The Board Pack Is Broken: 226 Pages, 4 Hours, and Now an AI Reading Over Every Director's Shoulder

The board pack has quietly become the least-read, highest-stakes document a company produces. It carries the quarter's strategy, the risks the executive team chose to surface, and the questions the board is meant to interrogate — and almost nobody reads it the way it was written to be read. That was a tolerable problem when the only reader was a tired director skimming on a Sunday night. It is not tolerable now that a second reader has joined the room: the AI that director is using to digest it, and the AI on the outside summarizing whatever you disclose before a human ever sees it.

Both sides of the boardroom table went machine-assisted in the space of about eighteen months. The board pack did not change to meet them.

The Pack Got Longer As Attention Got Shorter

Start with the artifact itself, because the numbers are worse than most executives assume. The average board pack now exceeds 226 pages — a 30% increase since 2019 — and Board Intelligence has documented packs that stretch past 1,000 pages, with directors spending roughly four hours trying to digest a single one. More pages have not bought more clarity. The opposite: only 36% of directors found their board pack useful in 2024, down from 48% the year before, and just 13% rated it "extremely effective." Fewer than one in five — 17% — said their pack effectively communicated the key messages it existed to communicate.

That is a document failing at its one job. And the people it fails are the people with the least slack to absorb the failure: 35% of US directors report not having enough time to review board materials thoroughly. The board pack has become a place where critical information goes to be technically disclosed and practically ignored — buried on page 140, in a chart with no headline, inside an appendix nobody reaches.

The instinct, when a board misses something, is to add a section. That instinct is exactly how you get to 226 pages. Length is not thoroughness. It is the absence of an editorial decision about what matters — and on a board pack, the absence of that decision is itself a governance failure.

Now There Are Two New Readers In The Boardroom

Here is what makes 2026 different from every prior year of board-pack bloat. The director is no longer reading alone.

Two-thirds of directors — 66% — now use AI for board work, according to McKinsey, pasting board materials into a model to get the summary the pack should have led with. Digital transformation and AI now top the board agenda outright: 65% of directors name it their single most-cited priority for the year ahead in the What Directors Think 2026 survey. The board pack is no longer being read by a human and a highlighter. It is being read by a human, a highlighter, and a large language model that will confidently compress your 226 pages into five bullets — and if your key message wasn't stated plainly, the model will pick the wrong five.

The same thing is happening on the outside, where the stakes are public. As one IR analysis of the 2026 proxy season put it, "poorly organized or ambiguous disclosures can be misrepresented by AI-driven summaries before they ever reach an analyst." The first reader of your earnings language, your shareholder letter, your governance disclosure is increasingly not a person — it is a model an analyst, a proxy advisor, or a retail investor pointed at the filing. Retail ownership is growing fast, and that audience expects communication that is accessible, not written for institutional consumption. AI-ready disclosure is no longer a sophistication; it is table stakes for not being misquoted by a machine.

So the board pack and the disclosure now share a brutal new requirement: they have to survive being read by an AI that has no charitable interpretation, no memory of last quarter's context, and no instinct to extend benefit of the doubt. Whatever the page says on its own is what the machine carries forward. There is no narrator to recover the moment.

What An AI-Readable Board Pack Actually Looks Like

The fix is not "use AI to write the pack faster." Plenty of tools do that, and a faster route to 226 mediocre pages is not progress. The fix is structural, and it is the same discipline that makes a page survive a skeptical human: every page leads with its conclusion, every number is traceable to a source, and every claim that matters is stated where it can't be missed.

An AI-readable page is an answer-first page. The headline is the takeaway — "Net retention fell to 104%, driven by two enterprise churns we flagged in Q2" — not "Customer Metrics Update." A model reading that page extracts the right signal because the right signal is the first thing on it. A human skimming it does the same. The properties that make a board pack legible to AI are identical to the properties that make it legible to a director with four hours and a long flight: state the message, support it with a verifiable figure, and cut the page that doesn't do both.

This is where the editing matters more than the drafting. A board pack earns trust by what it leaves out — and that is precisely the decision a generation tool won't make for you, because generation optimizes for plausible completeness, not for the hard call about what the board actually needs to decide this quarter.

The Risk No Board Can Carry

There is a reason board and IR communication can't be handed to an autonomous drafting tool and waved through. The governance numbers say the guardrails aren't there yet. Only 39% of Fortune 100 boards have any form of AI oversight — a committee, a director with real AI expertise, an ethics process — per McKinsey, and while two-thirds of directors are using AI for board work, only 22% have any governance process guiding that usage. The appetite raced ahead of the controls.

That gap is dangerous specifically because board materials are zero-tolerance documents. A hallucinated metric in a sales deck costs you a deal. A hallucinated metric in a board pack, or a number in the directors' summary that contradicts the figure in the appendix, is a fiduciary problem — the kind that surfaces in a deposition, not a deal review. When the board reads with AI and the AI was fed a pack with an internal contradiction, the contradiction doesn't get caught. It gets summarized, smoothed over, and carried into the decision.

So the durable model for high-stakes board and IR communication is not agent-writes-and-sends. It is the same three-layer shape that holds anywhere the cost of being wrong is the whole game: a system drafts the materials and keeps them current, a layer of review experts — grounded in the company's own numbers, prior disclosures, and brand — checks every page for the contradiction, the unsupported figure, the off-message claim, and then a human signs off. The drafting buys back the hours. The review buys back the risk. The board keeps the authority. Autonomy was never the goal for a document that goes to your directors.

What Changes For Board And IR Teams

If you produce board packs or run investor communication, the practical shift for 2026 is to stop optimizing for completeness and start optimizing for legibility under machine reading. Write every page assuming an AI will read it first and a busy director will read it second, and that neither will extend you the benefit of the doubt. Lead with the conclusion. Make every number traceable. Treat any page that can't state its own key message as a page that's failing — the same standard a senior chair would hold, applied to all 226 pages instead of the four anyone has time to scrutinize.

And check it before it ships, page by page, against everything the company already knows — because the SEC is fast-tracking a move to semi-annual reporting, retail investors are voting in numbers they never did, and proxy advisors are pointing models at your language the moment it's public. The board pack and the disclosure are being read by machines that quote you exactly. The only protection is to be exactly right, on every page, before you send.

The board pack stopped being a formality the moment AI started reading it. Build the one that survives the second reader, and the first one — the human who actually decides — finally gets a document worth the four hours.

— The Lurio Team