Research teams are losing influence to peers using AI, but the fix isn’t a prompt engineering course. The real gap isn’t technical. It’s the ability to critically evaluate what AI produces, and most upskilling programs aren’t built for that.
Qualtrics surveyed more than 3,000 research professionals last year. Teams not using AI were four times more likely to lose organizational influence than teams that were. The data is fueling a predictable response: more training budgets, more upskilling programs, more conversations about closing the AI skills gap. That’s a reasonable response. It’s also missing the more interesting question: why are so many teams already investing in AI and still not getting value out of it?
The Conversation Is at the Wrong Level
The standard answer is training. Learn prompting. Take a course on AI fundamentals. These are not bad suggestions. But they treat the problem as a technical one, and most of what is actually going wrong is not technical at all.
The same Qualtrics study includes a finding that gets far less attention than the four-times statistic: 39% of research leaders say AI has revolutionized their workflows, versus 19% of frontline staff. That twenty-point gap is not an AI literacy problem. It’s a translation problem. Leaders are measuring outcomes at the workflow level and seeing gains. The researchers doing the actual work are running AI on specific tasks and hitting the same judgment calls they always faced, just faster, with outputs they are not always sure they can trust.
What AI Literacy Actually Means in Practice
The GRIT 2025 Insights Practice Report found that 33% of research buyers now value AI literacy over traditional research expertise when evaluating agency partners. That sounds like a mandate to hire people who can code. It is not.
AI literacy in a research context is specific: the ability to critically evaluate what an AI tool produces. Not to build it, not to fine-tune it. To read the output and know whether it is right, where it is likely wrong, and what happens downstream if a team acts on a finding shaped by a model’s blind spots.
That is a harder skill than prompt engineering. It requires genuine depth in research methodology, not just familiarity with a tool interface. A researcher who understands sampling variability, response bias, and how large language models handle ambiguity is in a very different position than one who can write a fluent prompt but has no basis for questioning the result.
93% of organizational AI spending goes to the technology itself, while just 7% goes to training the people using it. The tool is not the constraint. The judgment is.
Why Platform Design Is Part of the Answer
This is also why the choice of research platform matters more than most upskilling conversations acknowledge. Generic AI tools hand researchers a blank interface and expect them to know the right questions. Purpose-built research AI is designed around the decisions researchers actually make: whether to trust a finding, how to handle anomalous data, when a sample looks off, where human review is not optional.
That design difference changes what AI literacy requires in practice. A platform that surfaces uncertainty, flags low-confidence outputs, and routes edge cases to human review reduces the demand on a researcher’s critical judgment. One that produces polished-looking outputs with no indication of where the model is extrapolating puts all of that burden back on the analyst.
Upskilling matters. The 70% of workers who complete AI training when employers actually make it available is a better number than most people expect. But training people on tools that were not designed around their real judgment calls is a partial fix at best. The more productive frame is not how to train researchers on AI. It’s whether the platforms they’re using make critical thinking easier or more necessary.
Most teams already know the answer to that question. They just haven’t been asked it out loud. See how Quali-Fi builds evaluative guidance into the research workflow ->