Humor makes it easier to forgive AI mistakes

This insight is brought to you by… Hubspot

You have more tools than ever. So why does it feel like critical context is still missing? 

HubSpot's 2026 Essential Apps for Marketers collection brings together 14 apps across creative, ads, events, and more - so the context you need travels with every campaign, decision, and handoff.

Want to sponsor Science Says? Here’s all you need to know.

📝 Context

Topic: AI | Customer Experience
For: B2C
Research date: April 2025
Universities: Hefei University of Technology

We recently launched our second Blueprint in collaboration with Wharton, about what’s slowing AI agent adoption. One key insight that emerged is this: 

”People are not as forgiving of AI errors as they are of human errors. They systematically misjudge AI partners by focusing on where the system fails on ‘easy’ cases. Those failures loom larger psychologically than the harder cases where the AI adds real value.”
- Kartik Hosanagar, Professor and Co-Director, Wharton Human-AI Research at Wharton

The bar is much higher for AI to perform than it is for people. Assuming the AI works well, we’d still be more critical of even the smallest mistakes. 

That means it’s difficult to get users to accept an AI, even if it’s extremely useful 98% of the time, but makes mistakes 2% of the time. If a junior employee did the same we’d love them, not so with AI.

So how can you make people more accepting of a good AI that makes occasional mistakes? 

Science shows us a simple tactic.

📈 Recommendation

Make your AI use humor when it makes a mistake (e.g. “Oops AI bloopers”), make sure the humor is self-deprecating (e.g. “I’m powered by algorithms but I’m definitely on dial-up speed today”) as that’s the most effective kind.

This is especially effective when the mistake is low severity (e.g. recommending a wrong product) but not so much when the problem is bigger (e.g. giving wrong information to an angry customer).

People will be more likely to forgive the mistake.

🎓 Findings

• People are more forgiving of AI making mistakes when it responds with humor, compared to no humor at all. Self-deprecating humor (e.g. “I apologize—turns out I’m a bit more 'artificial' than 'intelligent' today”) works better than positive and friendly humor (e.g. “AI needs coffee too, let me try again”).

• As part of 4 experiments with 1,919 people, researchers found that when AI used humor (vs no humor), people were:

  • Up to 33.9% more forgiving of a wrong recommendation when it used positive humor

  • Up to 47.8% more forgiving of a wrong recommendation when it used self-deprecating humor

  • 15.7% more forgiving in after-sales services when the humor was positive

  • 25.6% more forgiving in after-sales when the humor was self-deprecating

• The effect is:

  • Stronger when the mistake is low severity (e.g. a wrong color recommendation for a jumper)

  • Disappears when the failure is high severity (e.g. not reimbursing a rightful refund request)

🧠 Why it works

• Humor helps us regulate our emotions (e.g. making fun of a difficult situation).

• So in service contexts, it helps us get over our anger towards someone who makes a mistake.

• The same happens when AI messes up. Humor makes us feel less tension.

• This makes us see it more positively and be more forgiving.

🤖 The AI event for marketing strategists: BreakingSilos 2026. Join for free on June 9th.

Top marketing teams aren't winning AI search by optimizing channels.

The brands pulling ahead in AI search are doing something different: they’re building visibility ecosystems where SEO, social, PR, content, and brand reinforce each other. And AI is citing them for it.

BreakingSilos 2026 is an online conference where the practitioners building those ecosystems share exactly how.

This announcement was sponsored. Want your brand here? Click here.

✋ Limitations

• Researchers tested positive and self-deprecating humor. It's unclear how people might respond to different kinds of humor (e.g. self-enhancing, dark humor).

• The study was on online AI chatbots, people might respond differently in real-world offline interactions.

• The study only looked at customer service interactions, the effect might not apply in other contexts (e.g. a coding chatbot)

👀 Real-life example

Sephora helps customers find products through an AI chatbot. It handles complaints by being apologetic.

❌ Issue: The chatbot says sorry and gives in-depth explanations when it makes mistakes, which is not the best approach for this kind of low-severity mistakes.

✅ Solution: Sephora should change how its chatbot responds to mistakes, for example, by:

• Making it make fun of itself (e.g. “Oops my non-brain just glitched, let me retry”)

• Make the chatbot say “thank you” (e.g. “Thank you for bearing with me”), instead of “sorry”.

• Make the chatbot use interjections (e.g. “Oh no!”).

🔍 Study type

Online experiments.

📖 Research

Affiliative or self-deprecating? Exploring the effect of humor types on customer forgiveness in the context of AI agents’ service failure. Journal of Business Research (April 2025)

🏫 Researchers

• Yuguang Xie. School of Management, Hefei University of Technology. 

• Peiyu Zhou. School of Management, Hefei University of Technology.

• Changyong Liang. School of Management, Hefei University of Technology.

• Shuping Zhao. School of Management, Hefei University of Technology.

• Wenxing Lu. School of Management, Hefei University of Technology.

Remember: This is a new scientific discovery. In the future, it will probably be better understood and could even be proven wrong (that’s how science works). It may also not be generalizable to your situation. If it’s a risky change, always test it on a small scale before rolling it out widely.

🎁 Bonus mini-insight

Check your knowledge from previous insights (for paid Platform members only).

• 📈 Join the Science Says Platform to unlock all 300+ insights, real-world case studies, and exclusive playbooks

• 📘 Boost your sales and profits with topic-specific Science-based Playbooks (e.g. Pricing, Ecommerce, SaaS, AI Best Practices)

• 🔬 Get science-based thought leadership content for your brand. We help market leaders in AI, tech, and across B2B enterprises, build authoritative thought leadership using a scientific perspective. Find out more.

🎓 It took me, a PhD analyst, and a marketing pro 10.5 hours to accurately turn this 23-page research paper into this 3min insight.

If you enjoyed it please share it with a friend, or share it on LinkedIn and tag me (Thomas McKinlay), I’d love to engage and amplify! 

If this was forwarded by a friend you can subscribe below for $0 👇