How can a company secretary challenge an AI-generated recommendation before presenting it?

A company secretary challenges an AI-generated recommendation by interrogating its inputs, assumptions, and sources before it reaches the board. The test is simple: can the recommendation be explained, traced, and defended? If any answer is unclear, the paper is not ready for directors.
The role carries advisory weight in the boardroom, so the secretary answers for what is presented, whatever tool produced it. A disciplined review protects the board’s decision quality and its compliance standing. The sections below cover the questions to ask, the checks to run, and how to present the result.

What questions should a company secretary ask about an AI output first?

Start with five questions: What was the AI asked to do? What data did it use? What assumptions shaped the answer? What would change the conclusion? And what is the cost of being wrong? These questions expose the limits of any recommendation before it gains credibility through polished wording.

AI output reads fluently, and fluency creates a false sense of authority. A confident paragraph is not evidence of a sound conclusion. The secretary’s first task is to separate the quality of the prose from the quality of the reasoning.

Apply these questions in order:

  1. Scope: Did the prompt define the problem correctly? A poorly framed question produces a precise answer to the wrong problem.
  2. Inputs: Which documents, datasets, or time periods informed the output, and what was left out?
  3. Assumptions: What conditions must hold for the recommendation to be valid?
  4. Sensitivity: Which single change in facts would reverse the advice?
  5. Consequence: What are the legal, financial, and reputational implications if the recommendation proves wrong?

Record the answers. A short written trail shows the board that the paper was tested, not merely received.

How can you verify the sources and data behind it?

Verify AI output by tracing every material claim to a primary source, such as a statute, regulatory guidance, an audited figure, or an internal record. If a claim cannot be traced, treat it as unverified and remove it or flag it. Never accept a citation without opening the underlying document.

AI systems can produce plausible references that do not exist, or cite real sources inaccurately. For a role built on statutory and regulatory accuracy, this risk demands a firm rule: no unverified claim enters a board paper.

Check legal and regulatory references

Confirm every cited provision against the current text of the law or code. Governance requirements change, and an AI tool may rely on outdated or jurisdictionally incorrect material. Check the date, the jurisdiction, and the exact wording.

Check figures and internal data

Reconcile every number with the finance function or the original report. Confirm the reporting period, the definitions used, and whether the data is complete. A correct figure applied to the wrong period still misleads the board.

How do you spot bias or blind spots in AI advice?

Spot bias by testing whether the recommendation reflects the organisation’s actual context or a generic pattern. Look for missing stakeholders, unexamined alternatives, and conclusions that mirror the way the question was framed. If the output only confirms what the prompt implied, it has limited value.

AI tools draw on historical patterns. They tend to favour conventional answers and may underweight the specifics of your sector, culture, or ownership structure. They also lack awareness of boardroom dynamics, relationships, and unwritten constraints that shape a sound decision.

Use these checks:

  • Request the opposing case. Ask the tool to argue against its own recommendation, then assess whether the counterargument holds.
  • Test alternatives. Confirm that at least two credible options were considered, not one option dressed as a choice.
  • Map the stakeholders. Identify who is affected by the recommendation and whether the analysis accounts for them.
  • Look for framing effects. Check whether the way the prompt was worded steered the result.

Boards benefit most from honest, unbiased analysis. The same standard applies to a machine’s advice as to any adviser’s.

Who should review an AI recommendation before it reaches the board?

Route the recommendation to the subject matter owner, the relevant executive, and where the stakes warrant it, legal counsel or an independent adviser. The company secretary coordinates this review but should not be the only checkpoint. Accountability for the content stays with a named human.

The matter determines the reviewers. A financial recommendation needs the CFO’s scrutiny. A regulatory interpretation needs legal confirmation. A recommendation touching people, remuneration, or succession needs the right executive owner and, often, the Chair’s early view.

A workable review path looks like this:

  1. The function owner validates facts and feasibility.
  2. Legal or compliance confirms regulatory accuracy.
  3. The company secretary tests structure, sourcing, and fit with board procedure.
  4. The Chair is briefed on any material uncertainty before the paper circulates.

The Chair relies on the company secretary to flag weaknesses early. Surfacing a doubt before the meeting is far better than exposing it in front of the full board.

How should an AI-assisted recommendation be presented to directors?

Present it transparently. State that AI informed the analysis, describe what it did, show what humans verified, and set out the remaining uncertainties. Directors should see the recommendation as one input to their judgement, not a conclusion to ratify.

Transparency builds trust. Hiding the role of AI invites doubt if it later emerges, while overstating its reliability invites over-dependence. A measured disclosure respects the board’s oversight duty and supports sound deliberation.

Include these elements in the paper:

  • A brief note on how AI was used and which tool or method was applied.
  • The verification steps completed and by whom.
  • Key assumptions and the conditions under which the advice would change.
  • Alternatives considered and the reasons for the preferred option.
  • Residual risks, stated plainly.

Invite challenge in the meeting itself. Directors who question the analysis strengthen the decision. This mirrors the broader shift in board work, where attention moves from compliance alone toward engaged, forward-looking oversight. A board that tests AI advice rigorously is practising the same discipline it applies to every other significant decision, which is a central theme in any serious board effectiveness discussion.

How The Board Practice helps with challenging AI-generated recommendations

The Board Practice brings more than 19 years of refined methodology and board-level experience across industries and continents to the question of how boards should govern their use of AI. The firm works in close partnership with the Chair to tailor its approach to each board’s context, offering frank, unbiased counsel. Its AI-powered board platform, launching on 1 August 2026, pairs that expertise with scalable analysis that supports continuous board performance tracking.

  • Bespoke board effectiveness evaluations that examine how directors challenge information, including AI-generated input
  • Honest feedback on decision-making dynamics, strengths, and areas for development
  • Forward-looking development plans, typically spanning two to three years, monitored with the Chair
  • An AI-powered platform that generates evaluations and delivers actionable recommendations, designed to complement human judgement rather than replace it

To strengthen how your board and company secretary interrogate AI-assisted advice, contact The Board Practice to discuss an approach designed around your board.

Related Articles