The Wharton Blueprint for Effective AI Chatbots

Not all AI chatbots are created equal. Some drive satisfied customers, higher sales, and create positive brand experiences. Others leave customers frustrated, angry, and unlikely to return. How can you make your chatbot a customer winner? Wharton Human-AI Research and Science Says have teamed up to create a blueprint for effective chatbots. Based on the latest scientific research, the blueprint offers practical solutions for increasing chatbot usage, improving consumer trust, and deciding when and how to use chatbots that are human-like or machine-like.

Download the Blueprint

Some tasks are perfectly suited to AI chatbots: high-volume, repetitive tasks in lower-risk industries like travel, retail, logistics, and hospitality. But in order to increase AI usage, five psychological barriers must first be addressed.

Perceived opacity

Solution: Explain AI decisions clearly, including why it chose some options or excluded others. This will build trust.

Emotionlessness

Solution: In personal assistant roles, add human-like touches (e.g., give the AI a name) to make AI feel more relatable.

Rigidity

Solution: Highlight AI adaptability (e.g., “Learns your preferences over time”) to counter perception that the tool is rigid or outdated.

Autonomy

Solution: Allow users some control (e.g., give features to name or select AI avatar) to reduce their concerns about AI autonomy.

Non-Human Nature

Solution: Do not frame AI as “having human-like consciousness.” Focus instead on its practical tool-like abilities and benefits.

Customers need to know how chatbots benefit them rather than solely benefiting the firm.

▼
0
%
Chatbots were rated 8.5% lower than humans when providing the same exact service, because they are viewed as a cost-cutting measure.
▲
0
%
Highlighting superior chatbot performance, such as faster or error-free service, improved service evaluations by 37.2%.
Download the Blueprint

“Don’t underestimate the power of personalization, even with AI. Favorable decisions delivered by human-like chatbots are seen as more valid and appreciated because consumers feel their individual qualities are being considered.”

—Stefano Puntoni, Wharton Marketing Professor

Use human-like chatbots and machine-like chatbots in different situations.

▲
0
%
People rated a company 8.1% higher when favorable decisions (e.g., loan approval) were made by a human-like chatbot
▼
0
%
In high-pressure, time-sensitive contexts (e.g., travel rebooking), human-like chatbots reduced satisfaction by 15.7% compared to machine-like chatbots.
Download the Blueprint

“This Blueprint isn’t abstract theory. It’s rigorously tested scientific research that very few companies have discovered or applied.”

—Thomas McKinlay, Founder of Science Says

Select Takeaways

Increasing Usage
  • Use “learning” labels to make AI seem dynamic and able to gain experience over time (e.g., “This AI adapts and improves with use”).
  • Highlight chatbots’ benefits, such as faster response times or error-free service.
  • Recalibrate chatbots to recommend both popular and niche products to ensure suggestions feel personalized.
  • Entice less AI-savvy users to use the technology by framing AI tools as magical, transformative, and intuitive.
Improving Trust
  • Highlight evidence of AI’s accuracy through clear, measurable outcomes (e.g., “94% of users loved the products recommended by our AI”).
  • Design algorithms to deliver predictions quickly since speed increases the perception of AI accuracy.
  • Communicate human involvement in AI development rather than presenting AI chatbots as entirely autonomous systems.
  • If using humor in chatbots, be cautious of sensitive topics so that the chatbot does not inadvertently reinforce harmful stereotypes.
When to Use Human-like vs. Machine-like
  • Use human-like chatbots to deliver favorable decisions.
  • Use machine-like chatbots when handling embarrassing or sensitive information.
  • Prioritize human-like AI for emotionally impactful interactions, such as addressing complaints, resolving emotionally charged issues, or discussing sadness.
  • Use machine-like chatbots for situations where customers are likely to be angry, or for tasks prone to errors.
Making Chatbots More Human or Machine Like
  • Design your chatbot to use friendly language and interjections to feel more human.
  • Prioritize competence rather than personality alone by ensuring accurate and timely responses.
  • Design chatbots with human-like features such as empathy and gratitude when there is risk of fraud and to encourage moral behavior.
  • Program machine-like chatbots to use flattery with product recommendations. Machine-like AI is viewed as less manipulative than human-like AI.
Download the Blueprint