How we compile pay

• We read hourly rates straight from each platform's own job feed, read every hour, and take the median and the typical pay across the open roles that state a rate. Typical pay is the 25th percentile of what roles pay (the middle of each advertised range) to the 90th-percentile advertised ceiling, so it includes the well-paid expert, coding and agentic tier, and it is measured the same way for every list of roles: the whole market, a platform, a kind of work, a platform’s roles of one kind, or a search. Every pay figure opens the roles it is measured from, and that list states the same figure.
• We count distinct advertised rates after removing duplicate relistings of the same job, so one role posted many times cannot skew the median. A rate count is never reported as an open-role count: they measure different things.
• We exclude advertised ceilings ("up to $X/hr"), which are marketing maximums, not observed wages. A platform whose feed states only ceilings shows no live median, and we say so rather than dress a ceiling up as a going rate.
• We only call a figure a market rate when it spans several platforms with a large enough sample. A rate from a single platform is shown as that platform's rate card, clearly labeled, and never published as the going rate for the whole category.

Some links on AITraining.jobs are referral links. If you sign up through one, AITraining.jobs may earn a commission at no extra cost to you. It never changes what a platform pays you.

Referral fees are not our only income. We also run paid expert-data projects for companies building AI, staffed with professionals who join our expert bench under our expert terms: the company pays us for the work, and we pay the professionals who do it. That work is no input to any figure or order described here: a company we work for, or may work for, gets no lift in any list, and its pay figures are compiled the same way as every other platform's. Some companies whose roles we list may also be, or become, clients of these projects, and some of the platforms we list run similar projects themselves.

Here is exactly how the referral relationship does, and does not, affect what you see. It never touches the facts: a pay figure and its confidence label are compiled the same way whether or not a platform pays us, and the two sorts you can choose, highest pay and newest first, rank purely by the number and the date, with no referral input at all.

The Top Networks list on the home page is a featured list: it shows the networks we have a referral relationship with, in order of what a visit we send them is worth to us (how often the people we send end up paid there, times the referral fee), learned from each network's own referral records and updated as they change, and a line right under it says that we may earn a referral fee when you join through those links. The home page's platform filter and its “Platforms hiring right now” list are in the one order of every list of platforms, explained below, and the referral disclosure on that page covers it.

Every other list of platforms works the same way: the groups in the full platform directory, the platforms that hire for a kind of work, profession, language or country, the alternatives to a platform and the platforms to compare it with, the comparison pages, the platform guides in other languages, the platforms on this page about how the site works, the interview practice list, and the platforms to join in your workspace. Each is ordered by what applying there is worth, explained below, times one plus a referral boost for the networks we have a referral relationship with (from 20% for the network whose visits are worth least to us to 100% for the one worth most, in proportion), so a network is lifted, never put first whatever it is worth. A network gets no boost while it cannot credit anyone new (its referral limit is full), when you are already on it, or when it is closing to new joiners. Platforms of equal worth keep the list's own order (where a list shows how many roles each platform has, that number); where a sentence says which platforms list the most roles, it names them by their roles, with no referral input. Lists of roles, the similar roles on a job's page included, give those networks' roles their referral boost inside the one figure every list of roles is ordered by (below). Two things never carry any referral input: the pay figures, and the highest-pay and newest sorts.

Every list of roles on this site is ordered by one figure: the job board, the AI search and your matches, the lists of a platform, a kind of work, a country or a profession, the similar roles beside a role, and the roles in our emails and messages. The figure is how likely the role fits you, times how much people apply to it for the attention it gets, times how crowded it is, times one plus the referral boost of a platform we have a referral relationship with. How likely it fits you is our AI check's answer where it has checked the role for you (a strong or possible match fits, an unlikely one does not), else the chance a model reads from how close the role is to your profile, fitted on every check our AI has made; a list for nobody in particular counts every role the same here. How much people apply to a role is its apply clicks for the number of times people actually saw it (a role counts as seen when at least half of it is on screen for a second), leaning on the roles closest to it in meaning while it has had little attention, so a role moves with what people do when they see it rather than with its age or the place a list gave it. A platform gives a role out only so many times, so how crowded a role is counts how many of our visitors already applied to it and how much that lowers the next person's chance of being taken: the chance with that many people ahead over the chance with nobody ahead, measured on the applications people told us the outcome of, for each platform leaning on the average platform while it has few outcomes. A role nobody applied to yet counts in full, and a similar role with fewer people ahead rises, so applicants spread over the roles where each has the better chance. The referral boost is from 20% to 100%, by what a visit to that platform is worth to us, and none while it cannot credit anyone new, for a platform you are already on, or for any platform without a referral relationship. So a partner's role weighs more, but a role that fits you better, or that people want far more, still comes first, and partner roles appear where that figure puts them. Where a role genuinely pays above the market, we mark it with that fact (“Top-tier pay” when it is in the top quarter of what open roles pay, “Above-typical pay” when it pays more than the typical role), measured from the same live rates the rest of the site uses, so the emphasis rests on a real feature of the role, not a label. When you sort by Highest pay or Newest, the roles rank by the number or the date first.

When you describe the job you want, or match your LinkedIn profile, an AI model reads what you are looking for: a role matches when it is more likely to fit you than not, or when its title names a word you typed. The matches come first and every other role after them, each in the same figure. With Newest or Top pay, the matching roles rank by the date or the number first.

The roles like a role are ordered by the same figure, with how likely it fits you read as how likely someone who fits that role fits this one: a model fitted on every two roles our AI check has judged the same person on (how alike the two roles are in meaning, the language and field their titles name, the kind of work and the platform). As in the AI search, the roles more likely to fit than not come first, each in the figure, and the rest after them, so a role many people want never passes a closer one that fits. That is the list on a role's page, the open roles beside a role that has closed, the roles at our referral partners in a role's summary and under the roles of a platform we have no referral relationship with, and “Similar roles hiring now”, a small card after you apply to such a platform. The card lists up to three of them on other platforms, open where you are, never a role you were already shown, and never in a visit where you already applied through a referral link.

Clicking a role opens a short summary of it: its pay, where it is open, the start of its description and what the platform is. With a free profile, an AI model also checks the role against your background and tells you whether you are a strong, possible or unlikely match, and why. For a role at one of our referral partners the summary carries its apply link straight away. For a role at a platform we have no referral relationship with, the apply link is part of your free profile: create one (your LinkedIn link, your CV or a sentence about your work, an email and a password) and the link appears. Beside such a role the summary lists up to three of the roles most like it at our referral partners that you can apply to right away, open where you are, with their pay and how each platform pays; once it has checked them for you, the ones that fit you come first.

Our emails and messages follow the same figure: the new-role alerts for experts, the roles our assistant suggests, our Telegram channel and its alerts. The weekly roles email and the welcome email list the roles first listed this week, in that figure, as they promise what is new. An invitation to refer a colleague lists roles at the platforms that pay us a referral fee, the ones that would earn you the most first. Each of them says so. Apps that read our jobs inside an AI assistant or an app directory (our MCP server, the jobs API, Poe) get a neutral order, as those directories require.

Either way, this referral disclosure appears on every surface where partner status affects the order, and clicking any link takes you to that platform under its own terms.