Research integrity has occupied a lot of oxygen in scholarly publishing over the past few years. Between paper mills, fabricated references, and increasingly convincing AI-generated manuscripts, publishers have plenty to worry about.
But there's another question lurking behind all that integrity work: What happens when a manuscript isn't fraudulent, but it isn't particularly good, either?
That was the starting point for a conversation moderated by Hum's Dustin Smith at Silverchair's Platform Strategies 2026, featuring Miriam Maus of IOP Publishing, Dawn Melley of IEEE, and Jonathan Woahn of Cashmere.
The integrity problem is only part of the problem
Dustin opened with an interesting disconnect: By some estimates, only 3–5% of submissions raise genuine research integrity concerns; yet desk rejection rates are considerably higher.
That gap represents a much bigger editorial challenge. Publishers aren't just trying to keep fraudulent research out of the scholarly record, they’re trying to determine which legitimate research deserves to be in it. And unlike checking for a fabricated reference, assessing the scientific contribution of a manuscript isn't exactly a yes-or-no exercise.
"Does this piece of content make a contribution to scientific discourse?"
It's a deceptively simple question, especially when different journals, disciplines, and research communities have very different ideas of what constitutes a meaningful contribution.
Producing a paper has become the easy part
Submissions are climbing across the industry, with some journals experiencing year-over-year growth of 80–100% and certain conferences seeing even more dramatic increases. As Dustin said, "It's effectively free to create a research-shaped object."
AI has made it easier than ever to produce something that looks and reads like a scientific paper. That doesn't necessarily make the research fraudulent, but it certainly doesn't guarantee that it's valuable.
Publishers don’t just have to figure out how to process more submissions, faster. They need to focus on how to make better decisions about which submissions merit the limited attention of editors and reviewers.
On the panel, Miriam described a striking reversal at IOP Publishing. Five years ago, the priorities were attracting more submissions and speeding up peer review. Today, the conversation increasingly centers on setting appropriate hurdles for manuscripts that aren't ready for publication.
Dawn described a similar shift at IEEE, where volunteers are actively exploring ways to introduce more friction into the submission process.
After decades spent making publishing easier for authors, deliberately making parts of it harder is quite a change of direction.
Better judgment, not just faster workflows
Of course, friction for its own sake isn't the answer.
Jonathan brought an interesting perspective from lean manufacturing, where the goal isn't necessarily to slow a process down to improve quality, but to prevent mistakes from happening in the first place.
Miriam pointed to the role technology can play in applying that principle to peer review: performing routine checks earlier, before a manuscript ever reaches a reviewer, so human experts can focus on the scientific contribution.
She specifically mentioned Hum's Alchemist Review as an example of technology helping editorial teams do more of that work upfront.
The goal isn't to automate the judgment that makes peer review valuable. It's to give editors and reviewers better information, earlier, so they can spend their time on decisions that actually require their expertise. And those decisions won't always be straightforward.
A question from the audience raised one of the most important tensions in the discussion: What happens to novel research when we get too good at filtering submissions?
Scientific breakthroughs don't always announce themselves as breakthroughs. Sometimes the value of a piece of research only becomes apparent years later.
Dustin acknowledged the limits of comparing scholarly publishing to manufacturing. Science, he observed, is a much messier endeavor: "This is a collective of humans who are trying to validate our wisdom as a species." That's not something you can reduce to a defect rate or an automated checklist.
Making room for the research that matters
If there was a common thread throughout the discussion, it was that scholarly publishing needs to get more sophisticated about quality without becoming more rigid about what quality looks like.
Integrity checks remain essential. So do efficient editorial workflows. But neither can fully answer whether a manuscript advances its field, serves its research community, or might turn out to matter in ways we can't yet predict.
As it gets easier to produce something that looks like research, the ability to recognize research that matters becomes even more valuable. And that's a challenge worth putting some very smart humans (and a few very smart machines) to work on.