Bylaws Nobody Has Read Since 2011: An AI-Assisted Governance Document Review

How Bylaws Drift, and What the Drift Actually Costs

Drift is rarely a single decision. It accumulates. A board adds a development committee without amending the bylaws because the bylaws seemed like a formality. An executive director gains contracting authority through practice because someone had to sign the lease. A board that once had eighteen members shrinks to nine, leaving a quorum requirement that was reasonable at the old size and is now punishing. Meetings move to video and stay there. Notice moves from mailed letters to a calendar invite. Each step is sensible on its own, and none of them are reflected in the document that governs the organization.

The first real cost is decisions made outside authority. Bylaws typically allocate specific powers: which decisions require full board approval, which can be delegated to a committee, what dollar thresholds trigger board review, who may sign what. When the actual allocation differs, the organization has a governance gap that is invisible until someone looks. A committee that approves a major expenditure it was never empowered to approve has not committed fraud, but it has taken an action whose validity is genuinely questionable, and unwinding it later is expensive and embarrassing.

The second cost is quorum. This is the most common and the most consequential drift, because a vote taken without a quorum may simply not be a valid board action. Bylaws written for a large founding board sometimes set quorum at two-thirds of directors, or at a fixed number that no longer bears any relationship to board size. Boards in that position develop workarounds, counting proxies the bylaws do not permit, treating an email thread as a vote, or recording attendance loosely. Each workaround creates a category of decisions whose validity is arguable. If meetings are routinely delayed over attendance, or decisions are being made through informal channels between meetings, that is the symptom, and the bylaws are the cause.

The third cost is terms nobody tracks. Bylaws set director terms and sometimes limits, and in a surprising number of organizations no one maintains the roster that would reveal that four directors are serving in years that expired, or that the board has exceeded its own maximum size, or that an officer has held a position past the limit the bylaws impose. This matters most at exactly the wrong moment: when a contested question arises and someone asks who is actually entitled to vote. A board whose composition cannot be reconciled to its own bylaws is in a weak position to resolve anything.

The fourth cost is committees that exist only on paper, and their mirror image, committees that exist only in practice. Bylaws frequently enumerate standing committees, prescribe their composition, and sometimes grant them authority to act for the board. If the audit committee named in the bylaws has not met in five years, the organization is not doing something its own rules require. If the program committee doing real work was never established in the bylaws or by board resolution, its authority is unclear. Committee composition rules deserve particular attention, since some states restrict which committees may exercise board authority and who may serve on them.

The fifth cost is contradiction between documents. A nonprofit's governance is set by a stack: the articles of incorporation filed with the state, the bylaws, and board-adopted policies, in roughly that order of authority. Venable's widely cited survey of common nonprofit bylaw pitfalls notes that purpose clauses in bylaws frequently differ materially from the articles, and that bylaws are often patch-worked together over time rather than comprehensively reviewed. When the articles say one thing, the bylaws say another, and the board has adopted a policy saying a third, nobody can answer a simple question about what the organization is required to do.

The Review Workflow: Six Passes Through the Documents

Do not ask a model to review your bylaws. That prompt produces a bland summary and a list of generic suggestions. Instead run six distinct passes, each with a narrow question and a structured output. Separating them matters, because each pass produces a different kind of artifact and each one can be checked independently. A single mega-prompt produces output you cannot verify.

Pass one: extract every obligation into a checklist. Ask the model to read the bylaws and produce a table with one row for every obligation the document imposes, capturing the section reference, who bears the obligation, what exactly is required, any deadline or frequency, and any threshold or number. This turns a narrative document into an operational list: the board must meet at least quarterly, notice must be given at least ten days in advance, the treasurer must present financial statements at each regular meeting, officers are elected annually at the meeting following the annual meeting of members, and so on. Most boards have never seen their own bylaws in this form, and the list alone is frequently the most useful output of the entire exercise. Insist on section references for every row so each claim can be checked against the source.

Pass two: test the checklist against the minutes. Now give the model the obligation list and the last two years of minutes, and ask it to mark each obligation as evidenced, contradicted, or not evidenced, citing the meeting date for each. This is the pass that converts suspicion into findings. It surfaces the quarterly meeting requirement that was met three times last year, the officer election that does not appear in any minutes, the annual budget approval that happened in March when the bylaws say it happens before the fiscal year begins, and the standing committee that never reports. Be precise about the distinction between contradicted and not evidenced. Thin minutes are a different problem from a violated rule, and conflating them will send the board chasing ghosts.

Pass three: flag internal contradictions. Ask the model to identify every place where the bylaws contradict themselves, contradict the articles of incorporation, or contradict a board policy or committee charter. This is close-reading work at a scale humans do poorly. Classic finds include a purpose clause in the bylaws that differs from the articles, an amendment provision in one article that conflicts with the amendment reference in another, a quorum for the membership stated differently in two places, and a committee charter granting authority the bylaws reserve to the full board. Require the model to quote both conflicting passages side by side rather than merely asserting the conflict, so a human can confirm it in seconds.

Pass four: check against the state nonprofit corporation act. Ask for a list of provisions that appear inconsistent with the nonprofit corporation act of your state of incorporation, and explicitly ask the model to separate mandatory statutory requirements from default rules the bylaws are permitted to vary. Treat this output as a question list for counsel, not a conclusion. Statutory citations are the single area where models are most confidently wrong, and state acts have been amended recently in several states. What this pass is genuinely good for is producing a focused set of questions: does our state require an annual meeting, does it permit unanimous written consent if our bylaws are silent, does it cap or require a minimum board size, may committees exercising board authority include non-directors.

Pass five: identify what is missing. Ask what provisions are absent that state law, the IRS, funders, or standard practice would expect to see. Common gaps include no indemnification provision, no process for filling board vacancies, no removal procedure for directors or officers, no conflict of interest article cross-referencing the policy, no electronic meeting or notice authorization, no record-keeping provision, and no clear statement of whether the corporation has voting members. The dissolution clause deserves a specific check, since the IRS looks for asset dedication language as part of the organizational test, and it belongs in the articles rather than only in the bylaws.

Pass six: write the plain-language summary. Most board members will never read the bylaws, and the amendment discussion will go badly if the only available reference is the legal text. Ask for a two to three page plain-language summary organized by the questions board members actually have: how many of us are there and how long do we serve, how do we call a meeting and who has to be there, what can we decide by email, what needs a special vote, what are the officers responsible for, and how do we change any of this. Pair it with a one-page findings memo listing the top issues in priority order. This is the same discipline that makes board meeting packets readable, and it is the difference between a review that produces a decision and one that produces a deferral.

State Law Is the Controlling Authority, and Bylaws Cannot Override It

The most important conceptual point in this whole exercise is one that boards routinely get backwards. The bylaws are not the top of the hierarchy. The nonprofit corporation act of the state where the organization is incorporated is, and a state statute supersedes any conflicting provision of the bylaws. A bylaw that permits something the statute prohibits is simply unenforceable, and the board that relies on it is relying on nothing. Underneath the statute sit the articles of incorporation, then the bylaws, then board-adopted policies. Every finding from your review should be sorted into that hierarchy before anyone proposes a fix.

State acts work through a mix of mandatory rules and default rules, and understanding the difference is what makes a review useful. A mandatory rule applies regardless of what your documents say. A default rule applies unless your articles or bylaws provide otherwise, which means silence in your bylaws is itself a choice, and often a choice nobody made deliberately. The Revised Model Nonprofit Corporation Act, which many states have drawn on, illustrates the pattern: board action without a meeting by written consent is permitted unless the articles or bylaws provide otherwise, and directors may meet by any means of communication through which all participants can simultaneously hear each other. Where your state follows that pattern, your silence gets you the default. Where your bylaws say something different, your bylaws control, assuming the statute allows variation.

This landscape has been moving. Several states have modernized their nonprofit corporation acts in recent years, generally in the direction of authorizing electronic communication and notice, expanding remote meeting options, clarifying written consent, and aligning more closely with the Model Act. Florida's overhaul of Chapter 617, described in coverage from The NonProfit Times, updated default rules on director participation by communications technology, waivers of notice, unanimous written consent, remote member participation, and recordkeeping for remote votes. Washington enacted a substantially rewritten nonprofit corporation act earlier in the decade, and New York and Illinois have made their own revisions. If your bylaws predate your state's most recent revision, some of your provisions may be restating a rule that no longer exists.

There is a practical consequence for how you write the amended bylaws. Restating statutory text inside your bylaws feels helpful and creates a maintenance problem, because the statute changes and your bylaws do not. Where the statute already provides the rule you want, the cleaner approach is often to reference the governing law rather than reproduce it, or to say nothing and let the default apply. Reserve specific bylaw language for the places where you genuinely want to depart from the default, and make the departure explicit enough that a future board can see it was intentional.

Multi-state operation adds a layer worth flagging. Your state of incorporation governs your internal affairs, but the states where you solicit donations or operate programs impose their own registration and reporting obligations, which is a separate compliance track from your bylaws. Do not let a governance review blur into a charitable registration review. They both matter, and confusing them produces a document that tries to do two jobs and does neither well.

Where AI Output Stops and Counsel Begins

Everything described above produces a package: an obligation checklist, a practice comparison, a contradiction log, a question list, a gap list, and a plain-language summary. That package is a first draft for a nonprofit attorney. It is not a legal opinion, it is not an amended set of bylaws, and treating it as either is the way this exercise goes wrong.

Be specific about why. Models are strong at reading comprehension across long documents, at structured extraction, and at consistent comparison, which is why passes one through three produce genuinely reliable output that a board member can verify in an afternoon. Models are weak at exactly the thing that matters most here: stating what a particular state's statute requires today. Statutory citations are among the most common places for a model to be fluent and wrong, state acts differ meaningfully from one another, several have been amended in recent years, and the model has no way to know which version it absorbed. A confident paragraph about what your state requires is not evidence about what your state requires.

Judgment is the other boundary. Whether to shorten terms, whether to become a membership organization, how much authority to delegate to the executive committee, and how to balance nimbleness against deliberation are decisions about what kind of organization you want to be. A model can lay out the tradeoffs clearly and should be asked to. It cannot make the choice, and board discussions that begin from an AI recommendation rather than from the tradeoffs tend to skip the conversation the board most needed to have.

Then there is the drafting itself. AI can produce proposed amendment language, and that language is useful as a starting point for counsel. It should not be adopted as written. Bylaw drafting is precision work where a single ambiguous phrase creates the dispute the document was supposed to prevent, and where the interaction between provisions matters as much as any provision alone. An attorney reviewing and revising a coherent draft is a far smaller engagement than an attorney starting from nothing, which is the real economic argument for doing the review this way. Organizations already using AI for legal and contract review will recognize the pattern: the model prepares and organizes, the lawyer decides and signs.

A confidentiality note belongs here too. Bylaws and articles are generally not sensitive, and much of the content is already public. Minutes are a different matter. Board minutes can contain personnel discussions, litigation matters, donor information, and executive session content. Before uploading minutes to any AI tool, confirm what the tool does with your data, whether inputs are used for training, and what your own retention policy says. The safe approach for most organizations is a business tier tool with training disabled, plus redaction of executive session content before upload. This is the same diligence that belongs in your records retention schedule, and it is worth settling before the review rather than during it.

A Cadence That Keeps This From Happening Again

The reason bylaws end up fifteen years stale is that nobody owns them. Review is not on any calendar, it belongs to no committee, and the person who would notice the problem is the person who never reads the document. The fix is a modest, named, recurring commitment rather than a heroic one.

A practical rhythm has three layers. Annually, the governance committee or board chair runs a short compliance check: are we meeting as often as the bylaws require, did officer elections happen when they should have, does the roster reconcile to the terms, do the committees in the bylaws match the committees that met, and were any actions taken outside normal process. This is a thirty minute agenda item once you have the obligation checklist from your review, because the checklist is a reusable asset and the comparison against the year's minutes is precisely the work AI does quickly. Practitioners commonly suggest a light annual look of roughly this kind.

Every two to three years, run the fuller review: repeat all six passes, check for changes in your state's nonprofit corporation act, and take anything substantive to counsel. This interval matches what many advisors recommend and it maps well to how quickly practice drifts. Between the annual check and the triennial review, most organizations catch problems while they are still small enough to fix with a single amendment.

Then there are event triggers, which override the calendar. Review the bylaws whenever the board changes size meaningfully, whenever you hire a first executive director or change the executive structure, whenever you add or eliminate a standing committee, whenever you begin operating in a new state, whenever you take on a major government contract or a funder with governance requirements, whenever your state amends its nonprofit corporation act, and whenever a merger, affiliation, or fiscal sponsorship arrangement is contemplated. Each of these changes the organization the bylaws describe.

Assign the ownership explicitly. A governance committee is the natural home where one exists, and the board chair or secretary where it does not. Put the annual check on the board calendar in the same month every year, ideally shortly before the 990 is prepared so the findings feed the return. Keep the obligation checklist, the plain-language summary, and the current bylaws in one place, and hand all three to every incoming director. The broader point is the one that applies to every governance document: the rules that get followed are the rules someone reads, and the way to make bylaws readable is to maintain a current plain-language version alongside the legal text. If your organization is also building out its policy library, put the bylaws review on the same review rhythm so the whole stack stays consistent instead of each document drifting on its own schedule.