The diligence questions that didn't exist three years ago, and how sellers are getting marked down without realizing why
Something has changed in the diligence checklists landing in front of founders over the last year, and most sellers do not see it until they are inside a process. The headline numbers in an LOI still look familiar. The terms still look roughly market. But buried in the diligence packet is a set of questions that did not exist three years ago, almost all of them about AI. What is the company's exposure to OpenAI pricing changes. How much of the codebase was generated by Copilot. Whether any product features depend on a foundation model provider's API staying available at current terms. Whether the team has retention agreements for engineers who would otherwise be poached by AI labs.
Sellers are not always prepared for any of it. The buyer's questions are not designed to find a problem. They are designed to price the risk that the buyer is beginning to assume in any digital business acquisition. AI is no longer something buyers ask about because it is fashionable. It is something they underwrite, the same way they underwrite customer concentration or platform dependency. And like those older underwriting frameworks, it now shows up in the price.
What buyers are actually worried about
The AI questions in 2026 diligence fall into three rough categories, and sellers benefit from knowing which one any given question belongs to.
The first category is platform dependency. If the business runs on top of a foundation model API, the buyer wants to understand what happens if that API changes terms, gets restricted, gets more expensive, or disappears. They are not asking because they think it will happen. They are asking because the answer determines how much of the headline EBITDA is actually durable. A business with twenty percent gross margins because it pays a foundation model provider for every API call is being priced differently in 2026 than a business with the same revenue and a self-hosted model.
The second category is moat erosion. The buyer wants to know what the business does that an AI agent could do tomorrow. Customer support businesses, content production businesses, basic SEO businesses, copywriting businesses, and several flavors of consulting and research are all getting asked harder questions about why a buyer should pay a multiple for what looks like work that is going to be commoditized. The question is not always fair, and the answer is sometimes that the moat is real and not visible from the outside. But the seller who has not thought about it goes into the conversation flat-footed.
The third category is upside capture. If the business uses AI internally to reduce costs or improve margins, the buyer wants to know what is repeatable versus what is a one-time benefit. A company that cut its support team in half by deploying a chatbot has improved its margins, but a buyer underwriting the durability of that margin wants to understand whether the chatbot is going to need to be rebuilt in eighteen months, whether competitors have done the same thing and competed the savings away, and whether the cost reduction came from real automation or from understaffing that will eventually catch up to the business.
These three categories of question were rare in 2023 diligence. They are now standard.
Where sellers get marked down without realizing why
The way this shows up in pricing is rarely explicit. A buyer does not usually say "I'm taking half a turn off the multiple because of your foundation model exposure." They quietly adjust their offer, structure more of the consideration as earnout, lengthen the holdback period, or push for tighter reps on technology-related representations. The seller sees the deal as a whole and assumes the terms reflect market conditions. The buyer sees the deal as a stack of adjustments, several of which exist because the AI risk profile was not where the buyer wanted it.
The most common pattern is the substitution of cash for contingent consideration. A seller expecting eighty percent cash at close gets sixty, with the gap structured as a two-year earnout. The buyer is not being aggressive. They are pricing in their own uncertainty about whether the technology stack underneath the business will look the same in twenty-four months as it does today.
The second common pattern is around employee retention. Where two or three engineering hires used to be enough to satisfy a buyer that the team would stay through transition, buyers are now asking for retention agreements with senior technical staff and structuring earnouts to be contingent on key engineers remaining in place. The AI labor market is real and the buyer knows it.
The third pattern is the most subtle. It is the deal that simply does not happen. Buyers who would have moved forward two years ago are passing earlier in the process now, often citing concerns that the seller cannot fully address because the seller is hearing them for the first time during diligence. The deal that dies in diligence usually feels like bad luck. Sometimes it is preparation.
What sellers can actually do
Most of the work is not technical. It is documentary. A seller who can walk a buyer through the answers to the three categories of question in the first call is in a meaningfully better position than one who is hearing them at the LOI stage.
The platform dependency question gets answered by knowing your stack and your exposure. What models you use, what alternatives exist, what it would cost to migrate, what performance would look like. Sellers who have done this exercise present it as a slide. Sellers who have not present it as a series of uncertain answers under pressure.
The moat question gets answered by being honest about what the business does and where the genuine defensibility sits. Buyers are sophisticated enough to recognize when a seller is overclaiming. The seller who says "this is the part of our business that an AI agent will eat in three years, and here is the part that won't" is more credible than the one who claims everything is defensible.
The upside capture question gets answered by being precise about which of your cost improvements came from real structural changes versus discretionary spend cuts an acquirer would want to reinstate. Cutting a content marketing budget by sixty percent because AI lets you produce more content with fewer writers is a structural margin improvement. Cutting it by sixty percent because you were preparing the business for sale is going to get added back, and you will see the impact in the multiple.
The reframe worth carrying
The change in 2026 is not that AI is a topic in M&A. It is that AI is a category of underwriting, with its own questions, its own risk adjustments, and its own effect on price. Sellers who have not thought about how their business gets evaluated through this lens are not necessarily getting worse outcomes than they would have two years ago. But they are leaving negotiating room on the table they would have used if they had known to think about it.
The best preparation is not a strategy memo. It is a conversation, well before any process starts, about which AI questions are going to come up in your specific deal and how you are going to answer them. The buyer is going to ask. Knowing the answers in advance is what changes the price.