A 3-Page Proposal, a 28-Day Window, and a 10% Indirect Cap: OpenAI's $5M Teen Research Fund Closes October 6

Three pages. That is the entire research proposal OpenAI wants for a grant that can run to $1 million.

For anyone whose professional reflex was built on the NIH R01 — a 12-page Research Strategy, a Specific Aims page workshopped for a month, nine months from submission to Notice of Award — the AI and Teen Development Research Grant Program is a different animal entirely. Applications opened September 8, 2026 and close October 6, 2026 at 11:59 p.m. Pacific. Decisions go out on or before November 13. Start to funded is roughly nine weeks.

That speed is the program's most interesting feature and its biggest practical trap, and the trap is not the science. It is the 10 percent overhead cap.

What is on the table

OpenAI has committed $5 million total, with individual awards up to $1 million. The subject is narrow and specific: how generative AI affects the lives and development of young people ages 13 to 17.

The solicitation names four priority areas:

• Emotional development — emotion regulation, resilience, identity, agency, self-esteem, and social connection.
• Social development — friendships, family relationships, belonging, communication, empathy, social skills, and the relationship between AI use and human connection.
• Demographic variance — how outcomes differ by usage pattern, age, culture, socioeconomic status, and other factors.
• Mitigations and policy — safeguards, product interventions, AI literacy, and healthy-use practices.

Eligibility is deliberately wide. Applicants must be 18 or older and either affiliated with a research institution or organization or able to demonstrate significant experience in child and adolescent development, well-being, human-computer interaction, AI impacts, or an adjacent field. Multi-institutional collaborations are permitted. Students and post-doctoral researchers can be funded on a project. For-profit entities are explicitly not prioritized — the program wants research, not product development with a research label.

The required package is short: a project summary under one page, a research proposal of no more than three pages excluding references, a timeline, a budget justification, a statement of project dependencies, an ethics statement, team information with CVs, and a conflict-of-interest disclosure. English, submitted as a Google Doc, through OpenAI's SM Apply portal. Questions go to collaborativeresearch@openai.com.

Review is rolling, by internal OpenAI researchers plus an outside advisory panel, against ten stated criteria: scientific rigor, relevance to current AI discourse, proposal clarity, actionability of findings, ethics protections, feasibility, team expertise, inclusion, complementarity with existing research, and independence from bias.

The output obligation is a working draft suitable for discussion at a small conference within 12 months. Not a published paper. A draft, in a year.

The indirect-cost cap is the real eligibility filter

Here is the sentence most academic applicants will skim past: a maximum of 10 percent of each project's grant can support overhead or indirect costs.

Run the math on a $1 million award. At 10 percent, roughly $909,000 is direct and $91,000 is indirect. A research university with a federally negotiated F&A rate of 55 to 62 percent on modified total direct costs would ordinarily expect something closer to $375,000 on the same package. The gap is not a rounding error. It is a six-figure institutional subsidy that someone has to approve.

Most universities do have a path for this — a foundation-rate policy, a private-sponsor exception, or a formal F&A waiver request routed through the Office of Sponsored Programs or the VP for Research. What they do not have is a path that clears in four days. Waiver requests routinely take two to four weeks and often require a dean's countersignature.

So the operational sequence for a university-based PI is not "write the proposal, then route it." It is:

1. Today: email your sponsored programs office with the sponsor name, the 10 percent cap, the ceiling amount, and the October 6 deadline. Ask specifically whether the institution has a standing reduced-rate policy for corporate and foundation sponsors, and if not, what a waiver requires.
2. In parallel: build the budget at the size your institution will actually approve, not the size the ceiling permits. A $250,000 project that clears internal review beats a $900,000 project that misses the deadline in a routing queue.
3. Only then: write the three pages.

Independent researchers, think tanks, school-based research shops, and nonprofits with in-house research capacity have a genuine structural advantage here. A 10 percent overhead line is close to normal for a mid-size nonprofit and catastrophic for a large R1 budget office. That asymmetry is almost certainly intentional — it widens the applicant pool beyond the same dozen university labs that already dominate adolescent-technology research.

Why a private company is funding this at all

The honest context matters, because it should shape how you write the proposal.

OpenAI is funding independent research into the effects of its own product category on minors at a moment when that exact question is in front of regulators, state attorneys general, and courts. "Independence from bias" appearing as an explicit review criterion is the program acknowledging the obvious. The conflict-of-interest disclosure requirement is the same acknowledgment in procedural form.

Treat that as an opportunity rather than a disqualifier, but get the terms in writing. Before you accept funds, confirm three things with the sponsor and with your own institution's contracting office:

• Publication rights. Who decides whether and when results are published, and is there a sponsor review window? A courtesy review period of 30 days is standard and acceptable. A veto is not.
• Data ownership and pre-registration. If you intend to pre-register hypotheses and analysis plans — and for a study in this domain, you should — say so in the proposal. It is the cheapest credibility signal available and it hardens the finding against the "industry-funded" discount before anyone applies it.
• What "dependencies" means for you. The application asks for a project dependencies statement. That field is where you disclose whether the study requires anything from OpenAI — model access, usage data, a research API, cooperation on recruitment. Projects with a hard dependency on sponsor-supplied data are more fundable and less independent. Projects built on your own recruitment and instruments are more independent and slower. Know which trade you are making and name it explicitly rather than leaving the reviewer to guess.

The funding gap this is filling

There is no federal program that funds this question at this speed.

NIH has no dedicated mechanism for AI-and-adolescent-development research; the closest routes are behavioral science study sections where a topic this new competes against established programs of work and takes the better part of a year to reach council. The Institute of Education Sciences has been running under severe constraint. NSF's relevant programs are oversubscribed and are contending with the same end-of-fiscal-year obligation crunch affecting the rest of the federal research portfolio.

Meanwhile the private AI money keeps arriving on a different clock. OpenAI's nonprofit arm ran the People-First AI Fund at $50 million earlier this year, and the ten-foundation Humanity AI collaborative has a $10 million open call closing October 21. The pattern is consistent: short windows, light applications, fast decisions, and terms that trade institutional overhead for speed.

That is a real shift in the grant-seeking calendar. Federal deadlines are published a year out and reward long preparation. This tier of private AI funding publishes with four weeks of notice and rewards teams who already have an instrument, an IRB relationship, and a study design sitting in a drawer.

How to spend the four pages you get

With a one-page summary and three pages of proposal, there is no room for a literature review that establishes your general competence. Assume the reviewer knows the field. Spend the space on:

A falsifiable question, stated in the first two sentences. "How does companion-style AI use relate to friendship quality among 14-to-16-year-olds over six months" is a proposal. "Exploring the complex landscape of teen AI use" is not.

The design, in enough detail to judge feasibility. Sample, recruitment source, measures, comparison condition, analysis plan, power. Reviewers are scoring feasibility and rigor against a 12-month working-draft deadline — a three-wave longitudinal panel you cannot possibly field by next September will score badly on feasibility no matter how good the question is.

Ethics, concretely. You are proposing research on minors. Name the IRB, state its status, describe parental consent and youth assent procedures, and say what happens if a participant discloses self-harm. "Ethics protections" is a scored criterion and a generic paragraph will read as a red flag.

Actionability. The criteria list "actionability of findings" and "relevance to current AI discourse." State plainly what a product team or a policymaker could do differently if your result comes back one way versus the other. If nothing would change either way, the proposal is not competitive here regardless of its scientific merit.

Complementarity. They are scoring whether your project duplicates existing work. One short paragraph naming the two or three closest studies and stating what yours adds is worth more than a page of background.

The window is four weeks wide and the indirect-cost conversation with your institution is the long pole. If your research question is already formed, the constraint on whether you apply is administrative, not intellectual — which is exactly why it pays to start the routing conversation before you write the first paragraph. When the deadline is measured in weeks rather than quarters, tools like Granted earn their keep by getting a rough concept into submission-ready shape while the window is still open.