What to Tell Your LPs About AI

1. LPs Are Asking Now (DDQs and ODD)

A year ago, AI was not on the diligence questionnaire. Now it is showing up, in the DDQ a prospective LP sends before they commit, and in the operational due diligence review the institutional ones run on managers they already back.

The questions are still rough. Some are a single line bolted onto a cybersecurity section. Some are a full page. But the direction is clear: LPs have decided that how a manager uses AI is part of how a manager is run, and how a manager is run is exactly what operational due diligence exists to check.

So the practical position is simple. You will be asked. The only choice is whether you have an answer ready that builds confidence, or whether you improvise one under the gaze of someone whose job is to find the gap. This is practical guidance, not legal advice, and your own counsel and compliance team should review anything you put in writing. But the shape of a good answer is not a mystery, and it is worth having before the question lands.

2. What They Are Really Asking

It is easy to misread the question. An LP asking about AI is not asking whether you are innovative. They are not hoping for a story about how the firm is ahead of the curve. If anything, a manager who sounds too excited about AI raises the operational due diligence team's eyebrows, not lowers them.

What they are really asking is whether you are careful. An LP has handed you confidential information and capital, and AI is a new place both could leak or be mishandled. Behind every AI question on a DDQ is one of four worries: is our data safe, is anyone watching how this is used, does a human still own the decisions, and are you telling us the truth about it.

Read the question that way and the answer gets easier. You are not pitching a capability. You are demonstrating control. The LP wants to hear that you have thought about the risks they are thinking about, and that you have put boundaries in place before something went wrong, not after.

3. The Questions to Be Ready For

The wording varies, but the questions cluster into four areas. Be ready for all four, because a strong answer in one and a blank stare in another reads worse than competent answers across the board.

The first three are where most of the diligence weight sits, because they are about risk. The fourth is where firms get themselves in trouble, because the temptation is to oversell, and overselling AI is now a question the regulator can ask too.

4. Answers That Build Confidence

A confidence-building answer has a particular shape. It is specific, it names a boundary, and it does not reach for more than is true.

On data. The strong answer names the approved tools and the plan tier, states that confidential deal and LP material goes only through those, and says plainly whether the data is used to train outside models under that contract. "We use enterprise tools that do not train public models on our data, and consumer tools are not permitted for confidential material" is a real answer. "We take data security seriously" is not.

On supervision. The strong answer names the owner and points to a written policy. An LP wants to hear that one person is accountable, that there is a document, and that staff have been told what is and is not allowed. A firm that cannot name who owns AI has told the LP that nobody does.

The pattern holds across all four areas: a specific, bounded claim beats an enthusiastic vague one every time. An operational due diligence reviewer has read a hundred vague answers. The specific one is the one that lowers their pen. The fuller mechanics of the policy and supervision that sit behind these answers are the whole subject of the firm's AI governance work.

5. The AI-Washing Trap

Here is the part that has changed the stakes. Overclaiming about AI is no longer just a credibility risk with an LP. It is now squarely in scope for the regulator.

The SEC has made clear that the marketing rule covers what advisers say about their use of AI, the same way it covers any other claim about the business. Saying you use AI in a way you do not, or implying a level of sophistication that is not real, is the same category of problem as any other misleading statement to investors. The agency has used the term AI-washing for exactly this, and it has brought cases. The point is not the specific rule citation, which your compliance counsel will know precisely. The point is the principle: if you say it to an LP, it has to be true.

This cuts against the marketing instinct, which is to make the AI story sound as impressive as possible. The DDQ is the wrong place for that instinct. A claim that AI "drives our investment decisions" sounds good until an examiner or a careful LP asks to see where, and you are describing a tool a few analysts use to summarize documents. Underclaiming costs you nothing. Overclaiming can cost you a great deal.

6. What You Can Honestly Claim

The honest claim is usually stronger than the inflated one anyway, because it is credible. Describe what is actually true and you sound like a firm that knows what it is doing.

Most firms can honestly say something like this. We use AI to compress the assembly work around our process: summarizing documents, drafting first versions, organizing diligence, so our people spend more time on judgment and less on collation. We do it through approved enterprise tools, under a written policy, with a named owner, and a human reviews any output before it informs a decision. That is accurate, it is specific, and it describes control rather than magic.

Notice what that claim does not say. It does not say AI makes the investment decisions. It does not say the firm has a proprietary model. It does not promise a return uplift you cannot evidence. The honest claim is bounded by what you can show, and an LP trusts a manager who stays inside that boundary far more than one who steps outside it. Where AI genuinely moves the economics of the business, the durable version of that claim, is a separate and more careful conversation, covered in how AI is changing the economics of private equity.

7. The Governance That Makes Your Answers True

An answer is only as good as the thing behind it. The reason firms struggle with the AI DDQ is not that the questions are hard. It is that they have not done the work the questions assume.

The work is a short, real governance foundation. A written policy that says which tools are approved and for what. A named owner who is accountable. A clear line on confidential data and where it may go. And a record that staff have been told the rules. None of that is exotic, and none of it requires a large program. It requires a decision and a document, which is exactly what an examiner and an operational due diligence reviewer are both looking for.

The same foundation that answers the LP also answers the regulator, because the questions overlap heavily. The supervision and policy work that satisfies an examiner is the same work that satisfies an LP, which is why it is worth building once, properly. The exam-facing version of this is laid out in the firm's SEC exam readiness guide, and the data-handling specifics are in the AI security and data governance guide. For the regulator-facing package itself, that is what SEC AI governance produces.

8. Putting It in the DDQ and the Annual Letter

Once the governance exists, get ahead of the question instead of waiting for it. There are two natural places.

The DDQ. Draft a standing AI section before anyone asks, so the answer is consistent and reviewed by counsel rather than improvised by whoever happens to field the call. A prepared half-page that covers the four areas, in plain bounded language, signals a firm in control. It also stops the failure mode where two people at the firm give an LP two different answers.

The annual letter. A short, measured paragraph in the LP letter, framed as operational discipline rather than a technology boast, does quiet work. It tells your existing LPs you are thinking about this, on your terms, before they think to ask. Keep it modest and accurate, the same bounded claim from earlier, and it builds trust precisely because it is not selling.

Whoever owns this internally should hold both, so the DDQ answer and the letter say the same true thing. That single owner is the same role that runs governance day to day, the subject of who should own AI at your firm.

9. Turning the Question Into an Advantage

Most managers will fumble this for a while, which is the opportunity. When an LP asks about AI and gets a clear, specific, honest answer, the manager stands out, not because the answer is clever, but because most answers are not.

Operational due diligence is, at its core, a search for managers who are run well. A confident, bounded answer on AI is a small but real signal of exactly that. It says: we adopt new tools deliberately, we govern them, and we do not oversell. That is the same temperament an LP wants to see applied to everything else the firm does.

The advantage compounds with substance. A firm that can honestly say AI has changed how it works, with a specific example and a human still in charge, is more credible than both the firm that has done nothing and the firm that has oversold. The way to earn that honest claim is to actually do the adoption well, which is the entire point of an AI strategy and roadmap built around real work rather than a slide.

10. Where to Start

Write the half-page DDQ answer this week, even in rough form, and notice where you cannot answer cleanly. The blank spots are not a writing problem. They are the governance you have not built yet, and they are exactly what to fix first.

Then put the short foundation in place: the policy, the owner, the data line, the record. Have counsel review the language before any of it goes to an LP. Build it once and it serves the DDQ, the annual letter, and the next examiner who walks in, all from the same true source.

If you want the policy, the supervision approach, and the LP-ready answers built together, that is exactly what AI governance delivers, and where the regulator is the sharper concern, the SEC AI governance package is built for it. We can also run it with you over time as an AI Operating Partner, so the answers stay true as your use of AI grows.