The world of generative AI, led by platforms like ChatGPT and Gemini, is changing how we think about information access. For nonprofit professionals dedicated to connecting people with vital resources, the question isn’t if AI will change our work, but how we can use it effectively and responsibly.

At One Degree, we are focused on developing a One Degree AI Assistant to be a trustworthy and powerful tool for resource search and access. As this is our first entry into generative AI, we conducted a series of user tests and interviews with our Community Feedback Group. This group comprises a diverse mix of help-seekers, direct service providers, and healthcare workers. Through this process, we’ve gained critical insights into what makes a specialized AI solution stand out from general-purpose leaders in the market.

The Problem to Be Solved

Our community members face persistent, systemic challenges when searching for help, including:

• Wading through outdated and inaccurate resource information.
• Wasting time on long eligibility processes only to find they don’t qualify.
• Having to fill out the same paperwork repeatedly.

While a gentle optimism toward AI exists, citing our recent survey data, so do distinct fears. This includes concerns regarding resource verification and accuracy (“Will AI just make up information?”), the loss of human empathy, the potential for algorithmic bias, or unethical data use. This can be an incredible risk for people facing vulnerable situations. It takes an intentional and rigorous effort for AI-enabled solutions to have the community’s best interests at heart.

Comparing the AI Landscape for Resource Search

We sought comparison feedback on our in-development AI Assistant alongside general-purpose models like ChatGPT and Gemini. We did this through several rounds of user testing, comparison feedback, and in-depth remote interviews with our community. Here is what our participants found:

One Degree AI Assistant

The One Degree AI Assistant was described by one tester as: “a ChatGPT for resources for people.” The familiar chat interface was intuitive, and testers appreciated its focus.

• Strength: Focus and Actionability. “I think that it gives you what you need and then some,” noted one participant. Participants found the information level “enough to take action on a resource but not too much to be overwhelming.” Its core value lies in being intentionally focused on verified, actionable community resources.
• Weakness: Speed and Transparency. Compared to the high-speed, advanced back-ends of the larger platforms, some testers noticed a slowness in our Assistant’s responses. Furthermore, a tester noted that they missed the visual cues that platforms like ChatGPT use to show an “undergoing and thinking process” such as real-time typing. Though subtle, this provides transparency and signals to the people that “Something is happening.”

ChatGPT and Gemini

Many research participants stated that they had familiarity with ChatGPT, which made our Assistant easier to interpret initially. Even though those who said they weren’t particularly tech savvy shared that they had experience using generative AI platforms. However, when tested directly for resource access, the limitations of general AI became clear:

• Strength: Thoroughness and Layout. Some participants appreciated the sheer volume of information these platforms provided. ChatGPT was also noted for using tables to organize results, making resources easier to compare and contrast. Tags with linked sources on some text also offered a layer of trust by stating the source of the information, including a hyperlink.
• Weakness: Overwhelm and Unpredictability. A common critique was the sheer volume of text, which was seen as “long-winded and difficult to read.” One interviewee commented on Gemini: “It is visually difficult to process… It’s just a wall of text to me at this point.” Crucially, the information was often unpredictable. Sometimes contact numbers were present, other times they weren’t. Participants found themselves having to take an organization’s name from the AI response and “do more research just to find a phone number. It’s just a lot of waste of time.”

Traditional One Degree Search

Interestingly, the traditional, non-generative AI One Degree search with its card-based interface received positive feedback for its readability and predictable format.

• Strength: Readability and Consistency. Participants appreciated the structured format, which made addresses, phone numbers, and hours easy to locate. “I like the break up of different things… it just doesn’t look like a report I’m reading,” one person shared. The predictable layout eliminated the need to “hunt for information,” where time is a precious commodity in urgent situations.

Key Takeaways

Our research confirms a crucial point for the nonprofit sector: Given the existing field of generative AI tools, there is still a clear need for AI training and specialization to work effectively as a tool for social service navigation.

While general platforms like ChatGPT and Gemini offer global scale and flexibility across many topics, their data, user experience, and use cases are not optimized for social service navigation. Because of this, we are working to train a specialized AI Assistant to  focus on:

1. Trustworthiness: For our community, accuracy and relevance are critical, not just convenient. Standard AI models pose an unacceptable risk to vulnerable populations because they can hallucinate or provide out-of-date information. Even when an AI platform offers credible information, it must provide assurances to verify its accuracy. But the fact that it’s unclear where they’re pulling the data from and how accurate it is, significantly erodes the trustworthiness of the dominant AI chatbots.
2. Information experience: Throughout user testing and in-depth interviews, a clear preference emerged: participants preferred information to be succinct and well-structured. They don’t want to comb through a conversational chat or an in-depth analysis. They need specific details that allow them to take the next step toward accessing a resource.
3. Value proposition: The One Degree AI Assistant is powered by One Degree’s curated, human-verified database of nonprofit services and public benefits. Every response is grounded in accurate, up-to-date, and community-reviewed data — not scraped from the chaotic sprawl of the internet. This means our AI doesn’t hallucinate, doesn’t guess, and doesn’t send people down dead ends. Instead, it surfaces only verified, useful, and available resources. That focus is what makes One Degree AI a tool our community can trust by default.

One participant shared their experience using ChatGPT and Gemini to seek resources and stated in an interview, “I haven’t been blown away by anything by AI bots. I haven’t been blown away because I could do the same thing with Google.” This highlights the distinct challenge of having to offer a unique value proposition that existing tools like ChatGPT or Google search don’t specifically address.

We are incredibly thankful for the guidance from our Community Feedback Group. Their stories and insights are steering the development of the One Degree AI Assistant to ensure it is not just another piece of technology, but serves as an essential tool that genuinely reduces the friction between a person and the help they need.