Key findings from APPAM’s 2026 member survey on AI in public policy research and teaching
Of the many questions people are asking about AI, the one that keeps coming up in my role as a researcher is: how will AI impact research? Will it replace the need for data cleaning? Modeling? Even coming up with a hypothesis?
We’re now starting to see survey results that shed light on how researchers are using—and thinking about using—AI tools in their work. A May post from Anthropic revealed that more than 80% of social scientists (from a sample of 1,260 surveyed in February and March of this year) have tried using AI chatbots in their research, but only 20% have adopted coding agents into their work.
The Association for Public Policy Analysis and Management (APPAM) recently released a survey of 322 of their members (conducted between April 29 and May 20, 2026). The results reveal a general embrace of general-purpose AI tools for work, with nearly two-thirds of researchers incorporating them weekly. These results give us another glimpse into the uses (and the fears) researchers have when it comes to AI.
Below, I summarize what I see in the data. The dashboard, which was constructed with the help of Anthropic’s Claude, lets you explore it further on your own.
Most members are using AI frequently
Eighty-four percent of respondents have used a general-purpose AI assistant for research or policy analysis in the past year. Nearly two-thirds use these tools at least once a week, and a third use them daily. So, the technology is not on the horizon; it is already embedded in how many APPAM members are doing their work.
But the type of use matters. Adoption drops sharply beyond the basics: only 29% have used AI coding tools, 19% have used specialized research AI, and just 8% have accessed AI through an API. The majority of members are using AI as a general assistant, not as a specialized research instrument. It isn’t clear to me why: Are researchers not sure how to use these tools? Are their institutions not providing them with training or the appropriate guidance? Or maybe they are using restricted data and don’t have standalone server-deployed models they can use for those kinds of data.
Learning AI tools is informal
Perhaps the most striking finding is how members learned to use these tools: 80% taught themselves, and 43% learned from colleagues. Only 16% received training from their university or employer.
I wonder if that informal learning path has its limits. On the one hand, I feel like using AI tools for basic tasks just requires the user to keep asking the tool questions. But the APPAM survey results suggest something different. While nearly two-thirds of respondents report minimal or no formal training on AI tools, only one in five members feels very confident using AI effectively in their work. And a third describe themselves as not very or not at all confident.
Concerns are high
Researchers are hesitant about using AI. Eighty percent of respondents cite concerns about accuracy, reliability, or usefulness as a barrier to expanding their AI use. Data privacy and ethical concerns each register at 60% or above. These are not the objections of people unfamiliar with technology, but I think they reflect the concerns of researchers who understand what is at stake when AI-generated content finds its way into policy analysis.
And although there weren’t many, the open-ended responses do reinforce this. When members were asked to elaborate on their experiences, the most common sentiment was negative (66%), with themes centered on institutional AI policies, academic integrity, and questions of reliability.
The profession sees change coming and wants rules and safeguards
There is near-unanimous agreement that AI will transform the field: 81% of respondents agree or strongly agree that AI tools will significantly change how public policy research is conducted in the next five years.
And yet the profession lacks the norms to govern that change. Ninety-one percent of respondents agree or strongly agree that the field needs clearer standards for when and how AI tools should be disclosed in research. Nearly four in ten works at institutions with no formal AI policies at all. In the places where I teach as an adjunct, I see this as well—there is little in the way of guidance on how to write AI policies into my syllabi, let alone how to use AI tools responsibly in research.
APPAM (and other membership organizations) has a clear role to play
This is where APPAM as an association has work to do. Members are clearly hungry for guidance. Seventy-six percent said they would find APPAM programming on AI tools useful or very useful. When asked what kind of support would be most valuable, the top answers were best practice guidelines (66%), hands-on workshops on AI coding tools (53%), and guidance on data security and compliance (40%).
The message from the membership is consistent: they are using these tools, they have real concerns about doing so responsibly, and they want their professional association to help them do it better.
