Proposals are not read, they are scored — against published criteria with published weights, by reviewers working through a stack. Writing to the scoresheet is the whole craft, and almost nobody does it.
The central fact about grant writing is structural rather than literary. Your proposal will be assessed by reviewers working through a stack of applications, assigning points against criteria that were published in advance. They are not looking for the best piece of writing. They are filling in a scoresheet.
Almost everything that distinguishes a competitive proposal follows from taking that literally. The review criteria are published. In federal NOFOs the weights are published too — the Uniform Guidance requires every NOFO to state the relative percentages or weights wherever criteria differ in importance. Nobody is hiding the rubric — and yet the most common structural error in grantseeking is writing a compelling narrative that does not map to it.
Before writing anything, extract the criteria and their weights into a table. That table is your outline.
The scoresheet decides the ranking, not the award. Program staff select from the ranked list, and since Executive Order 14332 in August 2025 a senior appointee at each agency reviews discretionary awards before they issue; the order states that peer review recommendations are advisory. A high score gets you to the decision. It does not make it.
If Approach is worth 40 points and Organizational Capacity 15, then Approach deserves roughly two and a half times the space and attention. This sounds obvious and is routinely violated: applicants write at length about their organization’s history and founding story — genuinely interesting to them, worth a seventh of the total — and then compress the methodology into a page.
The same logic applies to what you cut. When you are over the page limit, cut from the lowest-weighted sections. Page limits are enforced, and proposals that do not meet the agency’s requirements can be returned without review at all.
NSF evaluates every proposal against two criteria approved by the National Science Board: Intellectual Merit — the potential to advance knowledge — and Broader Impacts — the potential to benefit society and contribute to specific, desired societal outcomes. A program solicitation can add criteria of its own. Broader Impacts is the one applicants treat as a formality, and it is not; it is half the published framework.
NIH changed its framework for application due dates from January 25, 2025. Most research project grants — R01, R03, R15, R21, U01 and a list of related activity codes — are now reviewed on three factors:
Reviewers still assign a single overall impact score from 1 to 9, where 1 is exceptional.
The third factor is the interesting change. Investigator and Environment used to carry numeric scores of their own. NIH made them a sufficiency check to reduce the potential for general scientific reputation to have an undue influence on review. For an early-career applicant or a smaller institution, that is a material shift: you now need to demonstrate you are sufficient to do the work, not that you out-credential the competition.
Two newer NIH rules change how you plan a submission cycle. Since the September 25, 2025 receipt date, NIH accepts at most six new, renewal, resubmission or revision applications per principal investigator per calendar year, and it will not treat applications substantially developed by AI as the applicant’s original ideas.
If you are a company applying to these agencies through SBIR or STTR, the deep tech funding guide tracks the current NSF, NIH and Defense solicitations and their next windows.
Anyone who has not yet checked eligibility. The most common reason proposals fail is that the applicant was never eligible for the program. No writing technique addresses that, and it is diagnosable in ten minutes.
Anyone hoping for a template. Templates produce proposals that read like templates. The funder’s own criteria, in the funder’s own order and vocabulary, is the only structure worth using.
Anyone whose evidence base does not exist yet. Reviewers are subject-matter experts. Confident prose about outcomes you have not measured reads as exactly what it is.
Use the funder’s headings verbatim. A reviewer scoring their eighth proposal should never have to hunt for where you addressed a criterion. Matching headings is not lack of imagination; it is consideration for the person holding the pen.
Answer the question as asked. If the criterion asks how outcomes will be evaluated, describe the evaluation. A paragraph about why the work matters scores nothing there, however true.
Get a cold score. Hand a colleague the criteria, the weights, and your draft, and ask them to score it without your explanation. Everything they cannot find, a reviewer will not find either. This is the highest-value hour in the whole process and it costs nothing.
Read a funded example. Many agencies publish abstracts of funded work; some publish full proposals. Reading two funded applications in your field teaches more about register and depth than any amount of general advice.
Find the review criteria and their point weights in the NOFO. Write them into a table. That table, not a template, is the outline of your proposal.
If Approach is worth 40 points and Organizational Capacity 15, your page allocation should look roughly like that. Applicants routinely write their best prose in the sections worth least.
Reviewers work from the criteria text. Where a criterion asks how you will evaluate outcomes, a paragraph about why the work matters scores zero against it, however well written.
Use the funder's own headings and vocabulary in the same order. A reviewer scoring their eighth proposal at the end of a long week should be able to find each criterion without hunting.
NIH's framework separates whether the work matters from whether you can do it, and NSF's separates knowledge from benefit. Most weak proposals argue one and assume the other.
Give a colleague the criteria and the point weights and ask them to score it cold. What they cannot find, reviewers will not find either.
