Should you follow an AI recommendation that conflicts with your own judgment?

31% of Executives Follow AI Over Their Own Judgment. Should You?

Only when the AI is acting as a tested formula, not a fluent narrator. Board's 2026 Planning Intelligence Report (September 16, 2026) found 31% of executives follow an AI recommendation that conflicts with their judgment, 48% among CFOs. Kahneman's rule in Thinking, Fast and Slow: formulas beat intuition only in regular environments with fast feedback, and a chatbot reasoning over data 75% of those executives admit is 30-plus days old fails it.

A founder I coach told me last month that he had overruled his own read of a hiring plan because ChatGPT disagreed with him, and that he felt oddly relieved doing it. The relief is the part I keep thinking about. Deferring to the model did not feel like a decision. It felt like putting one down.

He has company. Board's 2026 Planning Intelligence Report, released September 16, 2026 and based on a survey of 300 CFOs, CIOs and COOs at US companies with at least $100 million in revenue, found that 61% of executives now cite large language models such as ChatGPT, Claude, Gemini and Copilot among the external sources that most influence their strategic decisions, ahead of industry peers at 42%, technology vendors at 37%, external consultants at 28% and Gartner at 14%. More striking: 31% say they follow an AI recommendation even when it conflicts with their own judgment, and among CFOs that rises to 48%, against 33% of CIOs and 11% of COOs. Only 39% report formal governance and escalation processes for AI-driven decisions, and 36% expect AI to lead most enterprise planning decisions by the end of 2028. Board's own CFO, Gordon Pothier, told CFO Dive the same day that "you don't just go along with what AI is saying."

Here is where I have to be more honest than the headline wants me to be. The book I reach for on this question, Daniel Kahneman's Thinking, Fast and Slow, is the strongest argument in print for overruling expert judgment. Kahneman reviews some 200 studies in which simple formulas beat trained professionals, from clinical diagnosis to wine prices, and writes that "whenever we can replace human judgment by a formula, we should at least consider it." Read that way, the 48% of CFOs deferring to the machine are the enlightened ones. But Kahneman attaches two conditions that the survey never asks about. Formulas and intuitions are only trustworthy in what he calls high-validity environments: regular, predictable domains where the decision-maker gets fast, unambiguous feedback. And a formula is a fixed, tested rule applied the same way every time, with a track record you can inspect. An LLM recommendation is neither. It is a coherent story assembled from whatever you put in the prompt, which is Kahneman's WYSIATI, what you see is all there is, running at machine speed. It changes when you rephrase the question. It has no measured hit rate on your decisions. And it reasons over your inputs, which the same Board report says are stale: three-quarters of respondents admit roughly half or more of their planning decisions rely on data more than 30 days old, and 83% say their board has acted on a forecast it already knew was outdated.

So the question is not whether to defer to AI. It is whether this particular recommendation has earned formula status. I now run every AI-versus-me conflict through five questions before I decide who wins, and I have found the table more useful than any prompt trick:

Kahneman's condition	Ask yourself	If yes
Regular environment	Is this a repeating decision with stable rules (pricing, credit, inventory, scheduling)?	Lean toward the AI
Fast feedback	Will you know within weeks whether it was right?	Lean toward the AI
Tested rule	Has this exact method been scored on your past decisions, not just on a benchmark?	Lean toward the AI
Current inputs	Is the data it reasoned over newer than 30 days and complete?	Lean toward the AI
Facts, not values	Is the disagreement about what is true, rather than what you are willing to risk?	Lean toward the AI
Rule	Four or five yes: follow the AI and log the outcome. Two or three: run a premortem on both options before choosing. Zero or one: your judgment, and say so in writing.

What I actually do when a model disagrees with me is treat the disagreement itself as the valuable output and refuse to resolve it quickly. I ask the model for the base rate it is implicitly using and the three facts that would change its mind, then I run Kahneman's premortem on my own position: assume I followed my gut and it failed a year from now, and write the story of why. About a third of the time that exercise exposes that my judgment was System 1 pattern-matching on the last company I ran, and I switch. The rest of the time it exposes that the model's confident paragraph rested on an assumption I never gave it, and I hold. Either way I write down which one I chose and check back, because that log is the only way an AI recommendation ever earns the formula status Kahneman would grant it.

The limits of the evidence deserve stating. Board sells an AI planning platform, so the report has an interest in AI looking both inevitable and in need of governance. It is a self-reported survey fielded in May and June, and "follow an AI recommendation" was never defined, so some of that 31% may mean "I changed my mind after reading a better argument," which is exactly what should happen. The one figure I trust most is the smallest: 11% of COOs, the executives closest to the physical business and the fastest feedback loops, are the least willing to overrule themselves. They are living in Kahneman's high-validity environment, and they know what the model does not.

Sources: Board, "83% of Executives Say Their Boards Acted on Forecasts Known to Be Outdated," PR Newswire, September 16, 2026 (prnewswire.com); Board, The 2026 Planning Intelligence Report landing page (board.com/resources/the-2026-planning-intelligence-report); CFO Dive, "Business execs seize on AI as top tool for decision-making," September 16, 2026 (cfodive.com); Daniel Kahneman, Thinking, Fast and Slow (2011), chapters on intuitions vs formulas and the premortem.