included.

Families and postings: a posting is one listing text. A family groups one source’s core title after language, place and project names are removed. This deduplicates title variants, but does not establish that they describe identical work or contracts.

Pay: the top of each posting's listed hourly range in US dollars. Qualification comes from the job title. Pay is what each posting states; it is not a promise of work or income.

The named-category growth ranking excludes unclassified “other” families, which rose from an average 122.5 to 153.5 between the first and last two checks. Qualification and role groups are inferred from titles, not independently confirmed.

Related