AI and board papers: quality standards and lines of accountability

The executives and senior managers who write board papers now draft them with help from AI. An author who uses it well produces a cleaner first draft and spends the time saved on the thinking. An author who uses it badly produces twelve confident pages that answer nothing the board asked.

Boards feel the difference in paper quality. In Board Intelligence's Summer 2026 Board Value Index, UK directors ranked information quality as the single biggest barrier to board decision-making, ahead of process rigidity and skills.

Board Intelligence built the Question-Driven Insight (QDI) Principle on work with the boards and leadership teams of more than 3,000 organisations. It asks three things of a paper.

Critical thinking. The author asks the hard questions and structures the paper around them. How are we performing? Why? What are the implications? What are we doing about it? Where does this leave us? A director reading a paper built that way sees management taking a position and showing judgement, with good news and bad news both present, and with the options management considered and rejected set out alongside the recommendation.

Great communication. The author gets to the point early and writes plainly. A strong executive summary carries enough that a reader could contribute to the discussion from the summary alone. Charts carry one message each and are labelled.

Directors want confidence that the information in front of them does what the board needs it to do. When a paper arrives polished and says very little, a director reasonably wonders how it was produced. Some have started asking in the meeting whether AI drafted it.

Nobody can detect AI authorship reliably, and a board that spends meeting time on authorship has less time for the decision.

Directors face two risks from papers like these: information they cannot rely on, and weaker trust between the board and management. That trust thins quickly once a chair suspects the executive of sending filler. A chair who reaches that point has a reporting standard to agree with the executive, and a conversation to have about what the board expects.

Some organisations now ask authors to declare AI use. A board that requires the declaration learns how a paper was produced. A board that applies the three QDI tests above learns whether to rely on it. Both are useful, and each board decides on disclosure for itself.

Boards should oversee AI the way they oversee cybersecurity. Once AI sits inside a material control, the board should review it at regular meetings through the year, with a named executive accountable for each use.

Directors have started this work. In Board Intelligence's Summer 2026 Board Value Index, 49% of directors globally said their board is actively reviewing which decisions should remain human-led as AI becomes more capable. UK directors trail at 42%, against 58% in the Middle East.

Fewer boards have the arrangements to support that review. Board Intelligence's Board AI Readiness Radar, based on submissions from 47 organisations across 17 countries in May 2026, found that 11% of boards meet a relatively conservative threshold for AI readiness, with the average organisation scoring 10.3 out of 30. The two widest gaps are infrastructure, where roles, responsibilities and risk frameworks have not changed to reflect AI, and information, where boards are not getting what they need to oversee AI-related decisions.

The simplest format for AI oversight is a short standing report from the accountable executive, covering:

• AI uses added or changed since the last meeting
• incidents, including errors the business caught and how it corrected them
• any change in who is accountable for each use
• decisions the board needs to take

Directors then question the report as they would any other risk paper. The chair decides how often it comes up, which should be more often than the annual deep dive it replaces.

A general-purpose AI tool writes whatever an author asks of it. It holds no view on what a board needs, so an author who feeds it thin material gets a well-organised version of thin material back.

Board Intelligence built Report Writer, an AI writing tool, for board papers specifically. It starts an author with the questions decision-makers need answered and structures the paper around them. As the author writes, it spots gaps in the argument, pulls insights out of the supporting documents, and suggests improvements to the writing. The questions come from 20 years of governance consulting work, which is the part a general-purpose tool has no access to.

Authors use Report Writer inside Microsoft Word, so nobody has to learn a new system.

"It has transformed the quality of thinking and writing in our reporting, and it's helping us have more focused and productive conversations."
Tim George, Deputy Company Secretary, Essentra

Authors starting a decision paper can follow Board Intelligence's guide on how to write a great business case.

The company secretary reads every paper before it reaches the board, which makes that the point where a weak paper gets caught. The chair sets what the board will accept. An author who gets a paper back at this stage still has time to answer the question it missed.

Three things are worth doing with a paper that falls short:

• Compare it with the brief its author received and send it back with specifics.
• Name the question it has not answered, such as what the board is being asked to decide.
• Tell the chair when an item is not ready, so the board does not find out in the meeting.

A director reads a capital investment proposal. The paper is a good one: it states the ask, sets out the options management rejected, and says what would change the recommendation. Reading it, the director remembers a similar proposal two years ago with a similar forecast attached, and wants to know how that one turned out before voting on this one.

No author could have anticipated that question. A board paper is a document, and a document is fixed at the moment it is written. Board Intelligence and the Chartered Governance Institute UK & Ireland found the average pack now runs to 220 pages in research published in October 2025. Even at that length, no pack answers every question a director brings to it, pulls in every external source, or recalls every past decision. In Board Intelligence's Summer 2026 Board Value Index, 84% of directors said poor information quality had caused at least one delayed, rushed or poor decision in the past six months.

Answering that question by hand means an evening with old packs.

IQ Experts sits inside the Board Intelligence portal, on the pack a director is already reading. The director asks the question and has an answer in moments, cited to the paper it came from and limited to what their own permissions allow them to see. It brings 130+ expert perspectives and 80+ boardroom capabilities to any paper, from cybersecurity risk and capital allocation to CEO succession, drawn from 20 years of board advisory work. It searches reliable external sources too, each one cited. Board Intelligence configures it for the organisation, and client data never trains the AI models.

The director forms their own view and makes the call. A director who finds the figures do not hold can ask the board to defer, and a board acts legitimately when it defers until the information is good enough.