Share with your CHRO
BetterUp Labs, in collaboration with Stanford’s Social Media Lab, has identified leadership communication framing as the single strongest predictor of low-quality AI output in organizations, outranking psychological safety, employee confidence, and individual personality traits. When AI use is mandated rather than invited, employees are significantly more likely to produce “workslop,” a term the researchers use for polished-looking but substanceless AI-generated output that creates downstream work. Pfizer and Aon are cited as counterexamples, both distributing AI accountability across the organization rather than centralizing it, and high leadership trust correlates with a 46% higher probability of landing on an augmentation path rather than pure automation.
What this means for your business
The uncomfortable finding here is that your AI rollout design may already be producing the behavior you’re trying to prevent. If the first thing employees heard about AI came packaged as a proficiency deadline or an adoption rate target, the research suggests they are more likely to be gaming compliance than building capability. Whether that describes your organization or not is less a question of culture and more a question of what the first formal communication about AI actually said.
The 46% trust multiplier on augmentation outcomes is the number worth pressure-testing internally. High leadership trust isn’t a soft cultural asset in this framing; it’s a structural condition that shapes whether employees treat AI as a tool they own or a threat they’re managing around. The Salesforce data point, that employees who feel included are 58% more likely to be confident with AI, suggests that inclusion programs and AI readiness programs are solving for overlapping conditions. Organizations that siloed those two efforts are probably paying for it twice.
The research comes from BetterUp, a coaching platform whose commercial interest sits squarely in organizations investing in human development alongside AI adoption, so the finding that human conditions outrank individual traits as AI performance drivers is the conclusion you’d expect from that angle. That doesn’t make the finding wrong, but it does mean the burden of proof for the mandate-versus-encouragement distinction should sit higher than a single preprint. The leading indicator to watch is whether your quality review processes are catching an increase in outputs that pass initial scrutiny but require significant rework downstream. If rework rates are rising alongside AI adoption rates, the framing problem is already measurable.
Based on reporting from How Leadership Communication Shapes AI Performance, originally published 2026-07-21 16:47:00.
