AI and Your Form 990: Drafting Narrative Sections That Tell Your Story to Funders and Watchdogs

Which Sections Are Narrative, and Who Actually Reads Them

The Form 990 mixes hard financial data with several sections that ask you to write in plain prose. Understanding which parts are narrative helps you focus your writing effort where it matters most. Part III, Program Service Accomplishments, is the centerpiece. It asks you to state your mission and then describe your largest programs by expense, including what each program does and whom it serves. This is your primary opportunity to explain your work in your own voice, and it appears prominently on page two of the form.

Schedule O, Supplemental Information, is the second major narrative surface. The IRS requires all Form 990 filers to submit Schedule O, and it is where you provide expanded explanations for questions elsewhere on the form. Organizations use it to elaborate on program accomplishments that did not fit in Part III, to describe governance practices, and to explain the process by which the board reviewed the completed return. Schedule O is where a thoughtful organization controls its own narrative rather than leaving readers to interpret bare financial figures.

The audiences for these sections are consequential. Foundation program officers read program descriptions to assess whether your work aligns with their funding priorities. Watchdog platforms including Candid, which operates GuideStar, and Charity Navigator pull narrative data from filings into the public profiles that donors consult before giving. Investigative journalists and researchers use 990 filings as primary sources. Board candidates evaluating whether to join your organization often start with the public filing. Each of these readers looks for the same qualities: clarity, consistency, and candor.

A frequently cited observation among nonprofit finance professionals is that the single most common and costly mistake is leaving narrative sections thin or blank. An organization doing excellent work can undermine its own reputation by describing that work in a way that sounds generic or evasive. The narrative sections are free real estate for building trust, and organizations that treat them seriously tend to have stronger relationships with funders and better public reputations.

Using AI to Translate Internal Jargon Into Public-Facing Language

Every organization develops its own internal shorthand. Program names, funding-stream acronyms, methodology labels, and sector-specific terminology all make sense to staff but mean nothing to an outside reader. When this internal language leaks into the 990, it makes program descriptions harder to understand and can make an organization seem inward-looking. A foundation officer skimming dozens of filings will not pause to decode your acronyms. A journalist may misinterpret them. The goal for the 990 is language a smart reader with no background in your field can follow on a first read.

Translating jargon is one of the most reliable uses of AI for this task. Language models are good at recognizing specialized terms and rephrasing them in plain language, and they can do it at speed across long passages. A useful prompt is to paste your draft and ask the model to rewrite it for a reader who is intelligent but unfamiliar with your sector, replacing acronyms and internal terms with plain descriptions. The result is usually a clearer version that you can then refine to make sure it still captures the nuance your work deserves.

Readability tools built into or alongside AI writing assistants can also flag sentences that are too long or too dense. Because the 990 is a public document read by a general audience, aiming for accessible, straightforward prose serves both comprehension and inclusion. Shorter sentences, active voice, and defined terms make your narrative easier to read for everyone, including reviewers moving quickly and readers using assistive technology.

There is a balance to strike. Plain language does not mean dumbing down your work or stripping out the specificity that makes it credible. The skill is removing needless jargon while keeping the concrete detail. AI can propose plainer phrasing, but a staff member must confirm that the rewrite still says exactly what you mean. The same jargon-translation habit pays off across your communications, and the discipline connects naturally to broader efforts to repurpose content across channels in a consistent, accessible voice.

Accuracy and Factuality Guardrails: Staff Review Is Non-Negotiable

Everything valuable about AI in this context depends on one non-negotiable practice: a knowledgeable staff member must review every word before it is filed. The Form 990 is signed under penalty of perjury, and it is a public document that funders and journalists scrutinize. AI language models can produce fluent text that sounds authoritative while containing invented figures, overstated outcomes, or subtle mischaracterizations of what your programs do. This tendency, often called hallucination, is not a reason to avoid AI. It is the reason human review must be built into your process.

The safest workflow treats AI as a drafting and editing assistant that works only from facts you supply. Rather than asking a model to describe your programs from general knowledge, provide it with your actual program data and instruct it to use nothing beyond what you give it. When you ask for a rewrite or a consistency check, verify the output against your source records. Any number the model produces should trace back to a figure you can document. Any outcome it describes should match something you measured. If a claim cannot be sourced, it does not belong in the filing.

Data privacy deserves attention as well. Schedule O is public, so it must never contain sensitive information such as individual names, Social Security numbers, or confidential beneficiary details, and you should be careful about what you paste into AI tools during drafting. Avoid entering donor records, client identities, or other confidential data into general-purpose AI systems, and prefer tools with clear data handling terms. Treating the drafting process with the same care you apply to the filing itself protects both the people you serve and your organization.

A sound review process assigns clear ownership. Someone from the program side confirms that descriptions are accurate, someone from finance confirms that numbers reconcile with the financial schedules, and, as the IRS process contemplates, the board reviews the completed return before filing. AI accelerates the drafting so this review time is spent on judgment rather than on wordsmithing from scratch. The distinction between AI-assisted work, which is entirely appropriate, and unreviewed AI output, which is a serious risk, is the presence of genuine human editorial judgment at every stage.

Common Mistakes to Avoid in Your 990 Narrative

The most common mistake is neglect. Organizations file thin, generic program descriptions or leave Schedule O nearly empty, treating the narrative as an afterthought. This wastes a valuable opportunity to build trust and can make even excellent organizations look evasive to the funders and watchdogs who read the filing. The second most common mistake is the opposite: overreaching. Some organizations, sometimes with AI assistance, produce inflated descriptions full of outcomes they never measured and superlatives that read as promotional rather than factual. Sophisticated readers discount this kind of language, and it can undermine credibility.

Inconsistency is a third recurring problem. When the 990 narrative disagrees with the annual report or the grant application, careful readers notice, and the discrepancy raises more doubt than any single document would on its own. A fourth mistake is jargon: filings so dense with acronyms and internal terminology that an outside reader cannot follow what the organization actually does. Each of these mistakes is avoidable, and each is exactly the kind of issue that a disciplined, AI-assisted drafting and review process is designed to catch before filing.

A final caution concerns over-reliance on AI without human ownership. AI that drafts unreviewed narrative can introduce invented figures, subtle inaccuracies, and a generic voice that does not sound like your organization. The antidote is the same throughout this guide: use AI to accelerate the work, but keep knowledgeable staff firmly in control of what gets said and filed. The organizations that get the most from these tools are the ones that treat AI as a capable assistant rather than an autonomous author.