How to write an NSF SBIR Phase I proposal with AI, section by section

How the NSF SBIR process works

NSF's SBIR and STTR programs, branded America's Seed Fund, fund early-stage research and development at US small businesses. NSF does not take equity. The process has three stages, and the first is not a proposal at all.

1. Submit a Project Pitch: a short description of the technical innovation, the market opportunity and the company. NSF says to expect a response in about one to two months.
2. If NSF invites you, register your company and prepare the full Phase I proposal for one of the published deadlines.
3. NSF says to expect a funding decision about six months after you submit.

The four headers of the Phase I Project Description

The Project Description is the heart of the proposal: at least 10 and at most 15 pages, under four required headers. NSF's own guidance recommends how to divide the pages:

Lead with the technical risk

The most common weakness in a first SBIR proposal is that it reads like a business plan. Phase I funds research that might not work. Reviewers want to see the specific technical question you do not yet know the answer to, why it is hard, and how Phase I will answer it.

Write the risk in one sentence before anything else, such as: "We do not know whether a capacitive sensor can measure soil organic carbon to within 0.5 percentage points across clay and sandy soils at a unit cost under $40." Everything in Intellectual Merit then serves that sentence. Draimo is useful here as a sparring partner: give it your technical notes and ask it to state the riskiest assumption in your approach, then argue with it.

Milestones with numbers, not intentions

NSF asks for quantitative success criteria, and this is where many drafts fall short. "Develop a prototype" is an intention. "A prototype that measures organic carbon within 0.5 percentage points of lab results on 40 field samples from three soil types by month 6" is a milestone a reviewer can believe.

Ask Draimo to turn your plan into a table of milestones with a measurable criterion, a month and a fallback for each, and to flag any milestone whose success cannot be measured. It will also check that the milestones add up to the risk you said Phase I retires.

Commercialization Potential that a reviewer believes

Reviewers do not expect a Phase I company to have revenue. They do expect you to know who pays, why they would switch, and how you reach them. Name the first customer segment precisely, give the price you expect and the evidence for it, and describe the path to Phase II and beyond.

Market facts need sources. Run the market research in Agent Mode, which cites every source, then press Make a document under the answer to carry it into the proposal draft. Keep the sources in your notes, because the Project Description itself must not contain URLs.

Rules that get proposals returned without review

Some failures are procedural, and they are the cheapest to avoid. Check each one against the active solicitation before you submit.

• A Project Description shorter than 10 pages or longer than 15.
• A missing required header, or headers renamed.
• URLs or links inside the Project Description, which NSF's guidance says are not acceptable.
• Submitting a full proposal without an invitation from a Project Pitch.
• Formatting that does not follow the solicitation and NSF's proposal guide.

Using AI and staying within NSF's rules

NSF's notice on generative AI encourages proposers to indicate in the project description the extent to which generative AI was used and how. It also prohibits reviewers from uploading proposal content to non-approved AI tools, because confidentiality matters in merit review. The responsibility for accuracy and originality stays with the people who submit.

In practice: use AI to structure, tighten and check your writing, and to test your argument. Do not let it invent results, citations or letters of support. Keep a short note of how you used it, so you can describe it honestly in the proposal.

How to draft it with Draimo

See a complete example first: this seven-page Phase I proposal for soil-carbon sensors was made on Draimo, and the gallery page shows the prompt that produced it.

1. Put your technical notes, preliminary data and team bios in one project.
2. Ask Draimo to state your riskiest technical assumption, and refine it until it is one precise sentence.
3. Research the market and competing approaches in Agent Mode, with sources.
4. Ask for the Project Description under NSF's four headers, with a milestone table, and set the citation style you use.
5. Edit in plain words: "cut Company/Team to one page", "make every milestone measurable".
6. Read every technical claim against your data, then download the Word file to finish in your own formatting.

Sources we checked

• NSF America's Seed Fund: how to apply
• NSF Seed Fund: Phase I Project Description guidance
• NSF: Use of generative AI in the merit review process

Checked October 21, 2026. Dates and amounts change every year. Ask Draimo for this year's figures, grounded in official sources.