NSF GRFP Research Plan: Two Pages, Both Criteria

NSF GRFP Research Plan: Two Pages, Both Criteria

The Graduate Research Plan Statement is the shorter of the two GRFP documents and the one applicants most often write in the wrong genre. Under NSF 25-547, "the maximum length of the Graduate Research Plan Statement is two (2) pages (PDF)," and those two pages "include all references, citations, charts, figures, images, and lists of publications and presentations."

Two pages is not a grant proposal, and NSF says so outright. From the program's own FAQ:

"The research statement in your application is not considered to be a proposal that you are obliged to carry out; it is used to evaluate the evidence of your potential."

That single sentence should change how you draft. You are not committing to a project. You are giving a panel two pages of evidence that you can find a real question, choose an approach that could settle it, and say what would count as an answer.

Verification note. Every NSF rule quoted here is from NSF 25-547, the GRFP solicitation posted September 26, 2025, which states that it "covers the Fiscal Year (FY) 2026 competition." It was still the solicitation linked from NSF's GRFP program page when this guide was retrieved on August 19, 2026; no successor solicitation was posted as of that date. Page limits, formatting rules, eligibility, and deadlines are solicitation-versioned. Re-verify every one of them against whichever solicitation is live when you apply — the numbered solicitation governs, not this page.

Table of Contents

• What NSF 25-547 requires, mechanically
• Not a grant proposal: what the document is actually for
• Who reads it, in NSF's own words
• Four jobs, two pages
• Failure one: written for the six people in your subfield
• Failure two: Broader Impacts bolted on at the end
• The eligibility screen that runs on this document
• GRFP, NDSEG, Hertz: three documents from one body of work
• A self-check for your own two pages
• Where an outside read fits
• Frequently asked questions
• Sources

What NSF 25-547 requires, mechanically

Two of those deserve a second look, because they change what a two-page budget actually buys.

Figures and references are inside the limit. A half-page schematic costs you a quarter of the document. That is sometimes the right trade — but it is a trade, not a free addition, and applicants who lay out a figure first and write into the remainder usually end up cutting the sentence that says what result would count.

A reference section is not required. The GRFP FAQ is direct about it: "No. A reference section is not required, although your reviewers may expect to see references if your statement cites background work needed to convincingly motivate your research plan." Read that as a conditional, not a permission slip. If your first paragraph asserts that the field has not resolved something, a reader will want to see what you read to know that.

One mechanical mercy: NSF 25-547 says compliance "will be automatically checked by the GRFP Application Module," and "documents that are not compliant will not be accepted" — but an automated check that rejects your file at 4:50 p.m. is not a safety net, which is why the same paragraph tells applicants to "proofread and upload their documents early." The full formatting surface across both statements — every rule, and everything "returned without review" covers — is collected rule by rule in the GRFP formatting rules guide.

Not a grant proposal: what the document is actually for

The most expensive misreading of this document is treating it as a miniature NSF grant. Applicants who have read a real proposal import its furniture — a budget-shaped work plan, aims numbered like specific aims, a risk-mitigation paragraph, a timeline table — and spend a page of two on scaffolding that answers no question the panel was asked.

NSF has said what the document is for — the FAQ answer quoted at the top of this page, given to applicants who do not yet know where they will attend graduate school. Take both halves of that sentence seriously.

Not obliged to carry out means the plan is not a contract about a project. You do not have to have the funding, the collaborator, the sample, or the advisor. You do not have to have chosen the lab. The same FAQ answer adds, in the same breath, "You are not obligated to attend the proposed institution listed in your application."

"Evidence of your potential" means the plan is a demonstration, and the thing being demonstrated is a capacity: can this person locate an unresolved question, choose an approach that could actually settle it, and recognize what an answer would look like? That is a skill you display by exercising it on two pages, not one you can assert.

There is a real limit on the freedom, and it is worth stating precisely rather than letting the word "non-binding" do too much work. The project is not binding; the field largely is. If awarded, you are required to enroll in a degree program aligned with the field you proposed, and you cannot change your Major Field of Study during the first Fellowship year — the sponsor's exact wording on both rules is quoted in the FAQ below.

So: propose the work you would actually want to do, in the field you actually intend to enter, and stop worrying about whether you will run exactly these experiments. The right level of specificity is the level at which a reader can tell you understand what the work would involve — not the level at which you could hand the document to a lab manager.

Who reads it, in NSF's own words

Applicants routinely picture a reader who knows their subfield. NSF describes a different reader, twice.

On panel composition, NSF 25-547 says: "The application evaluation involves the review and rating of applications by disciplinary and interdisciplinary scientists and engineers, and other professional graduate education experts."

On assignment, the same solicitation says: "Applications are reviewed in broad areas of related disciplines based on the applicant's selection of a Field of Study... Selection of a Major Field of Study determines the application deadline and the broad disciplinary expertise of the reviewers."

The FAQ closes the loop for interdisciplinary applicants: asked whether selecting multiple fields means multiple sets of reviewers, NSF answers "No. Your application will be reviewed in the Major Field of Study that you select... Your reviewers will be drawn from experts within that field."

Put those together and the reader is well defined. Not a random generalist — someone with real training in your field. Also not your subfield's fifteen specialists, not someone who knows your model system, not someone who will recognize your group's acronyms. "Broad areas of related disciplines" is the sponsor's phrase, and it is the design constraint on every technical sentence you write.

That is also why the panel is reading for capacity rather than for project merit. A reader with field-level expertise cannot adjudicate whether your specific hypothesis about a niche mechanism is the best available one. They can absolutely tell whether you know what an unresolved question is, whether your approach could produce evidence about it, and whether you can explain a technical decision to a colleague one door down the hall.

Four jobs, two pages

Strip the genre furniture and four jobs remain. Every sentence in the document should be doing one of them.

1. The question, and why it is open. Not the topic — the question. "I will study lithium dendrite formation" is a topic; the question is what specifically is not known and why previous work has not settled it. This is the paragraph that most often gets replaced by background, and background is the cheapest thing on the page: a field-level expert already has it.

2. The approach, at a level your field can follow. What you would do, to what, measured how. The test is not whether the method is impressive; it is whether a reader can trace a line from the procedure to the evidence it produces.

3. What would count as an answer. The single most commonly missing sentence in GRFP research plans is the one that says what result would mean what. A plan that describes only activity — data will be collected, models will be fit — reads as work rather than as inquiry. A plan that says "if segregation sets the threshold, the measured value will track boundary composition rather than bulk conductivity" reads as someone who has thought about being wrong.

4. Feasibility at graduate scale. Two pages is not room for a timeline table. It is room for the plan to be visibly the size of a dissertation rather than the size of a career. If the work as written would take a funded lab a decade, the document quietly says its author cannot yet judge scope — which is exactly the capacity being read for.

On top of those four, the two required criterion sections sit under their own headings. They are not a fifth and sixth job bolted to the end; done properly the Intellectual Merit section states what the expected results would add to knowledge, and the Broader Impacts section states a route from a result to people. What each criterion has to prove, in both GRFP documents, is worked through in Intellectual Merit vs Broader Impacts.

Failure one: written for the six people in your subfield

This is the most common failure in a GRFP research plan, and it is invisible from the inside, because the draft reads beautifully to you and to your labmates — the only two audiences who have seen it.

NSF's definitions describe what is at stake. "The Intellectual Merit criterion encompasses the potential to advance knowledge," and reviewers assess "the applicant's potential to advance knowledge and understanding in one or more fields." A reader who cannot follow your approach cannot form a view about your potential to advance anything. Density does not read as rigor to someone outside your subfield. It reads as noise, and noise is scored by nobody.

Constructed teaching material — written for this guide, not an excerpt from any real application, and not modeled on anyone's submitted work.

Specialist-only draft:

"Building on prior characterization of the regulator deletion strain, I will apply transposon-insertion sequencing across a nutrient-shift panel to resolve whether the general stress regulon's contribution to the stationary-phase transcriptome is epistatic to the nutrient-sensing node, quantifying fitness contributions as log-fold change against the wild-type reference under the conditions described above."

Everything in that sentence is true and precise and lands on nobody. Now the same science, same specificity, written for a reader with field-level training:

Rewritten for the reader NSF describes:

"Bacteria survive starvation by switching on a stress program, and two separate control systems can switch it on: a general stress regulator and a nutrient-sensing regulator. Which one is in charge when a cell loses its preferred carbon source is unresolved, because the two systems have almost always been studied in isolation. I will delete each regulator in turn and measure how well tens of thousands of mutants survive a controlled nutrient shift, using a transposon-sequencing screen that reports the fitness cost of every gene at once. If the general stress regulator acts downstream, deleting the nutrient sensor should erase its effect; if the two act in parallel, the double mutant will do worse than either single mutant."

Nothing technical was removed. The method is still named, the measurement is still named, the comparison is still named. What changed is that the second version supplies the one clause of context each technical term needs, and it ends with the sentence the first version never reaches: what result would count as which answer.

The mechanical version of this fix: find every noun phrase in your draft that you have never had to define out loud, and give it a five-word gloss. If the glosses push you over two pages, you have too many of them, which is itself the finding — a plan resting on six subfield-specific constructs is a plan a field-level panel cannot evaluate.

The related discipline, at a different length and for a different sponsor, is in our PhD research proposal guide, where the proposal is the primary admissions document rather than one of two statements.

Failure two: Broader Impacts bolted on at the end

The second failure is structural, and the structure invites it. Because NSF requires a separate Broader Impacts section under its own heading, the path of least resistance is to write the science, hit the bottom of page two, and append a paragraph. The result is recognizable across thousands of drafts: an appeal to downstream applications plus a promise of future outreach.

NSF's definition rules out both moves. "The Broader Impacts criterion encompasses the potential to benefit society and contribute to the achievement of specific, desired societal outcomes." And the factors reviewers consider are framed as "the likelihood that the applicant's activities will" produce them — a list of named outcomes running from increasing participation in STEM to enhancing infrastructure for research and education, quoted in full in Intellectual Merit vs Broader Impacts.

Two words in that definition do the work. Specific rules out "applications in medicine, agriculture, and industry." Likelihood rules out an intention with no track record behind it: the reviewer is being asked to estimate whether it will happen, and the only evidence available for that estimate is what you have already done.

Constructed teaching material — written for this guide, not an excerpt from any real application.

Bolted-on version:

"Broader Impacts. This research will benefit society by advancing our understanding of microbial systems, which has applications in medicine, agriculture, and biotechnology. I am also passionate about outreach and plan to participate in outreach activities and mentor undergraduate students in order to broaden participation in STEM."

Three diagnosable problems. There is no specific outcome, only a direction. There is no action already taken, so there is nothing for a reviewer to estimate likelihood from. And nothing in it connects to the research described on the page above — it would sit equally well under any research plan in the field, which is the definition of a bolted-on paragraph.

Rewritten from the research's own output:

"Broader Impacts. Screens of this kind are usually published as supplementary spreadsheets that only groups with a bioinformatics pipeline can reuse, so the data reaches a fraction of the labs that could use it. In my second undergraduate year I rewrote our group's screen-analysis scripts into a documented command-line tool and taught it in two sessions to eleven students at the campus makerspace; six of them now use it in their own projects. I will release the nutrient-shift dataset alongside that tool and a worked tutorial, and run the same sessions through the university's summer bridge program for incoming transfer students. The outcome I am aiming at is concrete: a lab with no computational staff member can reanalyze this dataset in an afternoon."

That version passes the tests the first one fails. It names a specific desired outcome. It rests on an action already completed, with a result attached, so "likelihood" has something to stand on. And it grows out of the research plan directly — it could not be pasted under anyone else's project, which is the strongest available signal that it was not bolted on.

The general rule: write Broader Impacts from your own output, not from a menu of good causes. Ask what your research produces — a dataset, a method, a tool, a measurement, a piece of instrumentation, a body of trainable technique — and then who cannot currently get at it, and what you have already done about that. Then check the seam. If the Broader Impacts section would still make sense with the Intellectual Merit section deleted, it is decoration.

The mirror image of this diagnosis, from the recommender's chair, is in the GRFP letter mistake: all merit, no broader impacts; the letter-side conventions are in the GRFP reference letters guide.

The eligibility screen that runs on this document

Here is the part almost no GRFP guide mentions, and it belongs specifically to the research plan rather than to the personal statement.

NSF publishes an additional FAQ on applications returned without review — NSF 26-205, "Additional FAQ: NSF Graduate Research Fellowship Program (NSF GRFP)", published February 27, 2026 and last updated April 13, 2026. Two of its answers matter here.

First, on the scale of the problem: asked whether NSF has returned applications without review in previous years, the answer is "Yes. Each year, hundreds of applications are determined to be ineligible for the reasons described in Section IV of the solicitation and are returned to applicants without review."

Second, and more pointed — asked "Are Applicant Personal Statements screened to determine eligibility?", NSF answers:

"No. NSF screened the Graduate Research Plan of the application to determine if an applicant's proposed research was eligible."

Your two-page research plan is the document that establishes whether your proposed research is eligible at all. The same FAQ lists the reasons an application may be returned — "including failure to meet formatting requirements, missing required documents or if the applicant does not meet the eligibility criteria related to academic status, degree program, field of study or proposed research" — and notes that "within eligible fields of study, there are ineligible areas of study and ineligible areas of proposed research."

NSF 25-547 spells out the research-side exclusions, and the pattern across its own examples is goal versus implication. Summarizing the solicitation: research whose goals are "directly human disease- or health-related" — including etiology, diagnosis, treatment, or interventions for physical or mental disease — is not eligible; basic questions in plant pathology are eligible while applied studies aimed at maximizing agricultural production or food safety are not; and research "with implications that inform policy" is eligible while research "with the expressed intent to influence, advocate for or effect specific policies and outcomes" is not.

The solicitation also records a limited exception: "Certain areas of bioengineering research directed at medical use are eligible... provided they apply engineering principles to problems in medicine while primarily advancing engineering knowledge," with biomedical engineering as the Field of Study. And NSF 26-205 records a change: "To be eligible, an applicant must not have completed more than one academic year in a graduate program. In addition, clinical psychology degree programs are no longer eligible."

The writing consequence is narrow but real, and it is easy to get wrong in the wrong direction. This is not an instruction to hide a health connection, strip the word "disease" from a draft, or dress applied work in basic-science vocabulary. NSF says the screening is done by "multiple NSF program officers with subject-matter expertise," and describing your work as something it is not is a worse problem than ineligibility. The consequence is the opposite: state the actual goal of the research plainly, early, and in the terms NSF's categories use, so that the eligibility question is answered on your page rather than inferred. If the knowledge goal is mechanistic and the health relevance is downstream context, say which is which. If you genuinely cannot tell which side of the line your project falls on, that is an advisor conversation and a question for the GRF Operations Center before the deadline — not a drafting problem.

If your work is health-facing by design, the eligible route is a different program. Our MD-PhD essay guide covers the research-plus-medicine application on the AMCAS side.

GRFP, NDSEG, Hertz: three documents from one body of work

Applicants who enter the fall STEM fellowship season usually have one research story and three sponsors. The documents are not interchangeable, and the differences are structural rather than stylistic.

Three specifics are worth having exactly right.

NDSEG's redaction rule is disqualification-grade, and it is stated by the sponsor like this. From the NDSEG application FAQ, retrieved August 19, 2026, quoted in full:

"Redaction is defined as no inclusion of any identifying information, such as your or any person(s) names, organization/institution names, or specific details that could tie the application back to you and any institution/organization or person. Failure to follow this guidance will prevent your application from moving forward in the evaluation process."

The mechanics are in the NDSEG research proposal and redaction guide; the point for a GRFP writer is the direction of travel — a GRFP research plan names your university, advisor, lab, and collaborators as ordinary practice, and every one of those names has to come out before the same science can go to NDSEG. Rewrite; do not paste.

Hertz asks for no research proposal at all, and the Foundation says why:

"The fellowship is intended to fund individual students, not a specific project. There is no requirement to submit a proposed research project as part of the application process. However, the application does provide the opportunity to summarize current and prior research experiences as well as interests you intend to pursue."

A compressed GRFP research plan submitted into a Hertz short essay is a strong answer to a question nobody asked. Hertz's published word ranges are suggestions you "may go under or over" — the opposite of NSF's automatically enforced page limit — and the current prompts live in the Foundation's own portal, not in any secondhand list.

The three-way sequencing problem — which document to write first, what evidence transfers, and where the calendars collide — is laid out in GRFP, NDSEG, Hertz: apply to all three in one season.

A self-check for your own two pages

Run this on your own draft, in this order. It is a reading pass, not a rewrite pass — the point is to find out what the document currently does, before you decide what to change.

1. Highlight the question. Find the single sentence that states what is not known. If you cannot find one sentence, that is the first fix, before anything else on this list.
2. Highlight the result clause. Find the sentence that says what outcome would mean what. If the plan describes only activities, add it and cut background to pay for it.
3. Count the undefined terms. Mark every noun phrase a well-trained person in your field, but outside your subfield, would have to look up. More than a handful and you are writing for the wrong panel.
4. Delete the Intellectual Merit section and reread. If nothing is missing, that section was restating the plan rather than saying what the results would add to knowledge.
5. Delete the Broader Impacts section and reread. If the plan is unaffected, the section was bolted on. Rebuild it from what the research produces.
6. Find the completed action inside Broader Impacts. Something you have already done, with a result attached. If everything is future tense, a reviewer has nothing to estimate likelihood from.
7. Read for scope. Would this be a dissertation, or a career? Say the honest answer out loud.
8. Check the mechanical contract last. Two pages including everything; "Intellectual Merit" and "Broader Impacts" as separate sections, under individual headings, separated by a line break, with no other terms in the headings; no URLs; journal names in any citations.

The order matters. Applicants who start at step 8 tend to finish there, and formatting compliance is the one part of this document that the application module will tell you about by itself.

Where an outside read fits

Your advisor owns the judgments nobody else can make: whether the question is genuinely open in your field, whether the approach could actually produce the evidence, whether the scope is a dissertation, and whether the work sits on the eligible side of NSF's research categories. NSF 25-547 points the same way, advising applicants "to consult a faculty member, academic advisor, mentor, or other knowledgeable source for guidance on preparation of their research plans, selection of Major Field of Study, and identification of subfields."

Advisor attention is finite, and it is worth protecting from the first draft's obvious problems — the plan with no result clause, the Broader Impacts paragraph that would fit any project in the field, the three paragraphs of background that crowded out the question.

That first pass is what a rubric review is for. On the text you paste in, the NSF GRFP research plan rubric reads for whether there is a stated question rather than a topic, whether the approach connects to the evidence it would produce, whether the plan says what result would count, whether the scope reads as graduate-scale, and whether the Intellectual Merit and Broader Impacts sections each carry their own argument rather than restating the plan. NSF 25-547 also requires applicants to certify that both statements "are their own original work" — so the useful shape of outside help is the one that keeps you the author: you write it, then you check it against published criteria and revise it yourself.

The boundary, stated plainly: a review reads only the text you paste. It cannot see your PDF, count your pages, check your margins or type size, confirm that your headings are separated by a line break in the exported file, or tell you whether NSF will consider your proposed research eligible. Those stay with you, your advisor, and the solicitation. Feedback is a revision signal, not a prediction about your application.

Two Quick Reviews are free each day; your first Full Review is $5, once per account, and Pro is $50 for ten Full Reviews. What a review does and does not do is set out in how a scholarship essay review reads a draft, and the NSF GRFP application review page lists both GRFP rubrics side by side.

Frequently asked questions

The mechanical questions — length, headings, references, who reviews — are answered in the table above and in Who reads it; the three questions below are the ones the table cannot answer.

Do I have to do the research I propose?

No. The GRFP FAQ says "the research statement in your application is not considered to be a proposal that you are obliged to carry out; it is used to evaluate the evidence of your potential," and adds "you are not obligated to attend the proposed institution listed in your application." The field is a different matter: NSF 25-547 says that if awarded, "Fellows will be required to enroll in a degree program most closely aligned with the proposed graduate program of study and research plan in the application," and the FAQ answers that a Fellow may change Major Field of Study "not in the first year of the Fellowship" — only after completing it.

Is the research plan the document that determines eligibility?

For the proposed-research part of eligibility, yes. NSF's additional FAQ (NSF 26-205) answers the question directly: "No. NSF screened the Graduate Research Plan of the application to determine if an applicant's proposed research was eligible" — the "no" refers to personal statements, which are not screened for this. The same FAQ notes that "each year, hundreds of applications are determined to be ineligible."

When is the GRFP deadline?

That depends on the solicitation in force and on your field, and it changes cycle to cycle. NSF 25-547, which covers the FY2026 competition, published field-specific application deadlines of November 10, 12, 13, and 14 of 2025, with a reference-letter deadline that fell earlier — see the GRFP formatting rules guide for that competition's full letter calendar. Those are that competition's published dates, not a forecast for any future cycle. Check the current solicitation and the GRFP program page for live dates; our GRFP cycle guide tracks what changes when a new solicitation posts.

More GRFP and STEM fellowship guides

• Intellectual Merit vs Broader Impacts
• NSF GRFP formatting rules: returned without review
• NSF GRFP cycle guide
• NSF GRFP reference letters
• The GRFP letter mistake: all merit, no broader impacts
• NDSEG research proposal and redaction guide
• NDSEG fellowship essays guide
• GRFP, NDSEG, Hertz in one season
• PhD research proposal guide
• How to write a scholarship essay that fits the award

Review your draft

When the two pages are written, paste them into the scholarship essay review and run them against the NSF GRFP research plan rubric. Reviews are of your own writing, and feedback is a revision signal rather than a verdict on your application. The solicitation in force governs everything on this page.

Sources

NSF sources retrieved August 19, 2026 from the solicitation and FAQs then live. NSF 25-547 covers the FY2026 competition; no successor GRFP solicitation was posted as of that date. Re-verify every rule against the current solicitation before you submit.

• NSF 25-547: NSF Graduate Research Fellowship Program (GRFP) — posted September 26, 2025; replaces NSF 24-591; page limits, formatting rules, heading and line-break requirements, return-without-review consequences, merit review criteria and factor lists, panel composition and field-of-study assignment, eligibility and ineligible research areas, due dates, original-work certification
• NSF GRFP program page — current solicitation shown as NSF 25-547
• FAQ: Graduate Research Fellowship Program (GRFP), NSF 23-154 — the research statement is "not considered to be a proposal that you are obliged to carry out"; institution not binding; Major Field of Study review and reviewer expertise; reference section not required; first-year field-change rule
• Additional FAQ: NSF Graduate Research Fellowship Program (NSF GRFP), NSF 26-205 — published February 27, 2026, last updated April 13, 2026; return-without-review reasons; the Graduate Research Plan as the screened document; ineligible proposed-research examples; eligibility change
• NDSEG — Application FAQs — the redaction definition and its consequence, quoted in full above
• NDSEG — Application, Evaluation and Award — proposal page counts and format, dual redacted-and-original upload, DoD strategy document alignment
• Hertz Foundation — Application FAQ — no required research project proposal; suggested essay lengths
• Hertz Foundation — Apply — published application timeline and award value

All illustrative passages in this guide were written for it. They are not excerpts from any applicant's essay, real or submitted, and no real application should be copied from any source.