NSF Intellectual Merit 2026: Criteria & Writing Guide

Full answer: NSF Intellectual Merit tests knowledge advance; in 2026, the 5 shared review elements apply to both it and Broader Impacts. Both criteria receive full consideration; neither substitutes for the other. Use the evidence map below to connect your scientific claim to a method, assessment, team, and resources.

This guide is for: Researchers rewriting an NSF proposal, graduate applicants adapting criterion language to their solicitation, and research offices checking whether a draft's claims are supported.

Last updated: October 6, 2026. Start with the NSF PAPPG hub, then the current solicitation. The hub lists PAPPG 24-1 with supplements; a draft guide is not a replacement for the policy in effect.

Data note: Source review completed September 27, 2026. The two-criteria answer, the GRFP definitions in NSF 26-526, and the National Science Board recommendations were checked on October 1, 2026, and the PAPPG version in effect was rechecked on October 6, 2026. Rules come from the linked current primary sources; the worksheets are Grantsights editorial tools. Examples are hypothetical and do not establish award outcomes or approval decisions.

NSF Intellectual Merit 2026: What It Is, How It's Scored, and How to Write It

What Are the Two Main Criteria NSF Uses to Evaluate Grant Proposals?

Intellectual Merit and Broader Impacts. That's the answer to the common quiz question, and it comes straight from PAPPG Chapter III.A: "Both criteria are to be given full consideration," and "each criterion is necessary but neither, by itself, is sufficient." NSF defines Intellectual Merit as "the potential to advance knowledge" and Broader Impacts as "the potential to benefit society and contribute to the achievement of specific, desired societal outcomes."

If you're choosing among answer options, Grantsights checked the usual distractors against PAPPG Chapter III.A on October 1, 2026:

Behind the two criteria, NSF lists three merit review principles: projects should "be of the highest quality and have the potential to advance, if not transform, the frontiers of knowledge," should "in the aggregate" contribute to societal goals, and should be assessed with appropriate metrics. Reviewers then apply five shared elements to both criteria, covered in the next section.

Two 2025-2026 developments haven't changed the answer. NSF's statement of priorities, first issued April 18, 2025, says "NSF continues to review all projects using Intellectual Merit and Broader Impacts criteria." And the National Science Board's merit review report recommends renaming Broader Impacts to "Societal Benefits" while keeping the definitions of both criteria. That rename is a recommendation, not policy; PAPPG NSF 24-1 still uses Intellectual Merit and Broader Impacts.

Intellectual Merit in the GRFP Solicitation

The Graduate Research Fellowship Program uses the same two criteria with a different subject: the applicant, not a project. NSF 26-526, the solicitation for the 2027 competition posted August 25, 2026, gives the GRFP definitions and guidance in Section VI.A.

The practical difference: a faculty proposal argues that a project will advance knowledge, while a GRFP application argues that a person will. The research statement still needs a real question, but reviewers read it as evidence of the applicant's ability to plan research, so the strongest GRFP Intellectual Merit material is concrete: what you designed, what went wrong, and what you did next. The solicitation's list says "may include (but are not limited to)," so treat it as a guide to evidence, not a checklist to quote back. For who has won in past cycles, see the GRFP winners and award history.

1. Five Review Elements Are Shared, Not Five Separate Merit Scores

According to NSF's merit-review guidance, reviewers consider both National Science Board-approved criteria and any additional criteria stated in the solicitation. According to PAPPG Chapter III.A, five elements guide evaluation of both criteria.

These are not a universal five-point grading sheet. They are not exclusively Intellectual Merit questions, and they do not imply that adding five headings produces a high rating. The draft must give the reviewer enough evidence to evaluate the work.

According to NSF Chapter III, both criteria receive full consideration, each is necessary, and neither is sufficient. This is not a numeric 50/50 allocation and not permission to compensate for a missing benefit argument with stronger methods. Keep the Broader Impacts plan connected to the project without collapsing it into the scientific question.

The practical source-delta is that the older “five Intellectual Merit sub-scores” framing confuses shared review elements with a fixed scoring formula. Grantsights analysis of the policy wording instead treats these elements as an evidence map: each important claim should have support the reviewer can locate.

2. The Project Summary Is One Page, Not the Whole Argument

According to PAPPG Chapter II.D.2.b, the Project Summary is no more than one page and includes an overview, an Intellectual Merit statement, and a Broader Impacts statement.

That one-page constraint creates an editing problem. A summary cannot contain the entire rationale, every method, and all the team's credentials. It should state the central question, the contribution, the approach, and the benefit clearly enough that the later proposal can supply the support.

For GRFP, read NSF 26-526 and the GRFP section above. Its personal statement and research plan have their own required Intellectual Merit and Broader Impacts sections. Do not transfer those exact formatting instructions to every NSF grant without checking the relevant rule.

Keep a policy-version line in your working notes: solicitation number, PAPPG version, applicable supplement, and access date. If two instructions seem inconsistent, resolve the precise requirement with the official program contact. Do not hold an entire proposal rewrite while leaving clearly fixable claims unchanged.

The Claim-to-Evidence Worksheet

What the official guidance does not show is a proposal-specific map between a scientific sentence and the evidence elsewhere in your draft. Our analysis uses this original worksheet to find gaps before polishing prose.

For each main claim, record the following:

This is an editorial tool, not an NSF submission form or an extra mandatory attachment. Keep it in your working files. Its value comes from the relationships it makes visible, not from presenting a score that NSF never assigned.

Begin with the claim in the summary. Follow it into the research plan. Then check the relevant personnel, facilities, collaboration, budget, and data-plan sections. A reviewer should not need to guess how those parts connect.

If a dependency is unresolved, narrow the claim or describe a feasible fallback. Do not convert a possible data source into secured access. If a collaborator has not agreed to a role, the draft should not portray that role as confirmed. Read the NSF budget guide alongside this worksheet when resources and task effort affect feasibility.

A Hypothetical Rewrite, Not a Funded-Proposal Story

For example, one research team might write: “We will use a new model to improve predictions of material behavior.” This sentence states a broad goal but does not show which prediction fails today, what change the model tests, or how improvement will be judged.

A more informative hypothetical version is: “We will test whether representing boundary conditions explicitly reduces prediction error for the material class studied here. We will compare the proposed model with the stated baseline on a held-out dataset and examine cases where neither model explains the observed behavior.”

The example is invented to demonstrate structure. It is not a quotation, a successful application, or evidence of a funding outcome. Replace its details with your actual research design.

The rewrite supplies a question, a comparison, and an assessment. It still needs a justified baseline, access to data, a credible team, and a reason the result matters scientifically. More precise prose does not cure a weak design.

Use three passes rather than one:

1. Identify the missing scientific or feasibility evidence.
2. Correct the design or narrow the claim.
3. Rewrite the sentence after the evidence is established.

A common mistake is to start at step three. Replacing vague adjectives with technical nouns may make a claim sound better while leaving its method or resource dependency unresolved. The worksheet catches that failure because the method and assessment columns remain empty.

Distinguish Knowledge Advance From Project Activity

Activity is not automatically an advance in understanding. Collecting samples, developing software, running interviews, or building a dataset may be necessary work. Explain what those activities allow the project to test, distinguish, or explain.

A useful drafting sequence is: current uncertainty, proposed contribution, test, and consequence for understanding. The order is advice, not a required NSF paragraph template. Some disciplines will need a different presentation, but the reader still needs the relationship between the question and the evidence.

Compare “we will build a database” with “the database will allow us to test whether the existing relationship holds across the stated conditions.” The latter is still incomplete unless the conditions, measures, and comparison are specified. Avoid promising that a resource alone will transform a field.

Do not hide a null or inconclusive result behind a success-only narrative. State what you would learn if the preferred explanation is not supported. Identify the assumption that result would challenge and how it affects the next task. This gives the assessment of success a scientific purpose beyond checking a deliverable off a schedule.

Use the NSF agency route map when the real problem is program fit rather than wording. A proposal that answers the wrong program's question can remain a bad fit after extensive editing. Selecting the correct route precedes optimizing the prose.

Team, Resources, and Data Must Match the Method

The qualifications and resources elements are not invitations to list every institutional asset. Select the capabilities the proposed work depends on. State who will perform an analysis, who will manage a technical task, and which facilities or datasets make it possible.

Distinguish availability from access. A facility on campus may require scheduling, training, fees, or permission. A public dataset may have limits that affect the proposed analysis. A collaborator's general expertise does not establish that they have agreed to deliver a specific task.

Review the schedule from the dependency backward. If access is needed before the first measurement, put its approval step before that measurement. Describe a fallback when it is realistic. Do not add a fallback that quietly changes the scientific question without acknowledging the change.

Use the policy hub's current supplements to verify data-management requirements. A data plan is not merely a storage list: connect data handling, sharing conditions, and limitations to the scientific plan. However, do not infer a universal requirement from a supplement's existence. Read its scope and effective date.

Before submission, cross-check the method against the budget and personnel documents. A task with no assigned time or resource explanation deserves attention. Conversely, a large budget item that serves no stated method can distract from the proposal's logic. How NSF review works covers the process after the file is assembled.

Read Reviews as Evidence, Not a Binding Revision Order

A panel summary and reviewer comments identify assessed strengths and weaknesses. They are not a guarantee of funding after the applicant completes a to-do list. Program staff also consider other factors in the award decision.

Separate three kinds of concern: unclear explanation, missing evidence, and program fit. A concern about confusing prose may need a rewrite. A concern about an unsupported assumption may need new data, a changed method, or a narrower claim. A concern about fit may require a different solicitation rather than another draft of the same one.

If comments appear inconsistent, map each concern to the part of the proposal it discusses. Two reviewers may be reacting to different assumptions, not giving mutually exclusive instructions. Prepare focused questions for the program officer without treating a discussion as advance approval of a future proposal.

Do not promise an acceptance probability based on an overall rating, two favorable criteria assessments, or a previous award to a similar institution. The NSF funding overview and official review guidance describe a process, not an automatic rating-to-award conversion. The Research.gov proposal information provides the official system context.

A Final Evidence Audit Before Submission

Use this checklist after the substance is stable:

• Does the first-screen claim identify a knowledge advance rather than only a field or activity?
• Does the central method answer that question?
• Is the success assessment meaningful even if the expected result does not occur?
• Are the responsible people and necessary resources identified accurately?
• Does the Broader Impacts plan stand on its own evidence?
• Are the summary and description consistent with the selected solicitation?
• Are claims of access, prior findings, and collaboration documented?
• Does every cited policy refer to the version and supplement actually in effect?

For a second hypothetical example, a team proposing a new instrument should not treat an equipment list as proof of feasibility. The draft should connect the instrument's capabilities, the planned measurements, the responsible operator, and the question those measurements address. If one link is missing, revise that link before adding another promotional sentence.

The measurable next action is a completed claim-to-evidence map and a revised paragraph whose supporting evidence can be located. This public guide does not require buying anything. The grant search and report records can support a broader comparison of programs or award context, but award history cannot verify a proposed scientific method.

Maintenance and Related Guides

This guide covers writing to the Intellectual Merit criterion, not NSF route selection or the whole review process. Use the NSF grants guide, Broader Impacts guide, CAREER guide, and NSF proposal-writing guide for their separate reader jobs.

Recheck the policy hub and solicitation before each submission. When NSF changes a version or supplement, compare the exact affected preparation or review requirement; do not rewrite unrelated criterion advice from a date change alone. Keep examples labeled hypothetical and remove unsupported claims about personal review experience or funded outcomes.

What NSF Doesn't Tell You: Trace Each Claim Through the File

Based on our analysis of the 5 shared elements, a draft is strongest when the reader can follow the same scientific claim across its summary, method, assessment, and resource explanation. Don't rely on a polished summary to carry an unsupported plan. You can't fix unavailable data with a stronger adjective, and a favorable comment isn't an award notice.

Use a hypothetical case study of one investigator at a university proposing a comparison between two datasets. The method assumes access to both, but only one is available. The next action is to establish access or redesign the comparison before rewriting the novelty sentence. No rating or funding outcome is implied by this example.

Assign a draft location to every worksheet entry. Use a section name and paragraph reference rather than only “see proposal.” Ask a colleague to follow the map without your explanation. If they cannot find the comparison, qualification, or assessment rule, determine whether the evidence is missing or merely hard to locate.

According to NSF Chapter III, 5 shared elements connect the contribution, concept, plan, qualifications, and resources. According to NSF Chapter II, the Project Summary must fit within 1 page. Those constraints serve different jobs: the former is an evaluation framework, while the latter is a preparation limit. A summary can comply with the limit and still leave an important scientific claim unsupported.

This audit avoids a false numerical score. Record resolved and unresolved dependencies instead. A critical missing resource is not equivalent to a punctuation error, even if both produce one checklist mark.

Sources

Accessed September 27, 2026:

• Current PAPPG version and supplements
• PAPPG 24-1, Chapter III: processing and review
• PAPPG 24-1, Chapter II: proposal preparation
• NSF merit-review overview
• NSF funding overview
• Research.gov proposal preparation and submission