86% Success Rate
$1.6B Secured
NSF, DoD & NIH Expertise
SBIR/STTR, Grant Strategy

The Language That Gets Your Grant Declined: Agency-Specific Writing Rules for NIH, NSF, and ARPA-H

SBIR grant writing by agency differs in vocabulary, scoring direction, and framing conventions -- and getting them wrong is one of the fastest ways to get declined. NIH scores 1 = best while NSF and ARPA-H score 9 = best. NIH uses "investigator" where ARPA-H requires "performer." Using hypothesis-driven framing (standard at NIH) at ARPA-H signals you don't understand the agency's culture. These differences are systematic, not cosmetic.

This guide is a structured reference for agency-specific grant writing rules -- covering banned terms, required vocabulary, scoring conventions, and a translation table for reframing the same technology across NIH, NSF, and ARPA-H. Every rule below comes from Cada's production grant writing playbooks -- the same playbooks used to write 50+ SBIR applications with an 86% success rate across NIH, NSF, and ARPA-H (source: Cada internal data, 2023-2026).


Do NIH and NSF Use the Same SBIR Scoring Scale?

NIH and NSF both score on a 1-9 scale, but the direction is reversed. At NIH, a score of 1 means Exceptional (best possible). At NSF and ARPA-H, a score of 9 means Exceptional. A "3" at NIH is excellent. The same "3" at NSF means weaknesses outweigh strengths.

NIH NSF ARPA-H
Score of 1 Exceptional (best) Poor (worst) Poor (worst)
Score of 9 Poor (worst) Exceptional (best) Exceptional (best)
"Fundable" range 1-3 7-9 7-9

This isn't trivia -- it shapes how reviewers think. An NIH reviewer giving you a "3" is saying "Excellent -- strong, only minor weaknesses." An NSF reviewer giving you a "3" is saying "Fair -- weaknesses outweigh strengths." Same number, opposite meaning.

How Do the Scoring Criteria Compare?

Each agency scores on 5 criteria, but they weight different things:

NIH (1=best) NSF (9=best) ARPA-H (9=best)
Significance (health burden, scientific premise) Intellectual Merit (scientific/engineering knowledge) Non-Incremental Innovation (25%)
Investigator(s) (PI qualifications, preliminary data) Broader Impacts (societal/national benefit) Health Impact and Scale (25%)
Innovation (challenges existing paradigms) Commercial Impact (market need, adoption path) Technical Feasibility and Milestones (20%)
Approach (methodology, potential problems) Technical Risk & Innovation (novel, unproven) Team and Execution Capability (15%)
Environment (resources, institutional support) Writing Quality (clear, no AI patterns) Writing Quality and PM Communication (15%)

NIH reviewers are scientists evaluating your hypothesis. NSF reviewers are evaluating whether your R&D is genuinely novel and nationally significant. ARPA-H reviewers want to know if your solution will transform health outcomes at scale -- and they'll decide in 60 seconds.


What Words Should You Avoid in SBIR Grant Writing?

Certain terms will hurt your SBIR application regardless of agency. Words like "game-changing," "revolutionary," and "leverage" trigger reviewer skepticism at NIH, NSF, and ARPA-H alike. Other terms are agency-specific: "hypothesis-driven" works at NIH but signals the wrong culture at ARPA-H, while "at scale" without specifics will cost you points at NSF.

Category 1: Hype and Buzzwords (Banned Everywhere)

These terms are banned at NIH, NSF, and ARPA-H. No exceptions.

Term Use Instead Why It's Banned
game-changing Describe the specific change with metrics Buzzword with no specificity
revolutionary State what the technology does differently Unsubstantiated hype
breakthrough Name the specific advance and evidence Overused, triggers reviewer skepticism
best-in-class Compare to specific benchmarks Empty superlative
next-gen Describe the generation gap specifically Vague marketing language
moonshot Describe the specific ambitious target Oversells without evidence
cutting-edge (for own technology) Describe what makes it current OK for baseline comparisons, never for own work
paradigm shift Describe the specific paradigm change Overused in academic and grant writing
state-of-the-art (for own technology) Describe specific capabilities OK when describing baseline to surpass
world-class Name specific qualifications or rankings Empty credentialing
unprecedented (without proof) Cite evidence of novelty Must be substantiated or removed
disruptive (without evidence) Show specific market/clinical disruption Requires market evidence at NSF, clinical evidence at ARPA-H

Category 2: AI-Telltale Words (Restricted Everywhere)

Reviewers at all three agencies are trained to spot AI-generated prose. These terms signal machine-written text and will raise flags in 2026.

Term Max Uses Use Instead
leverage 0 use, apply, employ
furthermore / moreover / additionally 1 total Vary transitions or eliminate
"it is worth noting that" 0 Delete -- state the point directly
holistic 0 Name the specific scope
seamless / seamlessly 0 Delete or be specific about integration
synergy / synergistic 0 Describe the specific interaction
robust 1 max Use a specific descriptor
utilize 0 use
facilitate 0 enable, allow
innovative / novel 2 total Describe what's specifically new
comprehensive 0 Be specific about coverage

Category 3: NIH-Specific Bans

These terms are specifically problematic at NIH. Some are acceptable at other agencies.

Term Context Use Instead Why
proprietary technology Research Strategy Describe the technical approach Commercial language banned from Research Strategy
competitive advantage Research Strategy Frame as scientific contribution Save for Commercialization Plan
market leader / first to market Research Strategy Frame as unmet clinical need NIH reviewers are scientists, not investors
ROI / revenue projections Research Strategy Frame as health impact Belongs exclusively in Commercialization Plan
TAM / SAM / SOM Research Strategy Frame as patient population size Pitch-deck language banned from scientific narrative
explore / investigate (alone) Aims determine, test, quantify, validate Too passive -- NIH wants hypothesis-driven verbs
"more research is needed" Gap statement Name the specific missing knowledge Not a gap statement -- must be specific

Category 4: ARPA-H-Specific Bans

These terms signal the wrong agency culture at ARPA-H. Many are standard NIH language that will hurt you here.

Term Use Instead Why It's Banned at ARPA-H
hypothesis-driven solution-driven ARPA-H funds solutions, not hypotheses
exploratory study proof-of-concept demonstration ARPA-H funds demonstration, not exploration
specific aims solution summary sections Not the ARPA-H format
investigator performer ARPA-H's required vocabulary
program officer program manager (PM) Different role title
R01-style (don't reference NIH formats) Signals NIH culture
pilot study proof-of-concept demonstration ARPA-H language convention
"preliminary data suggests" "preliminary data demonstrates" ARPA-H wants conviction, not hedging
basic research applied/translational research ARPA-H doesn't fund basic research
optimization / enhancement Describe the 10x improvement ARPA-H requires non-incremental: "10x not 10%"
incremental / iterative / evolutionary Describe the step-change Incremental language is disqualifying
grant / contract Other Transaction (OT) ARPA-H uses OTs, not traditional grants

Category 5: NSF-Specific Concerns

These terms require special handling at NSF.

Term Issue Guidance
"will benefit society" (alone) Fails Broader Impacts specificity test Name the population, mechanism, and plausible scale
Revenue projections as societal benefit "$2B market" is not a Broader Impact Describe technology's national/societal benefit
"explore" / "investigate" Too passive for R&D framing Use: develop, validate, test, quantify, optimize, demonstrate
woke / equity (alone) / climate justice Ideologically-coded language Use neutral framing: "health equity," "environmental sustainability"
"at scale" Vague without specificity State specific scale: N users, M records, X throughput
"real-time" Vague without specificity State latency spec: "within Xms"

NIH vs NSF Grant Writing Differences: The Translation Table

This is the core reference for multi-agency applicants. When you're reframing the same technology for a different agency, use this table to translate your language.

Vocabulary Translation

Concept NIH Term NSF Term ARPA-H Term
Lead researcher Investigator, PI PI Performer
Agency contact Program Officer (PO) Program Director (PD) Program Manager (PM)
Initial phase Phase I (R43) Phase I Base period
Follow-on phase Phase II (R44) Phase II Option period
Core document Specific Aims (1 page) + Research Strategy (6 pages) Pitch (4 sections, character-limited) Solution Summary (6 pages)
Success criteria Milestones and expected outcomes Measurable R&D objectives Go/No-Go milestones
Funding mechanism Grant (cost-reimbursement) Grant (cost-reimbursement) Other Transaction (milestone-based payments)
Award amount (Phase I / base period) Up to $314K total costs ($250K direct costs cap for modular budgets) Up to $305K $1M-$5M (base period)

Framing Translation

Dimension NIH NSF ARPA-H
Innovation framing Hypothesis-driven: "We will test the hypothesis that X enables Y" R&D novelty: "The innovation introduces a novel [method] enabled by [scientific principle]" Solution-driven: "[Technology] achieves [metric] by [mechanism], a [Xx] improvement over [standard]"
Impact framing Scientific contribution: "These studies will advance understanding of [mechanism]" Societal/national benefit: "If successful, results could enable [societal benefit] for [population]" Patient outcomes: "Patients with [condition] will [specific benefit]"
Risk framing Potential problems + alternative strategies: "If [problem], we will [alternative]" High-risk/high-reward R&D: unproven method, not market uncertainty Honest technical risk: "List 3-5 technical risks" -- pretending there's none signals naivety
Commercial framing Separated: 2-3 sentences in Research Strategy, full detail in Commercialization Plan In-pitch: market need quantified, customer pain clear, no pitch-deck language Patient impact: health burden in lives/QALYs, not TAM/revenue
Preliminary data Heavily weighted: even "not required" = expected by reviewers. Present per-aim. Feasibility evidence: enough to show R&D plan is credible Proof-of-concept: "demonstrates" not "suggests"
Review audience Panel of 15-20 scientists (study section) PD screening + technical expert panel Single Program Manager

How Do NIH, NSF, and ARPA-H Define "Impact" Differently?

This is the #1 framing difference that trips up multi-agency applicants. Get this wrong and your application reads like it was written for a different agency -- because it was.

Agency What They Call It What It Actually Means Common Mistake
NIH Significance Health burden quantified + scientific premise + how this changes the field Framing as market opportunity
NSF Broader Impacts How the technology itself benefits national interests (not education, not revenue) Using academic NSF Broader Impacts framing (student training, outreach)
ARPA-H Health Impact and Scale Patient outcomes: lives saved, QALYs, cost to health system Using market language (TAM, revenue) as primary framing

How Do NIH, NSF, and ARPA-H Define Innovation Differently?

Each agency has a distinct innovation bar. NIH evaluates whether your R&D challenges existing paradigms through hypothesis-driven science. NSF classifies innovations into three tiers, and engineering optimizations (Tier C) get declined regardless of quality. ARPA-H requires non-incremental solutions -- 10x improvement over current standards, not 10% optimization.

NIH: Innovation as One of Five Criteria

At NIH, Innovation is a scored criterion alongside Significance, Investigator, Approach, and Environment. The bar: does this challenge existing paradigms or use novel approaches?

NIH wants genuine R&D, not product development dressed as research. Frame aims as hypothesis-driven research ("determine whether X," "test the hypothesis that Y"), not product tasks ("build X," "test Y").

NIH also requires a "scientific premise" -- the body of prior research supporting your hypothesis, with a rigor assessment. This is NIH-specific and not required at NSF or ARPA-H.

NSF: Innovation Classification Is the "Single Most Important Determinant"

NSF classifies every submission into one of three innovation tiers:

Tier Description Score Implication
A New scientific principle or method Score floor: 7 (out of 9)
B Novel application of known science to a new domain Score floor: 5
C Engineering optimization of existing approaches Score ceiling: 4 -- these get declined

If your technology is a "better, faster, cheaper" version of something that exists, NSF will classify it as Tier C regardless of how well you write the pitch. The leap must be scientific or methodological, not engineering.

ARPA-H: Non-Incremental Is the Mandatory Bar

ARPA-H's standard: 10x improvement, not 10% optimization. If your innovation cannot be clearly distinguished from engineering optimization, it won't pass the PM's 60-second evaluation.

ARPA-H uses the Heilmeier Catechism (10 questions) as its evaluation framework. The first question -- "What are you trying to do? Articulate your objectives using absolutely no jargon" -- sets the tone. Plain language, ambitious scope, patient-centered outcomes.


What Are the Key Structural Differences Between Agencies?

PI Requirements

Requirement NIH NSF ARPA-H
Citizenship Not required Required (US citizen or permanent resident) Not required (SAM.gov registration needed)
Employment >= 51% by the small business during award >= 51% by the small business during award Varies by OT agreement
STTR partner University, national lab, or nonprofit research institution Same N/A (ARPA-H doesn't have STTR)

Budget Presentation

Dimension NIH NSF ARPA-H
Format Modular budget (up to $250K direct/year) Character-limited section within pitch Basis of Estimate within 6-page Solution Summary
Award range (Phase I / base period) Up to $314K total costs Up to $305K $1M-$5M (base period)
Duration 6-12 months 6-18 months 12-24 months (base period)
Subcontracting cap 33% max (SBIR) Similar No fixed cap (but >50% = red flag)

Review Process

Dimension NIH NSF ARPA-H
Who reviews Study section (15-20 scientists) PD screening, then technical expert panel Single Program Manager
Writing for A committee -- balance multiple perspectives Two gates -- PD first, then technical experts One person -- convince a domain expert
Deadlines Fixed: April 5, August 5, December 5 Varies by topic Rolling -- no fixed deadlines
Resubmission One A1 allowed (must address prior critique) Resubmit as new No formal resubmission process

Agency Quick Reference Cards

NIH SBIR Phase I -- Quick Reference

  • Tone: Academic/scientific. First-person plural ("we"). Not commercial.
  • Core document: Specific Aims (1 page) + Research Strategy (6 pages: Significance + Innovation + Approach)
  • Scoring: 1-9, where 1 = Exceptional. Fundable range: 1-3.
  • Key vocabulary: Investigator, Program Officer, specific aims, hypothesis-driven, scientific premise, study section
  • Innovation bar: Must challenge existing paradigms with hypothesis-driven R&D
  • Commercial language: Strictly separated. 2-3 sentences in Research Strategy; full detail in Commercialization Plan only.
  • Preliminary data: "Not required" but heavily expected. Present per-aim.
  • Review process: Panel of 15-20 scientists (study section). Three assigned reviewers per application.

NSF SBIR -- Quick Reference

  • Tone: Calibrated confidence. "Will test" not "will prove." "Targets X%" not "achieves X%."
  • Core document: 4-section pitch (character-limited: 3,500 / 3,500 / 1,750 / 1,750 characters)
  • Scoring: 1-9, where 9 = Exceptional. Competitive range: 7-9.
  • Key vocabulary: PI, Program Director, R&D objectives, Broader Impacts, intellectual merit
  • Innovation bar: Must be Tier A or B (new principle or novel application). Tier C (optimization) gets declined.
  • Broader Impacts: Societal/national benefit of the technology itself -- not revenue, not education.
  • Commercial language: Market need quantified in-pitch, but no pitch-deck language (no TAM/SAM/SOM).
  • Review process: PD screening (5 questions), then technical expert panel. Decision: Invite / Decline.

ARPA-H -- Quick Reference

  • Tone: Conviction-based. Direct assertions. "Will demonstrate" not "aims to explore."
  • Core document: 6-page Solution Summary (11pt Calibri, 1" margins, single-spaced)
  • Scoring: 1-9, where 9 = Exceptional. Competitive range: 7-9.
  • Key vocabulary: Performer, Program Manager, base period, option period, milestone, Go/No-Go, Other Transaction
  • Innovation bar: Non-incremental mandatory. "10x not 10%." Evaluated via Heilmeier Catechism.
  • Impact framing: Patient outcomes (lives, QALYs), not market size or revenue.
  • Risk framing: Acknowledge 3-5 technical risks. Pretending there's none signals naivety.
  • Review process: Single PM decides. No panels. Writing must convince one domain expert in 60 seconds.

Frequently Asked Questions

Can I reuse my NIH narrative for an NSF pitch?

No. Beyond the format difference (NIH's 6-page Research Strategy vs. NSF's character-limited 4-section pitch), the framing is fundamentally different. NIH wants hypothesis-driven science with a scientific premise. NSF wants R&D with a national significance framing. The same technology needs to be re-argued from scratch for each agency. Reusing NIH language at NSF is one of the most common reasons multi-agency applicants get declined.

What's the biggest language mistake multi-agency applicants make?

Carrying hypothesis-driven framing from NIH to ARPA-H. "Hypothesis-driven" is NIH's native language -- it signals scientific rigor. At ARPA-H, it signals you're proposing basic research, not a solution. ARPA-H funds solutions that achieve 10x improvements, not hypotheses that need testing.

Do scoring directions actually affect how reviewers write their critiques?

Yes. NIH reviewers trained on "1 = best" anchor their language differently than NSF reviewers trained on "9 = best." When you read summary statements from NIH, a score of "3" comes with language like "strong, only minor weaknesses." The same "3" at NSF means "weaknesses outweigh strengths." If you've seen feedback from one agency, don't assume the scale when interpreting feedback from another.

How do I handle preliminary data differently across agencies?

NIH: Present preliminary data for every aim, even though it's technically "not required." Study sections almost always penalize proposals without it. NSF: Show enough feasibility evidence to prove the R&D plan is credible -- the bar is lower than NIH. ARPA-H: Frame as "proof-of-concept demonstration" using conviction language ("demonstrates" not "suggests").

Which agency is most forgiving of commercial language?

None are forgiving, but they each want commercial context in different places. NIH strictly separates it into the Commercialization Plan (a separate document). NSF allows market context within the pitch but penalizes pitch-deck language (no TAM/SAM/SOM). ARPA-H wants impact framed as patient outcomes and health system cost, not revenue or market size.

How long does it take to adapt a proposal from one agency to another?

For a thorough rewrite -- not a find-and-replace -- expect 20-30 hours to adapt a complete NIH application for NSF or ARPA-H. The vocabulary changes are the easy part (2-3 hours). The framing changes (impact, innovation, commercial context) require rethinking the narrative structure. Most founders underestimate this, treat it as a quick edit, and get declined.


Next Steps

If you're applying to multiple agencies, understanding SBIR grant writing by agency is the difference between a competitive application and a declined one. The rules above come from Cada's production playbooks for NIH, NSF, and ARPA-H -- the same system used to write 50+ SBIR applications with an 86% success rate.

Not sure which agency fits your technology? We do a free 15-minute agency-fit assessment -- we'll tell you which programs match, which framing to use, and where the language traps are. No pitch, no obligation -- just a straight answer on agency fit.