Helping Users Make Confident Solar Choices with Personalized AI Advice from Sunny

During user research, we found that people struggled to access relevant, trustworthy information to make the leap to solar. I guided the vision for Sunny, a ChatGPT–powered solar advisor, leading the design of its interface and prompts. Collaborating with an engineer, we created a tool one user called "a very helpful resource."

Role
Research, strategy, prompt engineering, design

Challenge

Homeowners considering solar felt uneasy about making a major investment without trustworthy, personalized information, which they could access only through pushy, commission-seeking salespeople.

Fragmented Information

Users face overwhelming, fragmented information about solar energy with highly regional variations (cost, laws, incentives).

Uncertainty & Hesitation

Many don’t know what questions to ask, leading to uncertainty and hesitation in starting their research.

Trust Issues

The industry’s reputation for scams reduces trust, and users are reluctant to engage directly with installers too early in the process.

Advice Wanted

Users want actionable, personalized answers to key questions like "Is solar worth it?" and "How much will it cost?"

Solution

A ChatGPT-powered solar advisor packed full of personalized data, available any time to answer any question.

  • Integrated with proprietary data to provide personalized, AI-driven solar advice.
  • Designed to answer user questions in plain language, using data such as energy usage, solar production estimates, and state-specific incentives.
  • Combines the convenience of self-service tools with the personalization of human expertise.
  • Serves as an entry point for users, helping them explore solar on their own time and terms.
Try Out Sunny

Impact

This is a very helpful tool and all in one space to discuss. Much appreciated. Sunny is great and it would be nice to get a full report that can be printed out. Smart move to be a part of [partner name redacted].

⸺Submitted through Sunny's feedback form

Sample questions asked by users

"I plan to install an air source heat pump next year. How do I know what size rooftop solar I need to produce 100% of my total electricity consumption?"

"I have a quote for $90,000 in the Arkansas area 72450. Is that too much?"

"Why won’t installers put solar panels on a mobile home roof?"

So what?

Metrics for new and GenAI products are tricky, but logs revealed Sunny's value: answering specific, detailed questions not found on Google. Refined prompts shortened chats, with users leaving after getting needed estimates rather than from frustration.

More Case Studies

Doubling Lead Gen Conversions Through Strategic A/B Testing and Microiterations Uncovering Opportunities Within the Solar Journey Through User Research

See older work

Details

Using insights from a Conversational AI course, I wrote Sunny’s prompt, refined it through feedback, and worked with an engineer to create a user-friendly assistant.

Process

1User Research

Research revealed key pain points: scattered info, lack of trust, and cost confusion. I then led brainstorming sessions during which Sunny was born.

2Education

I deepened my expertise by taking Reforge’s Conversational AI course, which helped shape Sunny’s strategy and development.

3Mad Science

I wrote the prompt defining Sunny’s personality, purpose, and data sources, while an engineer implemented the chatbot. I also designed a clean, user-friendly interface.

5Monitoring & Iteration

By reviewing conversations and user feedback, I refined the prompt to make Sunny’s responses clearer, faster, and more useful.

Mo’ Prompting, Mo’ Problems

Problem

Solution

ChatGPT can’t handle too much instruction in the prompt

I went through the prompt with a fine-tooth comb, rephrasing it to leave only the essentials and adhering to best practices, such as putting the most important instructions in the beginning or end.

Being too aggressive about asking for address

Sunny was scaring off users by demanding their address upfront. I adjusted the prompt to first share general estimates and explain the benefits of providing an address, building trust before asking for personal details.

Unclear why conversations ended

Were users leaving because they got what they needed or out of frustration? I added a quick feedback option so users could share why they left and provide further input if they wanted to.