What to Ask When Hiring a Chief AI Officer

For a commercial Chief AI Officer mandate, AI fluency is only part of the interview. Ask for production work, the candidate's contribution, and evidence of the business outcome.

IBM's 2026 CEO Study reports 76 percent of surveyed organizations with a Chief AI Officer. The 26 percent baseline comes from a separate 2025 study with a different sample. That difference does not tell us how many seats were created in a year.

I am a Certified Chief AI Officer and I build the systems. For a commercial mandate, I would test production judgment and commercial work alongside technical understanding. The questions below are a first screen, followed by an artifact review and reference calls.

So here are the questions I would ask, what the good answers sound like, and the one question you have to answer yourself before you talk to a single candidate.

First, the question that is not for the candidate

Before any interview, finish this sentence out loud: the person in this seat is accountable for turning our AI into ______.

If the mandate is commercial, name revenue, margin, cost to serve, or another defined business outcome. If the mandate is engineering or governance, specify the corresponding delivery or risk outcome. Innovation, transformation, and enablement still need a measure and an owner. Why Chief AI Officer hires fail examines the risks of leaving that work undefined.

The candidate cannot fix an undecided mandate. Neither can the search firm. Only you can.

The four questions that separate operators from narrators

Look for work the candidate owned: a production system, a decision to stop or change a project, a business result, and a conflict they helped resolve. Speaking well about AI does not supply that evidence.

Use these four questions as an initial screen, then check the artifacts and references.

1. What did you ship, and who used it?

Listen for: a system in production, real users, a date, and what broke. Operators go specific fast because they were there. Red flag: pilots, proofs of concept, and pronouns that drift. If every win is "we" and every failure is "they," you are hiring a narrator.

2. What did you kill, and who was angry about it?

Listen for: a specific project they shut down, why, and the political cost. Red flag: nothing. If they have never stopped an initiative, ask how their authority was bounded and which decisions they did own.

3. What number moved, and how do you know it was you?

Listen for: a business number, not a model number. Revenue, margin, cycle time, cost per case. Then attribution honesty: a good operator will tell you what they cannot cleanly claim. Red flag: accuracy scores and adoption rates offered as business outcomes, or a number so clean it clearly skipped a finance conversation.

4. Who fought you, and how did that end?

Listen for: a named function, a real conflict, and a resolution that involved the CEO or the board. Legal, security, a division head protecting a budget. Red flag: "everyone was aligned." Ask what competing priorities existed and how the candidate resolved them.

The answers help you decide what to investigate next. They do not replace an artifact review or reference call.

The technical bar, stated honestly

People ask whether a Chief AI Officer needs to be technical. My answer: technical enough to be lied to and notice.

They do not need to write production code. They do need to sit in a room while an engineer explains why something will take six months, and know whether that is true. They need to read an evaluation result and spot the benchmark that was chosen to flatter. They need to inspect a demo for omitted constraints, simulated steps, and limits on what it demonstrates.

Put a real artifact in front of them: an architecture, vendor contract, or model evaluation. Ask what they would challenge and what evidence they would need before making the decision. Compare their reasoning with the constraints your team knows.

The governance questions, which are not about policy

Ask how the candidate has made review work in a real workflow. A committee can set policy or resolve conflicts, but someone still needs authority and evidence at the decision point:

Ask these instead:

Who signs off before something risky ships, and what happens if they say no? You are testing whether they have built a real gate or a rubber stamp.
Tell me about a time your own system was wrong in production. How did you find out? The answer reveals whether they built monitoring or got a phone call from a customer.
What data would you refuse to put into a model, and who have you had to say no to? Saying no to a powerful internal stakeholder is the entire job on the worst day.

I run AI inside an export-controlled defense environment where controlled technical data never touches a public model and a human signs off before anything sensitive moves. That discipline is not a slide. It is a set of decisions someone makes under pressure, usually while a revenue team is asking why it is taking so long. Ask for the version of that story from their own career.

The reference calls, done properly

Most reference calls are a formality where you confirm the highlight reel. Waste of a good phone call.

Ask the reference about the thing that died. Which initiative got killed, who made the call, and how did the candidate take being overruled. Ask what the candidate could not get done, and why. Ask whether the candidate's budget went up or down over their tenure, because that is the cleanest available signal of whether leadership believed them.

And ask one more: would you give this person kill authority over a project you personally sponsored. The pause before the answer tells you more than the answer.

What you are actually hiring for

For a commercial mandate, the hire needs to connect capability with the buyer, price, delivery, and cost to serve. What a Chief AI Officer actually owns expands that scope. An engineering or risk mandate needs a different evidence test.

IBM's CEO study also reports only about a quarter of the workforce uses AI regularly, while 86 percent of surveyed CEOs believe their people have the skills. That gap does not establish whether ownership, training, access, or another factor caused low usage. Ask the candidate how they would investigate it before proposing the fix.

Use the four questions to select what to inspect next. Verify the candidate's contribution, the outcome, and the limits of their authority through artifacts and references.

Frequently asked questions

What questions should you ask when hiring a Chief AI Officer?

Ask what they shipped, what they killed, what it earned, and who fought them. Use the answers as a first screen, then inspect artifacts and call references. A strong candidate answers with systems that run in production, a decision they reversed, a number tied to revenue or margin, and a specific organizational fight they had to win. A weak one answers with frameworks, vendors, and pilots that never reached a customer.

Should a Chief AI Officer be technical?

Technical enough to be lied to and notice. They do not need to write production code, but they must be able to challenge an architecture, judge a build-versus-buy call, read an evaluation result, and know when a demo is hiding the hard part. The failure mode is a leader who cannot tell when the answer they just got was nonsense.

What is the biggest red flag in a Chief AI Officer interview?

Fluency without ownership. The candidate speaks beautifully about strategy, models, and governance, but every example is a pilot, a framework, or someone else's build. Ask who owned the number and watch the pronouns. If the wins are we and the losses are they, you are hiring a narrator. If the candidate never stopped a project, ask which decisions they had authority to make.

How do you check references on a Chief AI Officer?

Skip the highlight reel and ask about the project that died. Ask the reference which initiative got killed, who made the call, and how the candidate handled being overruled. Ask what the candidate could not get done and why. The useful reference call is about friction and failure, because those are the conditions the role actually operates in.

What should you ask yourself before interviewing a Chief AI Officer?

Who is accountable for turning our AI into revenue and margin, and what levers will that person hold. If you cannot answer in one sentence, no interview will save you. An undefined job leaves the hire without an agreed outcome. Decide the mandate, the budget, the kill authority, and the scoreboard before the first candidate call.

Do you need a Chief AI Officer in 2026?

A dedicated role makes sense when AI outcomes need senior ownership and the company can define its authority. IBM's 2026 study reports 76 percent prevalence versus 26 percent in a separate 2025 sample; that does not count newly created seats. Define the work before using adoption figures to justify a hire.

Sources

Adoption figures, the workforce-usage gap, and the CEO skills confidence number are from the IBM Institute for Business Value 2026 CEO Study (released May 4, 2026; 2,000 CEOs and equivalent senior leaders across 33 geographies and 21 industries, surveyed February to April 2026, conducted with Oxford Economics). The interview structure, the archetype distinctions, and the governance questions are drawn from my own operating experience, not from the study.

The Chief AI Officer: the complete guide · Why Chief AI Officer hires fail · What a Chief AI Officer actually owns · The first 90 days · The fractional option

About the author

Jeff Brokaw is a sitting CMO and Certified Chief AI Officer who ships AI in production, not slideware. He built server-side visitor identification that names the companies landing on a page, a de-identified version of which runs live at unignorable.jeffbrokaw.com, and answer-engine optimization that produces measurable pipeline. He has been building AI systems commercially since 2016. He built the engine behind $185M in new business for a defense manufacturer, and authored the go-to-market behind a $114M institutional raise that came together in under 30 days.