Filing Form 1023 With AI Help: Drafting a Defensible Exemption Application
Almost everything written about AI and nonprofits assumes the nonprofit already exists. Somebody has an EIN, a board, a determination letter, and a Form 990 to file. But every organization starts before that, with a group of people who have decided to do something and now face a federal application that asks them to describe, in detail and under penalty of perjury, activities they have not yet carried out. Form 1023 is where a good idea becomes a legal entity with tax consequences, and it is a document AI can help you write far better than it can help you decide.
The application for recognition of exemption under section 501(c)(3) is not a registration. It is a request for a determination, and the IRS is under no obligation to grant it. What you submit becomes the factual basis on which exemption is recognized, which means the application is simultaneously a writing exercise, a financial projection, a governance document, and a sworn statement. Founders tend to experience it as the first of those and are surprised by the other three.
This is exactly the kind of task where large language models look miraculous and are quietly dangerous. A model will produce a fluent narrative description of your activities in ninety seconds. It will also, if you let it, invent a program you do not run, assert an eligibility criterion you never set, describe a partnership that does not exist, and produce a three-year budget whose numbers do not reconcile with any of it. You will then sign that document attesting that it is true, correct, and complete. The fluency is real. The attestation risk is also real, and it does not care how good the prose was.
The useful framing is narrow and holds throughout this article. AI is very good at organizing what you already know into the structure the IRS expects, at finding the gaps and contradictions in what you have written, and at turning a founder's rambling explanation of a program into four paragraphs a reviewer can follow. It is not competent to decide whether you should be a public charity or a private foundation, whether your purpose clause satisfies the organizational test in your state, or whether your compensation arrangements create a private benefit problem. Those are legal and tax judgments, and you want a lawyer or a CPA who does exempt organizations work, not a chatbot.
What follows walks the process end to end: choosing between the long form and the streamlined version, the state and EIN prerequisites that come first, the organizing documents, the narrative that decides most applications, the projected budget, public charity classification and the public support test, the compensation and conflict of interest questions, submission through Pay.gov, and what to do when a letter arrives asking for more information. At each stage the same question gets answered. What should AI touch here, and what must stay with a person who can be held responsible for it.
Form 1023 or Form 1023-EZ, and Why the Shortcut Often Costs More
There are two paths to recognition. Form 1023 is the full application, a long document with schedules, a narrative, multi-year financial data, and attached organizing documents, carrying a $600 user fee. Form 1023-EZ is a short online form with a $275 user fee that asks almost nothing and takes most people under an hour. The eligibility gate for the EZ is the worksheet in the Form 1023-EZ instructions, which you must complete and keep in your records before filing. In broad terms it requires projected annual gross receipts of $50,000 or less for the current year and the next two, gross receipts of $50,000 or less in each of the past three years, and total assets with a fair market value of $250,000 or less. Several categories of organization are excluded outright regardless of size, including churches, schools, colleges and universities, hospitals and medical research organizations, and organizations seeking private foundation status.
The worksheet is where the first bad decision usually happens. Founders answer the projected receipts question optimistically in conversation and conservatively on the form, because the conservative answer unlocks the cheaper, faster path. If you genuinely expect to raise $40,000 next year and $45,000 the year after, the EZ is available to you. If your business plan involves a $200,000 government contract in year two and you check the box anyway, you have not saved time. You have made a false attestation on a federal form to obtain a filing status you were not entitled to.
Even when you are eligible, the EZ deserves more suspicion than it gets. Because it collects no narrative of substance, no financial data, and no organizing documents, the IRS is approving based almost entirely on what you attest. The National Taxpayer Advocate has studied this repeatedly and found the results poor. In one representative sample of approved Form 1023-EZ applicants, the Taxpayer Advocate Service reported an erroneous approval rate of 42 percent, with organizations that did not meet the organizational test receiving determination letters anyway. That is not a reassuring statistic if you are the organization. An approval that should not have been granted is not a shield. It is an exposure that surfaces at your first audit, your first major grant application, or your first state charity registration review.
There is also a practical funder dimension. Institutional funders, larger donors, and some government contracting offices have noticed that an EZ determination reflects almost no IRS scrutiny, and a subset of them ask for the application itself during due diligence. An organization with a thorough Form 1023 on file can hand over a document that explains its programs, budget, and governance coherently. An organization with an EZ has a two-page attestation and nothing else. That gap matters most at exactly the moment you are trying to grow.
The honest summary is that the EZ is appropriate for genuinely small, genuinely simple organizations that will stay that way for a while: a neighborhood arts group, a small scholarship fund, a volunteer-run animal rescue with no employees. Anything with paid staff, earned revenue, foreign activity, related-party arrangements, or realistic ambitions past the $50,000 threshold should file the full form. If the cost is the obstacle, a better answer than the EZ is often a fiscal sponsor for the first year or two, an arrangement our guide to fiscal sponsorship and compliance covers in detail.
Choosing between the two forms
Eligibility is the floor, not the decision
- Complete the eligibility worksheet honestly and keep it, whichever form you end up filing
- Project receipts against your actual plan, not against the threshold you would like to clear
- File the long form if you will have staff, earned revenue, insider transactions, or foreign activity
- Treat fiscal sponsorship as a legitimate alternative to a premature application, not a failure
What Has to Exist Before You Open the Application
Form 1023 assumes you are already an organization. That catches founders out, because the federal application is the part everyone talks about and the state formation is the part nobody mentions. The sequence is state first, then EIN, then the federal application, and doing it out of order creates rework.
State formation means filing articles of incorporation with the secretary of state or equivalent office in your state, or forming a trust or unincorporated association. Incorporation is the usual and generally advisable route, because it creates a separate legal entity and provides the liability protection that persuades people to serve on your board. Each state has its own fee, its own required provisions, and its own processing time, which ranges from same-day electronic filing to several weeks. The articles you file at this stage are the same articles the IRS will examine, so the language matters here rather than later.
Next comes the employer identification number, which every applicant needs regardless of whether it will ever have employees. It is obtained free from the IRS through the online application or Form SS-4, usually immediately. Two cautions apply. Do not apply for an EIN before you have decided on your legal name and structure, because changing it afterwards is administrative friction you do not need. And apply as the organization, using an authorized officer, never under a founder's personal identity in a way that blurs the line between the person and the entity.
Then there is the clock, which is the single most expensive thing founders do not know. The IRS generally requires that the exemption application be filed within 27 months from the end of the month in which the organization was formed. File within that window and recognition is generally retroactive to the date of formation, meaning contributions received since day one are covered. File late and exemption is generally recognized only from the filing date forward, leaving a gap during which donors have no deduction and the organization may have filing obligations it did not plan for. Groups that operated informally for years before deciding to formalize hit this constantly. If you are already past 27 months, that is a conversation to have with counsel before filing, not after.
Finally, state charitable solicitation registration is a separate obligation from federal exemption, required in most states before you ask the public for money, and it does not wait for your determination letter. Many founders assume the IRS approval handles it. It does not, and the requirements vary by state in ways worth mapping early. Our guide to multi-state charitable solicitation compliance covers how to track those obligations without a compliance department.
The order of operations
Each step depends on the one before it
- Incorporate in your state with articles that already contain the federally required provisions
- Adopt bylaws, seat an initial board, and hold an organizational meeting with recorded minutes
- Obtain the EIN in the organization's name through an authorized officer
- File the exemption application inside the 27-month window to preserve retroactive recognition
- Register for charitable solicitation in every state where you will actively fundraise
Organizing Documents and Bylaws: The Test You Pass on Paper
Before the IRS considers what you do, it considers what your organizing document says you are permitted to do. This is the organizational test, and it is mechanical. Your articles of incorporation must limit your purposes to one or more exempt purposes described in section 501(c)(3), and they must permanently dedicate your assets to an exempt purpose on dissolution. Two clauses, both required, and a surprising number of applications fail on them.
The purpose clause has to be limiting, not merely descriptive. Articles that authorize the corporation to engage in any lawful activity, standard boilerplate in many state templates, do not satisfy the test even if a charitable purpose is also mentioned, because the broad authorization permits non-exempt activity. The clause should state that the corporation is organized exclusively for charitable, educational, religious, or scientific purposes within the meaning of section 501(c)(3), then describe your specific purpose within that frame.
The dissolution clause must provide that on dissolution, assets are distributed for an exempt purpose, or to the federal government, or to a state or local government for a public purpose. Naming a specific successor organization is acceptable only if the document requires that the named organization be exempt at the time of dissolution and provides a qualified alternative if it is not. A handful of states supply this rule by operation of state law, in which case explicit language may not be required, but relying on that without confirming it for your state is a gamble with a $600 fee and several months attached.
Bylaws are a different animal. They are internal governance rules rather than the organizing document, and they are not where the organizational test is satisfied. They still matter to the application, because the IRS reads them for what they reveal about control. Bylaws that give the founder a permanent seat, a veto, or the power to appoint and remove every other director describe an organization controlled by one person, and that reads as private benefit risk regardless of intent. Bylaws that create a self-perpetuating board with staggered terms, a quorum requirement, and a real removal process describe an organization that can outlive its founder.
AI is genuinely useful here in a specific, limited way. Paste your draft articles and bylaws into a model and ask it to identify whether the purpose and dissolution language matches what section 501(c)(3) requires, to flag any provision that concentrates control, and to list internal inconsistencies such as a quorum defined two different ways. It is an excellent proofreader and a decent structural critic. What you must not do is let it generate your articles from scratch and file them, because state-specific requirements vary, models produce confident language that does not match your state's corporate code, and the document you are drafting is the constitutional foundation of a legal entity. Draft with AI assistance, then have a lawyer read it. That is a one-hour review, not a full engagement, and it is the cheapest insurance in this entire process.
What the IRS reads your documents for
Two required clauses, one control question
- A purpose clause that limits the organization to exempt purposes, not one that permits any lawful activity
- A dissolution clause permanently dedicating assets to an exempt or governmental purpose
- A board that can act independently of the founder, with real terms, quorums, and removal provisions
- Documents that are signed, dated, and, for articles, stamped as filed by the state
The Narrative Description of Activities, Where AI Helps Most
If one section decides your application, it is the narrative description of activities. The IRS asks you to describe your past, present, and planned activities completely and in detail, and its own guidance is explicit that you should describe actual or planned activities rather than possible or speculative ones. Everything downstream, the budget, the classification, the schedules, is checked against this narrative. When a reviewer sends a letter asking for more information, this is usually what prompted it.
The failure mode is abstraction. Founders write in mission language because mission language is what they have practiced for the pitch deck and the website. A narrative saying the organization will empower underserved youth through innovative programming tells a reviewer nothing they can evaluate. It does not say what happens, to whom, how often, run by whom, paid for how. The IRS is not judging whether your work is worthwhile. It is determining whether the described activity furthers an exempt purpose and does not confer private benefit, and it cannot do that against adjectives.
A usable narrative answers a fixed set of questions for every significant activity, and answers them the way you would explain it to a skeptical relative. What specifically happens. Who receives it and how they qualify. Who delivers it, whether staff, volunteers, or contractors, and whether any of them are related to each other or to board members. Where it happens and how often. How it is funded and roughly what share of your budget and staff time it consumes. Why it furthers a charitable or educational purpose. Practitioners generally recommend a substantial attachment rather than trying to squeeze this into the form field, and several pages is normal rather than excessive.
This is where AI earns its place, because the underlying problem is not that founders lack information. It is that the information is in their heads in the wrong shape. The workflow that works is dictation followed by structuring. Talk through each program for five minutes without worrying about form, transcribe it, then give the model the transcript with an instruction: reorganize this into a narrative that answers what, who, how, where, when, who delivers, how funded, and why it is charitable, using only facts present in the transcript, and list separately every question the transcript does not answer. That last clause is the valuable one. The gap list it returns is a reliable preview of the questions a reviewer would ask.
Then run the narrative back through for the opposite check. Ask the model to read it as an IRS reviewer looking for vagueness, unsupported claims, activities that sound commercial rather than charitable, and any benefit flowing to insiders. Models are good at this adversarial read, better than the founder who has lost perspective on their own text. The same technique works on the disclosure and narrative sections of the annual return once you are operating, a pattern our guide to writing Form 990 narrative sections with AI develops further.
The hard rule is that the model may only restructure facts you supplied. The instant it is allowed to make the narrative more compelling, it starts adding. It will invent a partnership with a school district, a number of people served, an evaluation framework, a sliding fee scale. These additions are plausible, well written, and false, and you are about to attest to them. Read every sentence of the final narrative and confirm you could substantiate it if asked, because being asked is a normal outcome, not a rare one.
A narrative workflow that holds up
Dictate, structure, stress test, verify
- Talk through each program aloud and transcribe it rather than starting from a blank page
- Instruct the model to use only supplied facts and to list every unanswered question separately
- Run an adversarial pass: where would a reviewer find this vague, commercial, or insider-benefiting
- Verify every factual assertion line by line before it goes near the submission
- Attach the narrative as a document rather than compressing it into the form field
Projected Budgets That Agree With Everything Else You Wrote
The financial data section asks established applicants for actual figures and new applicants for a projection covering the current year and the next two. Founders find this section absurd, and their instinct is understandable. You are being asked to forecast the finances of an organization that has not started. But the IRS is not grading your forecasting accuracy. It is checking whether the money you describe is consistent with the activities you described, and whether the revenue mix supports the public charity classification you requested.
Internal consistency is the whole test. If your narrative describes a tutoring program staffed by two part-time coordinators and your budget shows no salary line, something is wrong. If you claim to serve four hundred families and your total projected expenses are $9,000, a reviewer will want to understand how. If you describe a fee-charging program at length and your revenue is all contributions, the mismatch is visible. These are the contradictions that generate follow-up letters, and almost all of them are avoidable by reading the two sections side by side before filing.
Build the projection from the activities rather than from a target. For each program in your narrative, estimate what it costs to run at the scale you described: personnel with a realistic understanding of employment costs, space, insurance, supplies, technology, professional fees for the accountant you will need. Add the administrative costs founders systematically forget, including annual state registration renewals, directors and officers insurance, bookkeeping, and the audit or review that a funder may require. Then build revenue against a plan you could actually describe, naming sources by category rather than inventing precision you do not have.
AI is well suited to the mechanics of this. It will restate a narrative as a cost structure, catch arithmetic errors, flag categories you have omitted, and compare the budget against the narrative line by line to surface mismatches. Asking a model to review a draft projection as an IRS reviewer, listing every place the budget and narrative disagree, is one of the highest-value five minutes in the process. What it cannot do is know your local salary market, your state's insurance requirements, or what your fiscal sponsor charges. Feed it real numbers, and use it to check coherence rather than to supply facts. The same discipline applies to operating budgets once you are running, as our guide to AI-assisted budget forecasting explores.
One more thing worth saying plainly. Conservative projections are not a weakness in this document. An application projecting $60,000 in year one with a clear explanation beats one projecting $600,000 with no basis, because the second invites questions about how, from whom, and under what arrangement. Overstating your first year also has a downstream consequence, since your actual figures will appear on your first annual return and a very large gap between the projection and reality looks careless in a document anyone can read.
Budget errors that generate letters
Every one of these is caught by a side-by-side read
- Staffed programs in the narrative with no corresponding compensation in the budget
- Service volumes that cannot be delivered for the amounts projected
- Fee-based activity described in the narrative but missing from the revenue lines
- Revenue mix that contradicts the public charity classification being requested
- No line for insurance, professional fees, state registrations, or accounting support
Public Charity or Private Foundation, and the Support Test Behind It
Every 501(c)(3) organization is a private foundation by default unless it establishes otherwise, and the application is where you make that case. The distinction is not cosmetic. Private foundations face an excise tax on investment income, mandatory annual distribution requirements, strict rules on self-dealing and excess business holdings, a lower charitable deduction ceiling for donors, and the longer Form 990-PF every year. Nearly every organization running programs wants public charity status, and the route there for most is demonstrating broad public support.
There are two principal support tests and they behave differently. The test under section 509(a)(1), applying to organizations supported mainly by gifts, grants, and contributions, generally requires that at least one third of total support come from the general public and governmental units, with a facts and circumstances alternative available down to ten percent. The test under section 509(a)(2), which fits organizations earning a substantial share of revenue from program service fees such as tuition, ticket sales, or membership dues, requires more than one third public support and imposes a second hurdle: not more than one third of support may come from investment income and unrelated business taxable income. A third route, section 170(b)(1)(A) status by activity, covers churches, schools, and hospitals without any support calculation at all.
Choosing the wrong test is a real error with a delayed cost. An organization whose revenue will be mostly earned fee income but which requests 509(a)(1) status has set itself up to fail the calculation later, because the support test is recomputed on Schedule A of the annual return over a rolling five-year window. A new organization does not have to pass it immediately. It receives a determination on the basis that it can reasonably be expected to be publicly supported, and it has its first five tax years to establish the record. Founders often misread that grace period as permanent and are surprised when year six arrives.
The practical planning point is that the test punishes concentration. A single very large gift from one donor can push an organization past the limit on how much support from any one source counts toward the public support fraction, and an organization funded almost entirely by its founding family will not pass on gift volume alone. Sustained failure has consequences: an organization that falls below the threshold for two consecutive years can be reclassified as a private foundation, which changes its obligations rather than merely its label. The answer is to broaden the donor base early and to run the calculation annually rather than discovering it at filing time.
AI can model this competently, and it is worth doing before you choose. Give a model your projected revenue by source and ask it to walk through how each line would be treated under both tests, then show the resulting ratios. Treat the output as a thinking aid rather than an answer, since the rules on unusual grants, related donors, and support from governmental units carry exceptions a general model will not apply reliably. Looking at how comparable organizations in your field are classified is useful context too, and the returns are public, a research approach our guide to reading other nonprofits' 990s with AI lays out.
Classification questions to settle before filing
These decisions follow you for five years and beyond
- Will revenue come mainly from contributions and grants, or from fees for services you provide
- How concentrated is the expected donor base, and what happens if one large gift dominates a year
- Do you qualify by activity as a church, school, or hospital, which avoids the calculation entirely
- Who will run the support test each year once the first five tax years have elapsed
Compensation, Conflicts, and the Private Benefit Questions
A long stretch of the application concerns compensation and financial arrangements with officers, directors, trustees, substantial contributors, and people related to them. The questions can feel intrusive, and founders sometimes answer defensively as a result. They are asking about two specific doctrines. Inurement means no part of net earnings may benefit an insider, and it is absolute. Private benefit is broader, covering more than incidental benefit to any private individual or entity. Both are grounds for denial, and both are most often triggered by arrangements that seemed obviously fine to the people making them.
Startup nonprofits generate these arrangements naturally. The founder will be the paid executive director. The organization rents space from a board member at a favorable rate. The treasurer's accounting firm does the bookkeeping. The marketing work goes to a director's agency. None of these is automatically disqualifying, and the application does not ask you to avoid them. It asks you to disclose them and to describe how you determined the terms were reasonable and arrived at without the interested person participating in the decision. Disclosed and properly approved is a very different posture from disclosed and casually arranged, and undisclosed is worst of all.
This is why the conflict of interest policy question carries weight beyond its size. The IRS supplies a sample policy in an appendix to the Form 1023 instructions, and adopting it is not required but is close to standard practice. What matters is that the policy is real: adopted by the board before the arrangements it governs, requiring annual written disclosure, and requiring that an interested person be absent from the deliberation and the vote on anything in which they have a financial interest. The procedural discipline is the substance. A policy on file that nobody follows produces minutes showing the executive director present for the vote on their own salary, which is the exact fact pattern the question exists to surface. Our guide to managing conflict of interest disclosures with AI covers how to run that annual cycle without it lapsing.
Compensation itself should be set through a defensible process rather than by a founder naming a figure. That means comparable data from similar organizations of similar size in a similar location, a decision made by board members with no financial interest in the outcome, and contemporaneous documentation in the minutes. The data is available in public filings and sector compensation surveys, and this is a reasonable research task to accelerate with AI, provided you verify the sources rather than accepting summarized figures. Board composition is the related lever: a board with a majority of members unrelated to each other and to staff makes every one of these answers easier, which is one reason building it deliberately, as our guide to nonprofit board recruitment describes, is worth doing before you file rather than after.
Getting the insider questions right
Disclosure plus process, not avoidance
- Disclose every arrangement with an officer, director, substantial contributor, or their relatives
- Adopt a conflict of interest policy before the transactions it is meant to govern, not after
- Set compensation using comparable data, approved by disinterested directors, recorded in the minutes
- Recruit a board majority unrelated to staff and to each other before you file
Submitting Through Pay.gov, Then Waiting Well
Both forms are filed electronically through Pay.gov, and the mechanics deserve a rehearsal. You create an account, complete the form in the browser, upload your attachments as a single PDF, and pay the user fee by bank account or card. Assemble and name the PDF before you start, since the session is not a comfortable place to discover that your articles are a photograph of a printout. Save a complete copy of everything submitted, because you will be asked for it by funders, by state registration offices, and by the IRS itself if questions arise. The application, once approved, is a public document, so write it knowing it will be read outside the IRS.
Review the schedules before you conclude you are done. Additional schedules apply to particular categories of applicant, including churches, schools, hospitals and medical research organizations, supporting organizations, organizations providing low-income housing, those granting scholarships, and applicants filing late or requesting reinstatement after automatic revocation. Each has its own substantive questions, and missing one that applies to you is a straightforward way to earn a delay.
Timelines vary widely and you should set expectations accordingly. The IRS publishes current status information on its where is my application for tax-exempt status page, including the date range of applications currently being assigned to reviewers, and it is the only status source worth checking, since Pay.gov does not track progress and the IRS does not send updates. Clean applications that need no follow-up are sometimes resolved quickly. Applications that raise questions take substantially longer, and a year is not unheard of. Plan the organization's first year assuming the letter may not arrive quickly.
Waiting well means understanding what you may and may not do. You can generally operate, and you can generally accept contributions, though donors are relying on your expectation of retroactive recognition and you should say so honestly rather than implying a determination you do not yet have. Do not claim to be a 501(c)(3) organization while the application is pending. Do keep proper books from day one, because the financial record you build during the wait is the record you will report later. And be aware that an annual return may come due before the determination arrives, which catches organizations that assumed the clock starts at approval.
Then there is the letter. A request for additional information, often called a developmental letter, is a routine part of the process rather than a rejection. It typically arrives with a short deadline, frequently around two weeks, and asks pointed questions about the things this article has been warning about: vague activity descriptions, a budget that does not match the narrative, control concentrated in one family, an undisclosed relationship, a purpose clause that does not limit. Answer each question directly and in order, provide only what was asked, and do not use the response to restate your mission. If the questions concern classification, private benefit, or unrelated business activity, that is the point to bring in counsel if you have not already, because the answers become part of the record on which exemption is granted. Missing the deadline can close the case, forfeiting the fee and requiring a fresh application.
Submission and the months afterward
Prepare the package, then plan for the wait
- Assemble attachments into one clean, legible PDF before opening the online form
- Check every schedule for applicability, including the late-filing and reinstatement schedules
- Keep a full copy of the submission; the approved application becomes publicly available
- Watch the mail for a request for additional information and diarize its deadline immediately
- Answer follow-up questions precisely and in order, adding nothing that was not requested
Where AI Belongs in This Process, and Where It Does Not
Form 1023 is an unusually good match for AI assistance in one respect and an unusually bad one in another, and holding both ideas at once is the whole skill. The good match is that it is a structured document requiring a founder to say clearly what they already know, and models are excellent at imposing structure, finding gaps, checking internal consistency, and reading a draft adversarially. Small founding teams with no development staff and no budget for a full legal engagement can produce a far better application with that help than without it.
The bad match is that the document is an attestation. The person signing declares under penalties of perjury that they have examined the application, including accompanying documents, and that it is true, correct, and complete. That standard does not soften because a model wrote the sentence. An inaccurate application is not a writing problem to be corrected in the next draft, it is a statement to the federal government about facts, and the consequences run from a follow-up letter through denial to revocation of a determination obtained on a misstated basis. Read every word you are about to swear to, and be able to point to where each claim comes from.
A second limitation is subtler and catches sophisticated users. Models are unreliable on current, jurisdiction-specific tax and corporate law. They will cite revenue procedures that have been superseded, quote thresholds from an earlier version of the form, describe state incorporation requirements that belong to a different state, and reason confidently from all of it. Anything a model tells you about a rule should be verified against the current IRS instructions or the relevant state office before you act on it, and a model should never be your source for whether a given structure qualifies. The same caution our guide to using AI in nonprofit legal review sets out applies with extra force to formation, because there is no counterparty to catch your error.
The cost argument for skipping professional help entirely is the one worth answering directly, because it is the argument AI makes most tempting. Exempt organizations counsel is expensive, and a founding group scraping together the user fee may not have it. But the realistic comparison is not a lawyer against no lawyer. It is a full engagement against a targeted review, where you do the drafting with AI assistance and pay for a few hours of someone reading your articles, your classification choice, and your insider arrangements. That is the shape most small founders can afford, and it addresses the three areas where errors are expensive and hardest to unwind later. Building a working relationship with AI tools on this kind of structured, high-stakes documentation is also good practice for what comes next, a progression our guide for nonprofit leaders starting with AI maps out.
Hand to AI
Structure, drafting, and checking
- Turning a dictated program explanation into a structured activity narrative
- Listing the questions your draft leaves unanswered before a reviewer asks them
- Reconciling the budget against the narrative and catching arithmetic and category errors
- Proofreading articles and bylaws for missing clauses and internal contradictions
- Drafting a first response to a request for additional information, question by question
Keep with a person
Judgment, facts, and the signature
- Whether to file the full form or the streamlined version, and which support test to request
- Final review of articles of incorporation against your state's corporate code
- Any question touching insider compensation, private benefit, or unrelated business activity
- Verification of every factual claim, since the signer attests under penalties of perjury
- Any rule, threshold, or citation a model asserted, checked against current IRS instructions
Conclusion
Most applications that go wrong do not go wrong on hard questions of tax law. They go wrong because the narrative was abstract, the budget did not match it, the articles used a state template that permitted any lawful activity, a founder was on both sides of a transaction nobody disclosed, or the organization chose the streamlined form because it was faster rather than because it fit. Every one of those is a preparation failure, and every one is visible to anyone who reads the application carefully before it is filed.
That is the honest case for AI here. It makes careful preparation achievable for a founding group that does not have a development office or a retained lawyer. It turns a founder's spoken explanation into a narrative a reviewer can follow, tells you which facts you have not supplied, reconciles the budget against the programs, and reads your own draft back to you the way a skeptical stranger would. That is not a small contribution. It is most of the distance between an application that gets a letter and one that gets a determination.
What does not change is where responsibility sits. The signature on the application is a declaration under penalties of perjury that the contents are true, correct, and complete, and no amount of drafting assistance transfers that. Use AI to organize and to check. Use a lawyer or an exempt organizations CPA for the classification choice, the organizing documents, and anything that touches insiders. Then read the whole thing once more before you submit, slowly, asking of every sentence whether you could prove it. If you can, you have a defensible application. If you cannot, you have a problem that is far cheaper to fix now than in the letter that follows.
This article is general information about a federal process, not legal or tax advice, and formation questions turn on facts and on the law of your state. Treat it as preparation for a conversation with a professional rather than a substitute for one.
Start the Organization on Solid Ground
We help founding teams and early-stage nonprofits use AI to prepare thorough, consistent documentation, so the time you spend with professional advisers goes to judgment calls rather than typing.