Is AI slop overrunning editors and reviewers? Yes and no.

Charles Hemenway, Director of Publisher Relations, ChronosHub.

When every other industry blog is about paper mills, it’s easy to assume that research integrity is publishers’ biggest AI-related challenge. But the conversations I’ve been having this year challenged my assumption. There’s another concern that’s proving just as significant: the flood of legitimate research that AI is helping to produce.

Fraudulent research vs. AI-assisted research

SSP 2026 was the moment the penny dropped for me. I’d expected that research integrity would be the hot topic around the ChronosHub stand (aside from our fab merch of course). But what many publishers, of all shapes and sizes, actually wanted to discuss was capacity management. What gives?

In essence, the folks I spoke to are in the same boat as Organization Science. Back in April, the journal’s own internal task force released an analysis of their received submissions. Submissions were up 42% since 2022 (the year of ChatGPT’s public release). A big percentage of these showed signs of AI involvement. So far, so unsurprising. But the thing is, the article points away from paper mills or fraudulent research as the only cause.

The biggest tell was in where the surge came from. Universities that visibly changed their submission behavior after the UTD rankings launched in 2005, a scoring system that ranks institutions by publication in 24 prestigious journals, including Organization Science, showed the sharpest increase in AI-assisted submissions post-ChatGPT.

Paper mills don’t care which institution their authors come from. Instead, this paints a compelling picture of faculty at these schools using AI to supercharge their efforts. As the authors of the article write:

“AI adoption is not a purely technical phenomenon or institutionally agnostic. Instead, it appears to be correlated with career incentives that pressure researchers to increase their publication output.”

I have to admit, I hadn’t made the simple connection that the AI tools my team use on a daily basis (Claude, Chat GPT, etc.) were being put to use in research for all of the same reasons. First drafts, sourcing facts and references, and basic forms of data gathering are all being compressed with the introduction of AI.

We can debate the ethics of using AI in scholarly work till the cows come home. But that’s where we are now. And it leaves publishers in a bind. AI-assisted submissions can’t be categorized as fake slop. There’s good science in there. But the sheer volume of research coming in is making it impossible for publishers to sort the wheat from the chaff. And that’s a capacity problem.

An ounce of prevention

I’m going to call on a Founding Father for some inspiration. In what was basically a 1736 PSA on fire safety, Benjamin Franklin wrote, “An ounce of prevention is worth a pound of cure.”

When it comes to combatting paper mills, that’s certainly where the industry’s thinking is. In recent years, industry bodies and third-party vendors have built screening tools to flag fake research before it enters the review process. To take one example, the STM Integrity Hub claims to currently intercept around 1000 paper mill submissions a month. That’s a lot of labor, retractions, and frustration avoided.

There’s a growing consensus that the same proactive approach is needed for the capacity problem too. Mary Miskin, Operations Director at Crimson Interactive/Enago, a leading provider of author services, sees the demand first-hand:

“The solution customers are consistently asking for, is to find reliable methods to eliminate poor quality upstream; it’s no longer just paper mills or fraudulent research they need to identify, but a growing volume of genuine, often AI-assisted submissions that may still be unsuitable or fall short of the quality required.”

For Mary, the answer is moving from screening to triage, with “clear, evidence-based signals around technical quality, scope fit, content quality and ethical fit”, deployed upstream rather than after submission.

By embedding these earlier, you can surface the information to authors, reviewers, and editors at the right time in the manuscript journey, improving submission quality and velocity, and freeing editorial attention for where human judgement matters.

This “reordering” of the publishing process has huge potential. It certainly generated steady SSP booth traffic! It’s the future-state we’re building here at ChronosHub with our integration framework; a data and user experience layer that empowers publishers to selectively deploy author, reviewer, and editor tools where they create the most value, while enhancing the user experience.

Reasons to be cheerful

Ultimately, having a surplus of decent research is a “nice to have” challenge. The publishers I spoke to were refreshingly aspirational about it, in contrast to the gloom still surrounding the paper mill discourse.

Having said that, they are concerned that good research is in danger of being overlooked or delayed. Solutions need to be implemented to make serious strides in the early stages of the manuscript submission process. It was reassuring to hear publisher’s motives for action were truly in service of the common good as they chart their course ahead.

Let’s view AI-assisted research for the systems and process challenge it is. At the same time, let’s also consider it a blessing. Striking a better balance between automated safeguards and human expertise will go a long way to solving these challenges and harvesting the benefits.