Dear Suzannah: Can AI pick the grants we should pursue?

Here’s the deal

Microsoft’s October 28, 2026 Copilot harness announcement describes systems that can use tools and carry out complex tasks. Useful technology news, certainly. But a more capable agent is not automatically a better grants officer. Software still needs rules about what evidence counts, what it may recommend, and what a person must approve.

Grants.gov’s eligibility guidance says each federal notice of funding opportunity defines who may apply. A broad search category such as “nonprofit” is not a promise that your nonprofit meets that specific notice’s requirements.

And Grants.gov’s applicant registration guidance explains that an organization applying for federal grants must register through SAM.gov first, obtain a Unique Entity ID, and maintain active registration. Those are operational requirements an attractive AI summary can overlook. A grant listing is a lead to investigate, not approval to submit.

What this actually helps you do

• 1. Protect staff timeReject clearly ineligible opportunities before anyone drafts a narrative or prepares attachments.
• 2. Produce a useful shortlistKeep requirements, source links, open questions, and an accountable reviewer in the same record.
• 3. Make funding decisions clearerSeparate promising opportunities from verified fits and see which applications justify your team’s limited time.

The small nonprofit use case

Picture an after-school arts nonprofit with two employees, several volunteers, a program calendar, and a spreadsheet of grants. An AI assistant spots a funding announcement for youth education. The title fits, but its eligibility section requires a state agency or eligible local government. Your organization is a nonprofit. The topic is relevant; the applicant type is not.

The useful output is not “Apply immediately.” It is: “Potential program match, applicant type does not match. Put on the no-go list, cite the eligibility section, and consider whether a partner application is even permitted.” That last part still needs human review of the notice and partner rules. Never assume teaming up repairs ineligibility.

Evidence rule: Each answer needs a source location in the current notice or official application instructions. If you cannot find it, mark the answer “unverified,” not “yes.”

How to build the screening process

1. Write a one-page organization profile. Include legal type, mission, service area, programs, budget range, current registrations, and staff capacity. Date it so stale details do not linger.
2. Collect official opportunities. Start with the funder’s current announcement and complete package. Store the direct URL, opportunity number if available, source date, and deadline. Skip unreliable third-party summaries.
3. Ask AI to extract requirements, not invent answers. Request eligibility, required attachments, allowable costs, match, geography, deadlines, and the section or page supporting each item.
4. Compare requirements to the nonprofit profile. Label each verified yes, verified no, or needs review. Missing or conflicting information never means automatic approval.
5. Route exceptions to a person. Assign ambiguous requirements to someone who can read the original wording and use the funder’s official clarification process when appropriate.
6. Check capacity before writing. Even an eligible grant may be a poor fit if delivery, reporting, matching funds, or the deadline exceeds what the team can support.
7. Record the decision. A named reviewer marks pursue, investigate, or decline, with source and reason. Only approved opportunities move to the application calendar.

Tips and tricks

Require a source section beside every eligibility verdict. Keep the notice’s version date with the screening record. Sort by deadline only after passing eligibility, because a calendar full of impossible applications is an impressively organized way to waste a week.

Common mistakes

Matching mission words instead of legal criteria. A youth program is not eligible for every youth grant. Ignoring registration lead time. Grants.gov says SAM.gov registration can take time, so do not begin on the deadline. Confusing award size with available revenue. An opportunity is neither a grant award nor cash. Copying last year’s instructions. Confirm the current notice and amendments.

The human review checklist

• Did we read the actual notice and amendments rather than only an AI summary?
• Is our applicant type expressly permitted?
• Are geography, beneficiaries, activities, and expenses eligible?
• Do we understand matching, partnership, registration, and reporting requirements?
• Are deadlines and submission steps taken from the current source?
• Do we have time and capacity to apply and deliver the work?
• Has a named person approved the pursue or decline decision?

NIST’s Generative AI Profile provides voluntary guidance on managing generative AI risks. Here, the practical application is to make important AI-produced conclusions traceable and reviewable before funding decisions.

How to measure success

Track opportunities discovered, notices reviewed, verified eligible matches, declined opportunities with reasons, applications submitted, and awards confirmed. Record staff hours spent screening and writing. Keep possible award amounts separate from funding formally awarded and money actually received.

After a month, ask whether the team spent less time evaluating opportunities that never qualified and whether the shortlist produced better application choices. If not, revise the screening questions before adding new software.

For another example of checking AI-generated evidence, see AI Can Research the Market in Minutes. Make It Show Its Work. For permissions, read Your Maintenance AI Needs Its Own Access Log.

FAQ

Glossary

NOFO

Notice of Funding Opportunity, the official document describing a federal grant and its application requirements.

UEI

Unique Entity ID, the identifier issued through SAM.gov for federal registrations.

Eligibility gate

A requirement to satisfy before considering an opportunity an application candidate.

Human review

A named person’s examination and approval of a decision using underlying records.

Sources and further reading

1. Microsoft, Introducing the Microsoft Copilot harness (October 28, 2026). Official announcement about tool-using agents and managed processes.
2. Grants.gov, Grant Eligibility. Official explanation that each notice defines eligibility.
3. Grants.gov, Applicant Registration. Official SAM.gov and organizational registration guidance.
4. NIST, Generative AI Profile (July 2024). Voluntary risk-management guidance.