Skip to content
HumanSourcer

AI-training market data

6920 active listings across 35 AI-training networks, rebuilt on every daily deploy. Every rate figure here is computed over 791 distinct advertised rates, not per listing - providers repeat one pay sentence across a whole batch, so counting listings would let a single advertisement vote thousands of times.

As of 2026-09-09. For the median rate itself and how it is derived, see the Pay Index.

Share this findingShare this findingShare this finding

How much do advertised rates vary?

The middle half of advertised AI-training rates falls between $25–$100/hr, but the full range runs from $1 to $400 per hour, with 9.0% of distinct advertised rates below $10/hr and 11.4% at $150/hr or above. Measured over 791 distinct advertised rates, not per listing.

$25–$100/hr · n=791

The rates split by how they were derived. 654 were stated outright, with a median of $65/hr; 137 came from text that hedged (“approximately”, “on average”), with a median of $25/hr. The two differ by more than double, so the cheap end of the distribution is substantially an artefact of how the figure was derived rather than what the market pays.

Why one pay sentence covers a thousand listings

5,732 listings state an hourly rate, but they carry only 791 distinct advertised rates - each one repeated across 7.2 listings on average. Providers post a batch of roles quoting one pay sentence, so counting per listing would let a single sentence vote thousands of times. Every rate figure on this page is computed over the distinct rates instead.

7.2× · n=791

Those rates are also unusually round: 72.2% are exact multiples of $5. These are advertised prices rather than negotiated ones, and advertised prices cluster.

How concentrated is this market?

Across 35 platforms with live listings, the largest single source accounts for 44.2% of indexed listings and the top three for 62.5%, giving a Herfindahl–Hirschman Index of 0.225 - moderately concentrated. This describes the listings HumanSourcer indexes, not the AI-training market itself: a platform that publishes no scrapeable board is absent from it entirely.

0.225 · n=35

How is pay actually quoted?

96.3% of listings that disclose pay quote an hourly rate. The other 221 quote per task, per unit, per project or as a salary, and are excluded from every hourly figure on this site - there is no honest way to convert a per-task price into an hourly one without knowing how long the task takes.

96.3% · n=5,953

How often is pay disclosed?

86.0% of active listings state some pay signal - but that is listing-weighted, and one large source posting thousands of near-identical roles moves it (Prolific Participants alone is the largest). Counted per platform instead, 28 of 35 (80.0%) publish pay on at least one open role.

86.0% · n=6,920

Where do listings say applicants must be?

Of the 3,031 listings whose title names a country (43.8% of 6,920), the most frequently named are USA (36.5%), United Kingdom (21.7%), Canada (6.0%). The remaining 3,889 state no country in their title, which is reported as unknown rather than assumed to be worldwide.

USA · n=3,031

How long do listings stay open?

Of 666 listings first indexed between 2026-08-26 and 2026-09-02, 41% had closed within 7 days. Only listings whose full 7-day window has elapsed are counted, so a listing posted yesterday cannot drag the figure down; the earlier bound is where closures stop being visible in the published data, so the cohort is a window rather than everything older.

41% · n=666

Which platforms advertise what

Platforms with at least 50 open roles. A median is shown only where the platform advertises at least 5 distinct rates - below that a single listing would set the figure, so it reads an em-dash instead.

PlatformOpen rolesMedianMiddle halfDistinct ratesPay disclosed
AfterQuery Experts206$105/hr$80–$1437999%
Mercor Experts381$90/hr$60–$140120100%
Handshake AI200$83/hr$59–$11328100%
micro1 Experts Opportunities312$80/hr$44–$12010398%
Sepal Expert Hub88$75/hr$60–$937698%
Rise Data Labs113$45/hr$41–$501075%
DataForce jobs153$41/hr$30–$52851%
SME Careers249$40/hr$26–$7037100%
Mindrift145$40/hr$34–$5013100%
Prolific Participants3060$30/hr$21–$805895%
Meridial831$30/hr$12–$655899%
Turing AI Advancement Work227$20/hr$15–$60713%
Welo Data434$16/hr$9–$2911443%
TELUS International AI Community portal72$15/hr$6–$251794%
Alignerr601100%

Explore the data

Filter by domain, use case, provider or date. Every figure recomputes against the filtered set, and the filter state is kept in the URL so a view can be shared.

Advertised rate distribution

Each bar counts distinct advertised rates, not listings - providers repeat one pay sentence across a whole batch, so counting listings would let a single sentence vote thousands of times. Grey shows rates whose source text hedged ("approximately $20/hr"); those run materially lower than stated rates, so the left tail is partly an artefact of how the figure was derived.

No listings match these filters.
Show data table
Advertised hourly rate distribution, counted over distinct advertised rates
Hourly rateDistinct ratesStatedHedged
$0–$10000
$10–$20000
$20–$30000
$30–$40000
$40–$60000
$60–$80000
$80–$100000
$100–$150000
$150–$200000
$200+000

Median and middle half by group

Median advertised rate with the middle half (25th–75th percentile) as the bar. Computed over distinct advertised rates; a group with fewer than 5 of them shows an em-dash rather than a figure a single listing could set. Click a row to filter.

No listings match these filters.

Pay bands by provider

Share of each provider's listings by published hourly rate, top 10 providers by volume. "Unspecified" means no hourly rate could be determined from the listing. Click a row to filter by that provider.

No listings match these filters.

Demand heatmap

Darker = more listings for that use case × provider combination. Click a cell to filter by both.

No listings match these filters.
Listing counts by use case and provider. Empty cells mean no listings for that combination under the current filters.

How these figures are derived

Listings are scraped daily from the public boards of 35 AI-training platforms. Pay is read from a structured field where one exists, parsed by rule where it does not, and otherwise extracted from the listing prose by a model that must quote the sentence it took the figure from.

Rates are deduplicated before any percentile is computed. The key is the provider plus the verbatim pay sentence, so a batch of listings sharing one advertisement counts once. This is the single most important thing to know about the numbers above: the per-listing figures are materially different and would overstate the sample by roughly seven times.

Only rates quoted per hour are comparable, so per-task, per-unit and salaried listings are excluded from every rate figure and counted separately under how pay is quoted.

No trend is published. The index begins at 2026-07-19, so there is not yet enough history for a rate or volume trend that would mean anything. The closure figure above uses a 7-day horizon and counts only listings whose full window has elapsed.

Concentration describes the listings indexed here, not the AI-training market as a whole - a platform that publishes no scrapeable board does not appear at all.

The aggregated figures on this page are also available as JSON. Reuse terms are the same as the Pay Index.

Common questions

How concentrated is the AI-training listing market?
Across 35 platforms with live listings, the largest single source accounts for 44.2% of indexed listings and the top three for 62.5%, giving a Herfindahl–Hirschman Index of 0.225 - moderately concentrated. This describes the listings HumanSourcer indexes, not the AI-training market itself: a platform that publishes no scrapeable board is absent from it entirely.
Why do so many AI-training listings quote the same pay rate?
5,732 listings state an hourly rate, but they carry only 791 distinct advertised rates - each one repeated across 7.2 listings on average. Providers post a batch of roles quoting one pay sentence, so counting per listing would let a single sentence vote thousands of times. Every rate figure on this page is computed over the distinct rates instead.
Do all AI-training jobs pay by the hour?
96.3% of listings that disclose pay quote an hourly rate. The other 221 quote per task, per unit, per project or as a salary, and are excluded from every hourly figure on this site - there is no honest way to convert a per-task price into an hourly one without knowing how long the task takes.
How often do AI-training platforms publish pay at all?
86.0% of active listings state some pay signal - but that is listing-weighted, and one large source posting thousands of near-identical roles moves it (Prolific Participants alone is the largest). Counted per platform instead, 28 of 35 (80.0%) publish pay on at least one open role.
Are advertised AI-training rates round numbers?
Yes - 72.2% of the 791 distinct advertised rates are exact multiples of $5. Rates are advertised, not negotiated, so they cluster on round figures rather than spreading evenly.