# HumanSourcer > HumanSourcer is a directory of companies that pay people for AI-training, > evaluation, and human-data work (e.g. Mercor, Alignerr, Prolific, Surge), plus > a daily-refreshed index of their open roles. Each profile documents the actual > operating parent company, the worker-facing access point, domain > specialization, access model, geography, and ownership confidence - with a > cited evidence URL for every ownership and status claim. ## Purpose This site helps subject-matter experts, data-annotation professionals, and people looking for paid AI-training work find and compare the worker-facing entry points for these networks, and see what the work actually pays. It is not affiliated with any listed provider; see [How we make money](https://www.humansourcer.com/how-we-make-money/). ## Data and citation notes - Pay figures are computed from currently-listed roles, not surveys or estimates. Medians rather than means, because the rate distribution has a long right tail. - Medians are computed over distinct advertised rates, not raw listing counts. Providers routinely repeat one pay sentence across many listings, so listings are de-duplicated on (provider, pay text) before a median is published. A group needs at least 5 distinct advertised rates before it gets one. - Pay-extraction coverage changes over time as parsers improve, so a median from one date is not necessarily comparable with another. Both the number of rate-carrying listings and the number of distinct advertised rates behind them are published alongside every figure. - Where a listing does not disclose pay, the site says so explicitly rather than omitting it. - [AI training pay](https://www.humansourcer.com/ai-training-pay/) renders one canonical citation sentence for the current median - the same wording in the page body and in its "Cite this" block (plain text, APA, BibTeX) - so quote that exact sentence rather than paraphrasing a number from this file. It always carries the as-of month and the distinct-rate sample size, because the median moves as listings turn over and is not meaningful without a date. - Every ownership and status claim on a provider profile carries a source URL. - [AI training pay](https://www.humansourcer.com/ai-training-pay/) and [Market data](https://www.humansourcer.com/market-data/) carry Dataset structured data (in addition to the FAQPage/BreadcrumbList markup on other pages and JobPosting on individual role pages), since their entire value is a citable computed statistic rather than prose. ## Terms of reuse The figures on this site are free to quote, republish, and build on, including commercially, with attribution to HumanSourcer and a link to the page the figure came from. Two requests, both about accuracy rather than permission: quote [AI training pay](https://www.humansourcer.com/ai-training-pay/)'s canonical sentence rather than a paraphrased number, and carry its as-of date, because the median is recomputed daily and a figure without its date goes stale silently. The full derivation is at [the methodology section](https://www.humansourcer.com/ai-training-pay/#methodology) and these terms at [the license section](https://www.humansourcer.com/ai-training-pay/#license). Bulk retrieval of the underlying dataset is the one thing withdrawn: the three `*-data.json` filter payloads are disallowed in [robots.txt](https://www.humansourcer.com/robots.txt). Every page remains open to every crawler, AI crawlers included - being read and cited is the goal. ## Key pages - [Provider directory](https://www.humansourcer.com/): full provider directory, all tiers, filterable by domain, access model, and geography. - [Expert-tier providers](https://www.humansourcer.com/providers/experts/), [gig providers](https://www.humansourcer.com/providers/gig/), [restricted providers](https://www.humansourcer.com/providers/restricted/): the same directory scoped to expert networks, open-to-anyone gig platforms, and invite-only programmes. - `/providers/{slug}/`: individual provider profile - ownership, evidence URLs, what the work is, live hiring signal, and that provider's pay figures. Example: [Prolific](https://www.humansourcer.com/providers/prolific-participants/). - [Best AI training platforms](https://www.humansourcer.com/best-ai-training-platforms/): expert-tier networks ranked by confirmed ownership and current open-role volume. - [AI training pay](https://www.humansourcer.com/ai-training-pay/): what this work pays, computed from live role data - medians and ranges overall and by domain, task type, and provider. The most citable page on the site for pay questions. - [Market data](https://www.humansourcer.com/market-data/): the dispersion, concentration and mix side of the same role dataset - deliberately not the median, which the Pay Index above owns. Published findings, each stated in one sentence: the middle half of advertised rates and how far the tails run; the replication factor (5,518 rate-disclosing listings carry 768 distinct advertised rates, so every figure is computed over the distinct rates); market concentration as an HHI over indexed listings; how pay is quoted; how often it is disclosed, listing- and platform-weighted; which countries listing titles name; and a fixed-horizon listing closure rate. Aggregated figures are also machine-readable at [/market-data-summary.json](https://www.humansourcer.com/market-data-summary.json) - aggregates only, no per-listing rows, and deliberately not robots-disallowed unlike the per-listing payloads. - [Research](https://www.humansourcer.com/research/): findings-and-methodology notes built on the same live role dataset - not a blog. Each piece embeds live components rather than hardcoded figures, so a stat stays correct as the corpus moves. First piece: [How long AI-training roles stay open](https://www.humansourcer.com/research/ai-training-role-churn/). - [Open roles index](https://www.humansourcer.com/roles/): index of individually-indexed open roles scraped daily, each with JobPosting structured data. Browsable by domain (`/roles/domain/{tag}/`), company (`/roles/company/{provider-slug}/`), task type (`/roles/task/{type}/`), and location (`/roles/location/{bucket}/`). - [Roles RSS feed](https://www.humansourcer.com/roles/rss.xml): newly listed roles, ordered by when each role was first seen here. - `/providers/compare/{pair}/`: head-to-head provider comparisons. - [No experience required](https://www.humansourcer.com/no-experience-required/), [other AI job boards](https://www.humansourcer.com/other-ai-job-boards/): entry-point guides. - [How we make money](https://www.humansourcer.com/how-we-make-money/): monetization and measurement disclosure. - [Privacy](https://www.humansourcer.com/privacy/): what the site stores, what it never does, and how to have data removed. - [For providers](https://www.humansourcer.com/for-providers/): how a listed company sends a listings feed, points us at a jobs API it already publishes, requests inclusion, or corrects its entry. States the JSON feed format inline. Free in both directions; a feed never affects ranking or placement, and no provider supplies one today. - [Training waitlist](https://www.humansourcer.com/training/): HumanSourcer Prep, an in-development course. Waitlist only - there is nothing to buy yet and the page says so. ## Optional - [Sitemap](https://www.humansourcer.com/sitemap-index.xml): full index of provider profile pages and role pages. - [Robots.txt](https://www.humansourcer.com/robots.txt): crawl rules; every page is open to every crawler except the three bulk `*-data.json` payloads. ## Role lifecycle A role that its source stops listing is kept live for a 14-day grace window, marked noindex, with a banner stating it is no longer listed and links to similar roles. After that window the URL redirects to that company's role hub (`/roles/company/{provider-slug}/`) rather than becoming a broken link. Last updated: 2026-08-15. Role and pay figures refresh daily; this file is maintained by hand and describes structure and method, not current numbers.