The question in every newsroom right now is what to do about AI-generated content. Media outlets are deciding when disclosure is necessary to establish standards for the use of generated text and images, all while navigating a more fundamental challenge: maintaining reader trust at a time when audiences have more reasons than ever to question the information in front of them.
At the same time, synthetic research, i.e., the practice of using AI to simulate real respondents and generate ostensibly human insights, is entering the media ecosystem with considerably less scrutiny. Rather than recruiting real people to take surveys, companies use AI to create simulated participants designed to match certain demographic or behavioral characteristics. These AI-generated participants then answer survey questions as if they were real respondents. But their answers do not come from actual people. They are based on the information and assumptions used to create the simulated participants, which raises a number of questions, including how accurately they reflect what human audiences actually think.
The issue here isn’t that everyone using synthetic data is attempting to mislead journalists. The problem is that we’re treating synthetic research as a simple question of research methodology and looking past the question of whether synthetic research is trustworthy (hint: it’s not). Further, once those findings are being offered to journalists as evidence, it becomes a question of journalistic integrity, too. In short, it’s an ethical mess.
What a journalist is actually “buying” when they cite a stat
I’ve spent a lot of time thinking and talking about what makes research useful to a journalist, and I’ve realized it’s easy for people to reduce that value to the headline number: 64% of consumers say X, or four in five CEOs believe Y.
But a flashy number can only get you so far. When a journalist cites a statistic, there’s an implicit expectation that the number holds up. They may want to understand the sampling approach or request the methodology before deciding whether the finding belongs in a story.
That can sometimes feel like administrative detail, but it’s actually what makes the research usable. A reporter does not have to personally rerun a study because the PR person has supplied enough information for them to decide whether it’s worthy of publication.
That’s the implicit contract between PR and the media: verifiability in exchange for coverage. A contract that’s becoming increasingly important as AI gets used to pump out more and more content.
“How do you know that?” has no good synthetic answer
The challenge isn’t necessarily that synthetic research produces obviously absurd results. In fact, it doesn’t, and that’s a huge part of what makes the whole thing difficult to navigate. A synthetic panel can generate answers that sound remarkably plausible, making the resulting data look much more like conventional research than it actually is.
But when a journalist asks, “How do you know that?” the answer is fundamentally different.
With traditional research, you can point to the sample, the field dates, the questionnaire and the methodology. There’s a chain of evidence connecting the statistic in the story to people who actually answered a question.
With a synthetic panel, the methodology is ultimately a model and a prompt.
When the objective is exploration, that answer may be perfectly acceptable. But it becomes much harder to defend when the resulting data is presented to a journalist as though it carries evidentiary weight. After all, we’ve all had experience with AI inventing its own facts.
The disclosure gap
By the time a journalist sees a statistic, the fact that the underlying “respondents” were simulated rather than surveyed may no longer be obvious (or may not be mentioned at all).
That is the disclosure gap.
If newsrooms believe audiences should understand when an image or article has been generated by AI, it seems reasonable to assume the same principle should apply when a statistic has been AI-generated.
The PR-journalist relationship is built on trust, just as the journalist-audience relationship is. If any one of those parties jeopardizes that trust, the whole system breaks down.
Why research needs to survive scrutiny, not just land coverage
This is something we think about when we produce original research studies at Scribewise. If we’re going to put a number in front of a journalist, we have to assume that at some point someone is going to ask whether the finding holds up beyond the headline.
In other words, we’re not just thinking about whether a finding is interesting enough to get coverage. We’re thinking about whether it can survive the coverage—and that distinction is becoming more important as AI becomes a bigger part of the information environment.
If newsrooms are increasingly asking, “Was this AI-generated?” then data-driven PR should be prepared to answer a similar question about the evidence behind its stories: “Was this AI-simulated?”