PhDs and grad students are the highest-paid tier of AI training contributors in 2026 by a wide margin — $80–$200+/hr for part-time evaluation and reference-authoring work that fits around dissertation deadlines. The application bar is real but structurally easier for academics than for any other applicant pool: the platforms specifically request the credentials you already have, and the AI-led interview is built around the kind of detailed project-talk that grad students do every day.
This guide is for current master's students, PhDs, and postdocs deciding whether to take AI training work as a side income, and which platforms to apply to first.
Why PhDs are the highest-paid tier
Frontier AI labs need expert evaluation of model outputs in narrow domains — mathematical proof correctness, biological reasoning, legal argumentation, medical diagnosis, physics problems beyond the textbook. The pool of qualified evaluators in each domain is small (often just hundreds globally), the labs are well-funded, and the work has direct measurable impact on model quality. Three structural forces push these rates up:
- Credential rarity. A biology PhD evaluating molecular biology reasoning can't be substituted by a generalist. Labs pay accordingly.
- Output value. A PhD's evaluation calibrates the model's behavior across thousands of similar future examples. The lab's revenue per unit of judgment is high.
- Time scarcity. PhDs and postdocs have limited hours to give. Platforms compete on rate to access them.
The three best platforms for PhDs in 2026
1. Handshake AI Fellowship — purpose-built for U.S. grad students
Handshake AI Fellowship is the program most structurally aligned with grad students. ~200 active opportunities at any time. Up to $125/hr. Fully remote, part-time, U.S. work authorization required. Disciplines covered span the sciences, engineering, humanities, and design. If you're at a U.S. institution and your school is a Handshake partner (most R1s are), you sign up via your school account. Non-partner schools and recent graduates can still apply with a personal account.
Why it's structurally easier: the application explicitly asks for your degree status, research area, and adviser — academic credentials are the primary signal. The AI interview, when present, is calibrated to the kind of structured-reasoning answer academics give naturally.
2. Mercor's specialist tier — the highest hourly rates
Mercor's listings include the highest-pay specialist work in the AI training market. Their domain-expert listings — Equity Research Expert ($120/hr), Pediatric Neurology, Patent Law, Senior Software Engineer review ($130–$200/hr), specialty mathematics and theoretical CS — all sit at $80–$200/hr+. Read our Mercor review for the application playbook.
For PhDs: apply to the most specific listing your subfield supports. Mercor's matching algorithm rewards narrow positioning — "Computational biologist, 4 years wet-lab + ML protein structure" beats "Biology PhD" by a wide pay margin on the same listing.
3. micro1's expert and engineering tiers
micro1 runs roughly 60–80 engineering and specialist roles at any time with pay in the $50–$150/hr band. Lower pay ceiling than Mercor but better acceptance bar for non-frontier-lab work and faster application turnaround. Good complement to Mercor for filling between project gaps. See our micro1 review.
Where each subfield should look first
- CS / EE / Math: Mercor for top-end, Handshake AI for steady volume, micro1 for engineering-leaning. Run all three in parallel.
- Hard sciences (Bio, Chem, Physics): Mercor's specialist tier surfaces these listings most consistently. Handshake AI Fellowship runs research-evaluation projects in each discipline.
- Medicine: Mercor explicitly hires for medical AI evaluation at $80–$200/hr depending on specialty. Active medical license required; the listings ask for it up front.
- Law: Mercor's law listings (patent law, contract review, regulatory) range $80–$200/hr. JD required.
- Humanities and social sciences: Handshake AI covers these well; Mercor less so. Mindrift also runs writing-quality and research-methodology projects across disciplines. See our Mindrift review.
- Languages and linguistics: Welo Data runs ~480 multilingual projects; pay scales with language rarity.
How much can a PhD realistically earn?
At common rate bands and part-time hours:
- $125/hr × 10 hrs/wk (Handshake AI): ~$5,400/month
- $150/hr × 10 hrs/wk (Mercor mid-specialist): ~$6,500/month
- $200/hr × 8 hrs/wk (Mercor top-tier specialist): ~$6,900/month
For grad students on stipend, even 5 hours per week at these rates roughly doubles take-home income. For postdocs, the delta is even larger. Most contributors work in evening or weekend blocks; the work compresses well because the tasks are bounded (5–20 minutes each) rather than open-ended.
One non-obvious advantage academics have
The AI-led interviews these platforms use are calibrated to catch generic, vague answers. Academics are trained to talk about their work in specifics — methods, results, limitations, next steps. That's the exact register the interviews score for. Most PhDs we've heard from clear the interview on the first attempt without specific prep; most non-academics fail at least once and need to re-record.
Tactical preparation tips for the AI interview live in our getting-accepted playbook — same playbook applies to PhDs, but you'll find most of the prep is already part of how you talk about research.
Time and tax notes for grad students
- Stipend conflicts. Some PhD programs cap or prohibit outside paid work. Check your specific funding agreement before signing on. Most programs allow part-time consulting; some require disclosure or approval.
- 1099 / self-employment tax. AI training pay is contractor income, not W-2. In the U.S. you'll owe self-employment tax (15.3%) on top of federal/state income tax. Quarterly estimated taxes are the standard. See our AI training jobs taxes guide for the full setup.
- International students on F-1. AI training contract work generally falls outside CPT/OPT employment authorization. Most platforms require U.S. work authorization explicitly. Check with your DSO before applying — getting paid for unauthorized work has visa consequences.
Where to start
Apply to Handshake AI Fellowship first if you're U.S.-based — it's the most credential-aligned and has the most consistent pipeline. Run Mercor applications in parallel; the AI interview practice transfers and the top-end pay justifies the effort. Then layer micro1 and Mindrift as you build track record.
For broader landscape context, see our 2026 best AI training platforms guide. For the full pay-tier model, see how much do AI training jobs pay.