Your AI goal assistant should challenge you, not flatter you

Key Takeaway: Most AI goal-setting tools are built to agree with you. They generate OKRs on request or polish whatever you type. A randomized trial at BT Group, reported in Harvard Business Review, suggests that's the wrong design goal. An AI tool built to challenge employees as they set their annual goals produced 3.5 times as many excellent goals as default AI or no AI, with no drop in enjoyment or psychological safety. It's one trial at one company, so treat it as a strong signal rather than settled science. If you're rolling out AI for goal setting, set it up to challenge ambition, specificity, and outcomes, then measure whether goal quality actually improves. Or use Perdoo's built-in AI Agent (Spark).

Most AI goal-setting features are built to be helpful in the most literal sense. You describe what you want and they write an Objective, or you paste in a draft and they polish it. In my experience, that rarely makes the goals any better. New research from BT Group, the UK's largest telecommunications group, now puts a number on that gap.

What BT Group tested

The study, published in Harvard Business Review on November 4, 2026, comes from researchers at BT Group and the behavioral-science practice MoreThanNow. In a randomized controlled trial, 2,400 participants set their goals for the year ahead in one of three groups:

• A control group with standard guidance and no AI.
• A default AI group, using a tool that mirrored typical enterprise AI and tended toward sycophancy.
• A challenger AI group, using a tool designed to be challenging, ambitious, and stretching.

The challenger tool kept nudging people to be more specific and to push further on scale or timeline. Three and a half times as many of its goals were rated excellent compared with the other two groups, and enjoyment and psychological safety were the same across all three.

It's one trial at one company, published in HBR rather than a peer-reviewed journal, and it tested individual performance goals, not OKRs. Still, it matches what I've seen with many customers: Hard questions are what turn a weak first draft into a good goal.

Why agreeable AI makes goals worse

The authors' headline concern is sycophancy: AI that affirms users instead of challenging their thinking. In goal setting, that's expensive. An agreeable assistant turns "Improve the customer experience" into a more polished version of the same vague goal. The authors also warn about less diverse ideas and "cognitive surrender," where people rely on AI output instead of their own judgment and skip the thinking that creates ownership.

What good pushback looks like

The study's challenger AI pushed on ambition and specificity. For OKRs, I'd add two more: outcomes and focus.

It raises the ambition. As our guide on how to write great OKRs puts it: A Key Result that doesn't make you a little uncomfortable isn't ambitious enough. And if you consistently hit 100%, the Ultimate OKR guide notes, your OKRs aren't challenging enough.

It demands outcomes, not activities. "Launch the new onboarding flow" is something you do, not a result. As we explain in outcomes vs. outputs, Key Results measure outcomes and the work belongs in Initiatives. This is where OKRs go beyond the study: BT's challenger AI asked people which actions they'd take, a good question for an Initiative but not for a Key Result.

It questions the number of priorities. The right number of OKRs depends on how many people are available to work on them, and most teams overestimate that. A good coach asks who will move each Key Result and what you'll stop doing to make room.

Take a typical draft: The Objective "Improve onboarding," with the Key Results "Launch the new onboarding flow" and "Run four customer webinars." A challenging coach would ask what changes for customers if this works, and where that number is today. The rewrite might look like this:

• Objective: Make new customers successful in their first 30 days.
• Key Result: Increase 30-day activation from 42% to 60%.
• Key Result: Reduce time to first value from 9 days to 4.
• Initiatives: Launch the new onboarding flow and run four customer webinars.

The numbers are illustrative. The activities didn't disappear. They moved to the Initiatives, where they belong.

Challenge doesn't have to cost psychological safety

In the BT trial, pushback didn't feel like criticism: People who were challenged enjoyed the process just as much. I suspect a tool's questions carry less social weight than a manager's. But the authors don't frame AI as a replacement for people, and they recommend investing in managers' ability to give feedback and have tough conversations. AI can improve a first draft, but commitment still forms in the conversation between a manager and their team. That's also where stretch goals go wrong if people think missing an ambitious target is a failure.

How to choose and roll out AI for goal setting

Here's what I'd focus on. The first two come from the authors' recommendations.

• Measure outcomes, not adoption. The authors warn that metrics like active users, token consumption, and training completion are easy to game. Compare goals before and after against a simple rubric, and track whether they move the metrics that matter.
• Test before you roll out. The authors recommend testing design choices, ideally through randomized trials, before a wide rollout. Start with a challenging setup for a few teams for one cycle and compare their goals with everyone else's.
• Check what the tool challenges. Does it ask about ambition, measurability, outcomes, priorities, and how a goal connects to your Strategic Pillars? If it only rewrites, it's a writing tool, not a coach.
• Have people think before they prompt. Ask teams to write their own first draft, then use AI to critique it.

Good goals are the starting point of strategy execution, so this is worth getting right. An AI that tells everyone their goals look great makes planning faster and execution weaker.

Try this before your next planning cycle

Take your three most important draft OKRs and give them to your AI tool with this prompt: "Act as a demanding OKR coach. For each Objective and Key Result, tell me what's vague, what measures activity instead of an outcome, and where the target is too safe. Don't rewrite them for me." If all you get back is praise, change the prompt or change the tool.

At Perdoo, we've built our AI around the same idea. Spark, our in-app AI Agent, doesn't just draft goals: It reviews your OKRs and KPIs and helps you refine your Strategic Pillars. And because it works inside your Strategy Map, it sees how every OKR, KPI, and Initiative connects to your Strategic Pillars and Ultimate Goal. That way, ambition and alignment get challenged in the same place. Start for free or Request a demo to see how it works with your own goals.