Beyond the Prompts: What Questions Restaurant Operators Ask AI

Operators are having a fundamentally different relationship with data and what most restaurant technology previously offered because a third of them are treating AI as analyst instead of just a reporting tool, according to PAR’s  e-book, “The Questions Behind the Counter.”

“When you follow up, drill down, and redirect, you're not looking for a number, you're building toward a decision,” explained  Diane Le, VP of Marketing- Restaurants, PAR Technology. “That one-third figure tells me operators understand intuitively that the first answer is rarely the whole answer. They want to understand the why behind what they're seeing, and they're willing to have a back-and-forth to get there. That's actually the behavior that makes AI most valuable, and it's emerging organically. Nobody trained them to do that.”

Among the top prompts are: 

• What were my Uber Eats sales in [month]?

• How does sales over the last 30 days show a trend of going up or going down?

• How do we improve the speed of service at this location?

PAR is labeling this a turning point, Le noted, because the question operators are asking has fundamentally shifted from "what is the number" to "why is the number what it is, and what should I do about it?" 

A Different Perspective

The team at PAR believed there were so many AI adoption reports that measured clicks on preset categories that tells what features are built well, not what operators actually need. 

“We wanted to know what a GM actually asks AI versus what they’re being told what to ask,” said Le. “The raw, self-composed questions are so much more revealing than any survey response. When someone types ‘where am I losing the most money?’ at 11 p.m,, that signals a business problem with real stakes behind it. We felt like that unfiltered question deserved to be the foundation of the whole report.”

Data Rich, But Context Poor

The questions tell you that operators are data-rich and context-poor and they can see the gap between their best stores and their worst stores, the report found. 

“They can see the discount total. They can see the slow drive-thru. What they can't see is why those things look the way they do. And for years, the answer was: pull another report, add another dashboard, hire someone to synthesize it. What's changed is that AI can now sit on top of that data and give you the context in real time.”

The biggest takeaway isn't really about prompt optimization. It's about permission, Le said. 

“Operators have spent years looking at dashboards that told them what happened. What this report shows is that when you give them an AI connected to their actual data, they immediately start asking why it happened and what they should do about it. That instinct is exactly right.”

Le’s advice would be: don't think about prompts as a formula to memorize. Think about the question you actually can't answer with the report sitting in front of you right now, and ask that. 

“The operators in this dataset who got the most out of AI were the ones who treated it like a conversation, not a search bar. They followed up, they pushed back, they drilled down. That's the behavior to replicate.:

A Number of Surprises

A few things in the report genuinely caught Le off guard include the striking loss prevention intensity. Not just that it was the biggest cluster, but how specific the questions were. 

“Operators weren't asking broadly about discounts. They were asking for cashiers with back-to-back zero-dollar transactions, credit card refunds run on the same card in the last 30 days. These are people who already knew the patterns of theft and manipulation. They weren't learning about it from AI. They were using AI to finally get the evidence fast enough to act on it.”

Loss prevention is a hot topic because margin pressure is relentless right now, and discounts are one of the easiest places for money to leak undetected, the report found. The problem has always been that investigating it manually is incredibly time-consuming. You'd need someone to pull transaction logs, cross-reference employee IDs, look for patterns across shifts. It's the kind of audit that only happens when something has already gone seriously wrong. AI changes that calculus completely. Now you can ask "show me cashiers with back-to-back zero-dollar transactions" and get an answer in seconds.

“The operators in our dataset clearly knew exactly what patterns to look for,” Lee said.
“They just finally had a tool fast enough to catch them in time to act. That's why loss prevention was the first instinct. Not because it's the most exciting use of AI, but because it's the most immediately valuable one for a business running on thin margins.”

The other one that stuck with Le was the jailbreak attempts. Two operators literally tried to get the system to reveal its own instructions.

“That sounds alarming but I actually read it as a really positive sign. You only stress-test a tool you're starting to depend on. That's not skepticism. That's the beginning of trust.”

Future Prompts

Operators are asking about their costs, their employees, their problem locations, but very few are asking what's happening with their guests between visits, the report revealed. 

“What's fascinating to me, coming from loyalty and marketing, is how much of this dataset is inward-facing,” Le said. “Operators are asking about their costs, their employees, their problem locations. Very few are asking what's happening with their guests between visits. Redemption patterns, lapsed members, which dayparts are losing loyalty engagement.”

The questions that connect operational performance to guest behavior are still largely missing, according to the data. Things like: which stores have the worst loyalty redemption rates and also the slowest drive-thru times? Is there a correlation between labor instability and guest retention at a location level? 

“Those cross-signal questions are where I think the next wave of AI use is going to show up, and the operators who get there first are going to have a real advantage.”

Trust is being earned in real time, and this dataset captures that process honestly, Le noted. 

“You can see it in the range of behavior. Some operators are running sophisticated forensic audits and following up with layered questions. Others are asking ‘is there an app for this?’ and figuring out where to click. Both of those are real, and both matter. The stress-testing behavior, probing what data the AI can actually see, checking how it calculated a number, that's due diligence. It's what you do with any tool before you start making real decisions with it.”

What gives Le confidence is that operators are clearly moving through that calibration phase quickly. 

“The sophistication of the questions in this dataset, from a group who had only had access for a matter of months, suggests the learning curve is steep but short. Once operators see that AI gives them answers they can actually act on, the relationship changes fast.”