Mercor is a two-sided marketplace that turns human expertise into AI training and evaluation data. Doctors, accountants, lawyers and engineers take paid remote projects on it, while AI labs and enterprises use it to source vetted specialists, commission RLHF data and deploy agents.
Used this tool? Rate it
Used this tool? Rate it
Mercor is a two-sided marketplace that converts human professional expertise into the training and evaluation data that modern AI systems need. The company states its mission plainly on its homepage: Mercor is organizing human intelligence to power the AI economy. That sentence is worth reading carefully, because it explains why this product is easy to misfile. It looks like a job board from one angle and like a data vendor from another, and both views are partially correct.
The clearest way to understand Mercor is to recognise that it has two distinct audiences who almost never read the same page. On one side are individual professionals — physicians, accountants, lawyers, quantitative traders, chemists, designers, engineers — who apply for paid remote projects. On the other side are AI labs and large enterprises that need those professionals' judgement captured as structured data. Mercor's own description of the business covers the second group: it describes itself as powering frontier research, RLHF data, and AI agent training at scale for the top AI labs and enterprises.
This document deliberately treats both sides separately. If you arrived wondering whether you can earn money here, the expert-side sections are the relevant ones. If you arrived evaluating a vendor for your company, the enterprise sections apply. Conflating them produces the confusion that most short descriptions of Mercor fall into.
The commercial mechanics matter for understanding incentives on both sides. Independent reporting describes the model as connecting AI labs with domain experts for foundation model training, and charging an hourly finder's fee and matching rate for their work. In other words, experts are paid by Mercor rather than paying it; the platform's revenue comes from the enterprise side. For a professional considering signing up, this is a meaningful structural fact: there is no subscription or listing fee to work here.
The underlying market condition is that frontier AI models have largely exhausted the easy data. What remains valuable is judgement that only qualified humans possess: how a physician reasons through an ambiguous case, how an auditor identifies a misstatement, how a securities lawyer reads an edge case. Reporting on the sector notes that Mercor's customers include major AI labs and that it recruits PhDs, lawyers, bankers, scientists and programmers to create specialized training data. That recruitment profile is the clearest signal of what the platform actually is: not a gig site for generic microtasks, but a channel for credentialed expertise.
The navigation itself encodes the split. The main site carries APEX … Research … Enterprise … Experts … Mission, and the hero offers two parallel calls to action: Start working … Learn more. The expert workflow then moves to a separate domain entirely — the expert side lives on a separate workspace at work.mercor.com, with its own account system, while enterprise conversations stay on the main site and run through a sales process.
The intake path for professionals is short and standardised. You build a profile, complete a brief AI-conducted evaluation interview, and then, if there is a fit, our team follows up and connects you to projects aligned with your skills. The AI interview is central to the model — it is what allows the platform to screen at a scale that human recruiters could not match, and it is also the step most applicants ask about.
Notably, the platform does not gate entry on AI experience. Its own guidance to applicants is explicit: You don't need any prior AI experience. What matters is depth in your own field. A tax accountant is valuable because of the tax knowledge, not because of familiarity with machine learning.
Three practical facts define the working arrangement. Payment is on a weekly cycle — the site states you Get paid weekly for contributing your expertise. Work is remote and self-scheduled. And terms are agreed before work begins rather than negotiated afterwards.
Rates are unusually transparent for this category. The expert page reports that the average contracted rate is $112/hr, with rates ranging from $60-$250/hr depending on expertise and project complexity. Live listings corroborate this range: a health insurance insight study at $120/hr, accounting roles between $80 and $120/hr, product and finance roles around $80–90/hr, and a Chemistry Expert (PhD) — AI Safety posting in the $65–70/hr band.
Not every engagement is hourly. The expert workspace also carries task-based listings paying around $2K per task, and some roles are restricted to US Based contributors, alongside a mix of project-based and one-time engagement types. Checking which pricing basis applies before accepting is worth the minute it takes.
The enterprise offering is organised around the AI lifecycle rather than around staffing. The site presents Agent Diagnostics to identify high-value AI opportunities, Agent Deployment to put AI into production, Agent Optimization to improve every agent already running, and Data Monetization to unlock new data revenue. This is consultancy-shaped work with a data supply chain attached, not a self-serve tool.
Delivery combines platform and people. The company describes its method as AI interviews, enterprise data, and workflow discovery, paired with Forward Deployed Engineers who embed with the client. The AI interview appears on this side too, used to map how work is actually performed inside an organisation before any agent is built — the same screening technology pointed at a different problem.
Mercor also publishes a research line built on the same expert base. APEX, the AI Productivity Index is its benchmark family, with public leaderboards including APEX-Agents, APEX-Accounting and APEX-SWE, plus off-the-shelf datasets for buyers who want data without a custom engagement. For a company evaluating Mercor, these benchmarks are a useful artefact: they show the kind of professional-grade evaluation the platform can construct.
The most common individual use case is a working professional monetising expertise part-time. A practising accountant reviews model outputs on financial reasoning; a physician evaluates clinical summaries; a chemist assesses safety-relevant responses. The work is remote, scheduled around existing commitments, and paid weekly. For professionals in fields where side income traditionally means consulting engagements with long sales cycles, the appeal is the absence of business development.
A second pattern is professionals who want exposure to AI work without retraining. Because prior AI experience is not required, a domain specialist can enter through the door of their existing expertise. Roles like a data engineer position requiring coding-agent experience sit alongside purely domain-based roles, so the ladder exists for those who want it.
For an AI lab, the problem is that improving a model in accounting, medicine or law requires practitioners in those fields, at volume, briefly, and with verified competence. Hiring them as employees is impractical. This is the demand Mercor was built to serve, and reporting indicates the platform has more than 30,000 experts on its roster available for exactly this kind of engagement.
Beyond data supply, enterprises engage Mercor to diagnose where agents would create value, deploy them, and keep improving them once running. Published examples on the enterprise side include incident response work and value assessments, delivered through the embedded-engineer model rather than as software licences.
A third enterprise use is measurement — determining whether an AI system actually performs at professional standard in a specific domain. The APEX benchmark family exists precisely because generic benchmarks do not answer that question for accounting or software engineering work.
The fourth line addresses organisations sitting on proprietary data with latent value. This is a distinct commercial relationship from both hiring experts and buying data, and it is worth naming separately because it changes who the counterparty is.
Create your profile on the expert workspace with your actual professional credentials. Because matching is driven by domain expertise, specificity helps: the difference between "finance" and "FP&A with SaaS revenue recognition experience" determines which projects you surface for.
Expect an AI-conducted interview rather than a human screening call. Treat it as a professional assessment: answer with the depth you would use with a colleague in your field, since the evaluation is trying to establish genuine expertise rather than communication polish.
You can apply to specific listings, and the platform also routes opportunities to you. The stated flow is that the team follows up when there is a fit. Because supply is project-driven, timing varies — listings visible today reflect current client demand rather than a permanent catalogue.
Payment terms are defined upfront, so read them. Establish whether the engagement is hourly or per-task, what the approval process is, whether geography restricts eligibility, and how long the project is expected to run. This step deserves emphasis because payment-basis ambiguity is the subject of documented complaints, covered in the limitations section.
Once engaged, work proceeds remotely on your own schedule with weekly payouts. The platform also runs a referral programme if you bring in other qualified professionals.
Decide which of the four service lines matches your need — expert-generated data, agent diagnostics, deployment, optimisation, or data monetization. These are different engagements with different deliverables.
There is no self-serve checkout. Enterprise engagement begins with a demo conversation, and pricing is scoped to the work.
Expect the engagement to start with workflow discovery and AI interviews inside your organisation before anything is deployed. Budget internal time for this: the quality of the diagnosis depends on access to how work is really done.
Before deployment, specify precisely which systems, data and permissions the agents may touch. The platform's model puts this in your hands, so treat it as a design decision rather than a formality.
Experts: be specific about sub-specialty. Matching runs on domain depth. Broad self-descriptions surface for fewer relevant projects than precise ones, because the demand side is usually looking for a narrow competence.
Experts: verify the payment basis in writing before you start. Confirm whether you are paid hourly or per approved task, and what "approved" means in practice. This single check prevents the most commonly reported dispute on the platform.
Experts: keep your own records. Save offer documents, correspondence and time logs. This is standard practice for any contract work and is prudent here given the documented complaints about changing terms.
Experts: treat the AI interview as a real assessment. It is the primary gate. Preparing to discuss your field substantively pays off more than preparing to sound polished.
Experts: size expectations to project supply, not to a salary. Work is project-based and episodic. The published rates are attractive, but availability fluctuates with client demand, so this suits supplementary income better than a guaranteed full-time replacement.
Companies: bring a real workflow, not a hypothesis. Since delivery starts with discovery grounded in your actual data and processes, engagements go faster when internal access is arranged in advance.
Companies: define permission boundaries early. Agent scope is yours to set. Deciding it deliberately at the start avoids retrofitting controls later.
Companies: use the benchmarks as an evaluation tool. The public APEX leaderboards indicate how the vendor measures professional-grade performance, which is useful for calibrating your own acceptance criteria before committing.
Both sides: expect a services relationship, not a product. Neither side of this marketplace is self-serve software. Plan for human coordination, matching time, and scoping conversations.
Credentialed professionals seeking flexible paid work are the clearest supply-side fit. Physicians, accountants, lawyers, financial analysts, product managers, designers, HR specialists and PhD scientists all appear in live listings. If you hold genuine domain expertise and want remote, weekly-paid project work, this is the audience the expert side is built for.
Specialists curious about AI without AI backgrounds fit well precisely because prior AI experience is not required. The platform converts your existing field knowledge into the qualification.
AI labs and foundation model teams are the core demand-side customer — organisations that need expert judgement captured as training and evaluation data at volume.
Enterprises deploying AI into real workflows are the other demand-side group, served by the diagnostics-through-optimisation service lines with embedded engineers.
Organisations with proprietary data assets may engage through the data monetization line, a different relationship again.
Who this is not for: individuals seeking stable full-time employment with benefits — this is contract, project-based work. Companies wanting a self-serve tool with published pricing will find the enterprise side is a consultative sales motion. And people looking for general microtask work without specialist credentials are not the target; the model depends on verified expertise.
Mercor operates as a web platform split across several domains, and knowing which one you need saves confusion. The marketing, research and enterprise content sits on the main site. The working environment for experts is separate: the expert side lives on a separate workspace at work.mercor.com, where profiles, listings, applications and account management happen. Documentation for contributors sits on a third domain, a dedicated talent help centre.
That help centre is a meaningful signal of operational maturity for the supply side. It covers interviews, payments and policies, and includes a Policies section spanning AI utilization, social media, payment procedures, and compliance regulations, plus guidance on how hours, referrals and payouts work. Contributors are operating under written rules rather than ad hoc arrangements.
The expert workspace includes profile and home views, and a referral programme with earning potential for bringing in other qualified experts.
On the enterprise side, contact channels are separated by function — support, press and sales each route differently — and there is a dedicated trust and security portal for customers reviewing the vendor. No mobile applications are advertised; both sides are browser-based.
Pricing works in opposite directions for the two sides, which is why a single "price" figure would be misleading.
For experts, there is no cost — you are paid. Mercor's revenue comes from the enterprise side, described in independent reporting as an hourly finder's fee and matching rate. There is no subscription or fee to join or apply.
Expert compensation is published with unusual clarity. The average contracted rate is $112/hr, with rates ranging from $60-$250/hr, varying by expertise and project complexity. Live listings show the range in practice, from roles in the $65–70/hr band to a $120/hr health insurance study. Payment is weekly, and terms are set before work begins.
Payment structures vary by engagement. Alongside hourly work, the workspace carries task-based listings paying around $2K per task, and some roles are restricted to US Based contributors. Project-based and one-time engagement formats both exist. Because the basis differs, the headline hourly average is not a reliable predictor of what any specific project will pay.
Scale of payouts, per independent reporting of what the company told investors, is that it pays more than $1.5 million per day to its contractors, with more than 30,000 experts on its roster earning over $85 per hour on average. These are company-reported figures relayed by media rather than independently audited numbers, and should be read as directional.
For enterprises, pricing is not published. Engagement runs through a demo and a scoped commercial conversation. Given that delivery involves embedded engineers and custom discovery, costs are necessarily project-specific. Any organisation evaluating this should expect a consulting-style commercial structure rather than per-seat licensing.
Data products including off-the-shelf datasets exist as an alternative to fully custom engagements, which may suit buyers wanting a lower-commitment entry point, though pricing for these is likewise not public.
The landscape splits by which side of the marketplace you are on.
If you are an expert looking for work, comparable platforms include Outlier, DataAnnotation, Alignerr, Prolific and Mindrift. These differ substantially in required qualifications and pay bands: several accept generalist contributors at lower rates, whereas Mercor's published range skews toward credentialed specialists. The practical approach is to register with several, since availability is project-driven everywhere in this category.
If you are an AI lab buying expert data, the historical incumbent is Scale AI, alongside Surge AI, Turing, Invisible Technologies and Handshake's AI arm. The competitive question is depth of credentialed expertise versus throughput on generic labelling.
If you are an enterprise deploying AI, the comparison set is different again: consultancies with AI practices, systems integrators, and forward-deployed engineering firms. Mercor's distinguishing claim here is that its expert network feeds the same engagements.
For traditional expert networks such as GLG or AlphaSights, the difference is output form. Those firms sell conversations with experts; Mercor converts expertise into structured data and deployed systems.
Documented disputes about changing payment terms. Public reviews include a specific complaint that the payment model was changed to payment per approved task after being presented as hourly, and that a written offer stated to be valid for fourteen days was withdrawn after two. Whether or not this is representative, the practical response is the same: confirm terms in writing and keep records.
Public review samples are small and fragmented. The marketplace's review profile carries only five reviews, and the platform itself flags that This company hasn't invited customers recently, so reviews may not be representative. Separately, a differently-registered profile for a Mercor domain shows several hundred reviews with a much higher score. Because the two profiles cannot be reconciled from primary sources, no star rating from either is a sound basis for judgement here.
Work is episodic, not employment. Listings reflect current client projects. There is no guarantee of continuous work, no benefits, and no fixed hours. Treat published rates as the price of available work, not as an income floor.
Approval processes affect actual earnings. Where compensation is per approved task, the approval standard directly determines effective hourly earnings. Clarify what approval requires before accepting task-based work.
Geographic restrictions apply to some roles. Certain listings specify US-based eligibility, so location can limit which opportunities are open to you.
The AI interview is a genuine filter. Not every applicant is matched. Applying is not equivalent to being engaged, and the timeline between application and first project is not guaranteed.
Enterprise pricing opacity. There is no published price for enterprise work, which makes budget planning impossible before entering a sales conversation and complicates like-for-like vendor comparison.
Vendor-reported metrics. Expert counts, average rates and daily payout figures come from the company via media reporting. They are plausible and consistently reported, but they are not independently audited.
Dual-audience confusion in third-party descriptions. Much external coverage describes Mercor as either a hiring tool or a data labelling company. Both descriptions omit half the business, so evaluate against the primary sources for whichever side concerns you.
Structurally it is both, which is why single-sentence descriptions mislead. It recruits and screens professionals like a hiring platform, but what it sells is the output of their expertise: training data, evaluations and deployed agents for AI labs and enterprises. Which half matters depends entirely on which side you are approaching from.
No. Experts are paid rather than charged. The platform's revenue comes from the enterprise side through fees on matched work, so there is no subscription, listing fee or commission deducted from a stated rate.
The published average contracted rate is $112 per hour, with a range of roughly $60 to $250 per hour depending on expertise and project complexity. Live listings span roughly $65 to $120 per hour, and some engagements pay per task rather than hourly. Actual earnings depend on project availability and how many hours you take on.
No. The platform states that no prior AI experience is required. The qualification is depth in your own professional field — accounting, medicine, law, chemistry, design and similar domains all appear in active listings.
You create a profile, complete a short AI-conducted evaluation interview, and are matched to suitable projects when a fit exists. The AI interview is the main filter, which allows screening at a scale human recruiters could not reach. Applying does not guarantee placement.
Payment is weekly, and terms are defined before work begins. Confirm at the outset whether your engagement is hourly or per approved task, since that determines how your effective rate is calculated.
Four lines: agent diagnostics to find high-value AI opportunities, agent deployment into production, ongoing agent optimisation, and data monetization. Delivery blends the platform with Forward Deployed Engineers who work inside the client organisation, informed by AI interviews and workflow discovery.
Enterprise pricing is not published. Engagement starts with a demo and is scoped commercially, which is consistent with a delivery model involving embedded engineers and custom discovery rather than licensed software.
The stated position is that agents operate only within the systems, data and permissions the customer defines, with the customer determining how agents interact with enterprise systems. Managed deployments to customer-controlled infrastructure are available, alongside monitoring and auditability of agent activity.
Treat them cautiously. The marketplace profile has a very small sample and carries a platform warning that reviews may not be representative, while a separately registered Mercor domain profile shows a much larger sample and a substantially different score. Since the two cannot be reconciled from primary sources, individual documented complaints are more informative than any aggregate score.
