How to hire AI research interns for summer 2027

Who is a research intern, in practice?

In AI research, the intern you want is usually a doctoral student in the middle of the degree. They have published once or twice, they know how to run an experiment end to end, and they still have a year or more before the defence. A summer with you is a chance to work on a problem their own lab cannot give them, often with more compute.

Two other groups apply too. Master's students can be excellent, but the index only sees them once they publish, and most have not yet. Final year doctoral students are a different case: they are about to choose a first job, and a summer project is a slow way to make them an offer. For them, read how long an AI doctorate takes and treat the conversation as a hire.

The first paper is the simplest marker of where someone stands. Between 4 and 5 years separate a first indexed paper from the end of a doctorate in this index. A first paper from 2024 or 2025 points at a student in the middle years, which is the internship window.

Where are the doctoral students?

The table counts the researchers in each country whose record reads as early career, typically a doctorate in progress or recently finished. A country is where the institution sits, so a Chinese student at a Swiss university counts in Switzerland.

Three readings matter for an internship. The first is volume: China and the United States together account for just over half of the early career profiles with a known country, so an unfiltered search will be dominated by them. If you cannot host a student from one of them, say so in the search before you read a single profile.

The second is Europe. Germany, France and the United Kingdom come next, with 47 602 early career profiles between them, and Switzerland reaches a share of 55% on a smaller base. For a company in Europe these are the students who can join without a long visa process, and their university calendars are close to yours.

The third is a caution about France. Its 65% is the highest share of any large country, but part of that lead is a measurement effect. France publishes a national register of doctorates, theses.fr, which lists doctorates still in progress, so French doctoral students are easier to see than students elsewhere. Read it as good visibility as much as a larger pool. Each country report breaks its row down by subject and by institution, and the atlas compares any two countries side by side.

When should you start for summer 2027?

Later than most teams think is the usual mistake. A summer internship is a decision the student makes months ahead, usually with their supervisor, and often against offers from labs that open their programmes in the autumn. If your first message arrives in March, the students you wanted have mostly decided.

• October and November 2026. Write the project down: the question, the mentor, the dates, whether the work can be published. Then build the list of students on your subject.
• November to January. First messages. This is when students compare options and when a precise offer stands out from a generic one.
• January to March 2027. Conversations, a technical discussion of their own paper, and offers. Leave room for the supervisor to agree.
• March to May. Contracts, and visas where they apply. A student coming from another country needs this time, and the paperwork rarely moves faster than the calendar allows.
• May to September. The internship itself. Academic years end at different times in different countries, so agree the start and end dates with the student rather than imposing yours.

One more date to ask about: the conference deadlines the student is working towards. A student who has a submission due in the middle of your summer will either miss it or work on it on your time. Agree on it in advance.

How do you find candidates on your subject?

Interns are recent arrivals in a field, so the useful measure is newcomers: researchers whose first indexed paper falls in a given year. The comparison below uses full years only, 2024 and 2025 against 2021 to 2023, and it compares each subject's share of the world's newcomers rather than raw counts, because the most recent years keep filling in as the index catches up. A ratio above 1 means the subject draws a larger part of the people starting out than it used to.

• Explainable AI: about 1 540 newcomers a year in 2024 and 2025, against 1 084 a year from 2021 to 2023. Its share of the world's newcomers is 1.5 times what it was.
• Multimodal machine learning applications: about 2 971 a year against 2 437, a share 1.3 times higher.
• Generative adversarial networks and image synthesis: about 2 216 a year against 1 879, a share 1.3 times higher.
• Adversarial robustness: about 1 600 a year against 1 394, a share 1.2 times higher.

Other subjects hold their place. Graph neural networks sit at 1.1, while reinforcement learning in robotics and natural language processing techniques both sit at 1.0; the last is the largest subject on this page, with about 3 363 newcomers a year. Two subjects draw a smaller part of newcomers than before: domain adaptation and few-shot learning at 0.9, and advanced image and video retrieval at 0.6, down from about 1 355 newcomers a year to 762. A researcher can carry several subjects, so these counts overlap and should not be added up.

A subject label is a starting point, not a search. The method that works for interns has four steps:

• Describe the problem. Write what your intern would work on in plain words, the way you would explain it to a colleague, and search on that rather than on a category.
• Keep the early career. Keep the people whose record reads as early career and whose first paper is recent. That leaves mostly doctoral students in their first years.
• Read the papers. A first-authored workshop paper on exactly your problem is worth more for an internship than a middle-author slot on a famous one.
• Look at the co-authors. The supervisor and the lab tell you what the student has been trained to do, and whether your project is a natural next step for them.

The country filter comes before all of this when you cannot host from everywhere.

How do you reach a doctoral student?

Use the address the student printed on their own paper or lab page. It is the one they read, and it is the one they chose to make public. The index carries it where it exists, with the document it came from, and says so where none is public.

The first message should be short and specific. Name the paper you read and what you found interesting in it. Then give the facts a student needs to decide whether to reply: the question you want worked on, who would supervise them day to day, the dates, the place or the remote arrangement, and your policy on publishing. A message that could have been sent to anyone gets the reply it deserves.

Expect the supervisor to be part of the decision. In many labs an internship needs their agreement, sometimes the agreement of the funder too. Offering to talk to the supervisor about the project is often what turns a polite yes into a real one. Send one follow-up after two weeks, then stop.

What should the internship offer?

Students compare internships on a handful of things, and pay is only one of them. The offers that win usually give clear answers on each of the following:

• A real research question, small enough to finish in a summer and close enough to the student's thesis that the work counts for them.
• The right to publish, with the review process stated in advance. For a doctoral student a paper is the main thing a summer produces.
• A mentor who does research and has time for weekly meetings, named in the offer.
• Compute and data the student could not get at their university, described concretely.
• Pay and practical help: help with housing and relocation, and visa support where the student is moving.
• A path after the summer: a part-time collaboration during the year, or a conversation about a full-time role when the doctorate ends.

Whatever you offer, write it down in the first serious exchange. A student comparing two labs compares the two emails.

An internship is the earliest look at a future hire

The student who spends summer 2027 with you will probably defend in 2028 or 2029. By then they will be fielding offers from every lab that read the same papers. A good summer gives you the one advantage those labs lack: you already know how they work. Keep the end of their doctorate in view and set an alert on it, so the conversation about a full-time role starts six to nine months before the defence rather than after it. The arithmetic behind that window is in how long an AI doctorate takes, and where AI researchers are gives the wider picture by country.

Common questions

When should I start recruiting AI research interns for summer 2027?

In October or November 2026. Lab internship programmes typically open their applications in the autumn, and strong doctoral students usually settle their summer during the winter. First messages between November and January leave time for interviews, the supervisor's agreement and any visa before a start between May and September.

Where are most AI doctoral students?

In this index, China and the United States hold 53% of the early career researchers with a known country. Germany, France and the United Kingdom follow, and France has the highest early career share of any large country, at 65%, partly because its national thesis register makes doctoral students more visible.

Should I hire a master's student or a PhD student as a research intern?

For research work, a doctoral student in the middle of the degree is usually the better fit. They have published, they can run an experiment end to end, and they still have time before their defence. Master's students can be excellent but are hard to assess before they publish.

How do I find AI PhD students working on a specific subject?

Search the literature rather than job boards. Describe the problem in plain words, keep the researchers whose record reads as early career and whose first paper is recent, then read their papers and look at their co-authors. A first-authored paper on exactly your problem is the strongest signal.

What makes a research internship offer attractive to a PhD student?

A research question that fits their thesis, the right to publish, a named mentor who does research, compute they cannot get at their university, fair pay with practical help on housing and visas, and a clear path to keep working together after the summer.