Submission volume is up! We should be excited! Right?
Not necessarily.
Since their popularization in 2022 with the launch of ChatGPT, large language models (LLMs) have transformed how people work, making tasks that once took hours, days, or even months possible in a fraction of the time. Scholarly publishing is no exception, and the growing use of AI is raising new questions about research integrity and editorial workflows.
One pressing concern is surging manuscript submissions. As journals face more papers to evaluate, determining which submissions merit further consideration (and identifying potential research integrity concerns) can put additional strain on editorial teams.
For example, the AI task force for Organization Science reported that their submission volume had increased 42% since 2022, while writing quality had simultaneously declined. And they’re not alone.
Some Scholastica customers have reported similar surges in submission volumes, likely related to AI.
Scholastica CEO and Co-Founder Brian Cody has been following these developments closely as part of Scholastica’s product roadmap planning and participation in broader industry conversations, including the 2026 ISMTE Conference session “Ready for Takeoff: Putting AI safeguards (and slop guards) in place.”
I sat down with Brian to get a better understanding of his current take on how journals are managing increased submissions from a process perspective and where he thinks software can help alleviate the most work. Below is a transcript of our conversation (edited for readability).
Discussion Transcript
ML: When a journal’s submissions increase significantly, where does the strain tend to appear first (e.g., technical checks, desk review, reviewer recruitment)? What workflow data should teams assess to identify bottlenecks?
BC: First, I want to validate that there is a strain. At Scholastica, we’re seeing submission volumes increase across disciplines this year, generally at a faster pace than last year. This is a real challenge for journals, but it’s also an opportunity because more scholarly work is being produced. Journals have relationships with researchers producing more good science, but they’re facing greater submission volumes without necessarily having more resources.
The specifics vary by team, but I think the strain tends to emerge as a people problem before a technical one. For some teams, one person handles technical checks, and an increase in submissions suddenly takes up more of their time. For others, a volunteer EIC may struggle to get through all the papers. Other teams might manage those stages well but find that their reviewer pool is suddenly saturated because the people who typically review for them are already handling another manuscript.
When assessing workflow data, I’d start with time to first decision and compare it year over year. If that time is increasing, it makes sense to break the process down by stage. Look at the workflow from submission to technical checks, reviewer invitations, completed reviews, and the final decision.
It can also be helpful to look at the data by person rather than by stage. If an editorial team has multiple handling editors, you might see significant differences in their turnaround times. I’ve seen publishers encounter this when a particular handling editor is responsible for a certain article type or subdiscipline that suddenly receives many more submissions, causing delays.
So, consider both the stage and the person responsible for it when investigating an increase in overall decision times.
Last, I’ll note that if the time to decision is not taking longer, it’s still worth checking in with the team. People might be maintaining that key performance indicator (KPI) by taking time away from other projects or working under constant stress. That’s not sustainable, and the overall metric might be hiding an underlying strain. Those findings could prompt teams to look for ways to improve efficiency or redistribute staff workloads to reduce pressure.
ML: What opportunities do journals have to use peer-review software more effectively to manage increased submission volumes? Are there particular features or workflow automations that editors may be underusing?
BC: One of the first things I’d consider as submission volumes increase is what can be addressed as early in the process as possible.
For example, at the submission stage, think about what information you’ll need and require it upfront. You can also educate authors earlier in the process. For example, some journals have added more validation to ensure that required files, disclosures, and other information are provided before a submission can proceed. We also see journals requiring datasets to be publicly available at submission for reviewers to assess and to promote transparency.
Thinking about these requirements upfront can save time and prevent situations where a manuscript moves through peer review only for the editors to realize something is missing. From a research integrity standpoint, tools such as Similarity Check can also help journals identify potentially problematic articles earlier in the process.
Another opportunity is to make full use of your peer review software’s functionality. At Scholastica, we roll out new functionality regularly, with larger updates every quarter. It’s very easy for editors to miss announcements about new features, especially if they’ve been using a system like Scholastica for years or even a decade. There may be useful functionality they simply don’t know about or haven’t enabled.
You can explore the available features, but I also encourage editors to contact their peer review software vendor and ask to review their workflow together. We do this when journals first join Scholastica, but revisiting that conversation every couple of years can be valuable.
You might discover that a task you’ve been doing manually could be partially or fully automated, reducing the number of clicks or steps involved. As submission volumes increase, software capabilities are also evolving, and those improvements may help you manage the editorial workload.
ML: What changes have you seen editors make to their desk-review workflows to weed out low-quality or poor-fit submissions earlier?
BC: One helpful approach I’ve seen is educating potential authors earlier in the process, particularly through submission guidelines. Review your guidelines and ask, “What problems can we solve before a manuscript is submitted?”
In the software world, we often think about bugs. If something goes wrong, we ask how we can fix it and prevent it from happening again. That’s the idea behind regression testing. I’ve discussed applying a similar approach with editors when they receive manuscripts that aren’t a good fit for their journals.
For example, an article might be well-written and scientifically sound but still fall outside a journal’s scope. Rather than treating each poor fit as an isolated issue, editors can ask what they could do earlier in the process to help prevent it.
One option is to clarify the journal’s submission guidelines. Another is to introduce questions or confirmations into the submission form. You could present the journal’s scope and ask authors to identify which areas their articles address. You might also ask them to confirm their understanding of several statements about the journal’s focus. These prompts can help authors recognize when their work may not be the right fit.
The goal is to educate authors earlier, both through the guidelines and through the submission form itself.
Another opportunity is to move certain technical checks earlier in the workflow. Some journals conduct checks after acceptance, before an article enters production. It’s worth considering whether any of those checks should happen at submission instead, before the editorial team and reviewers invest time in evaluating a manuscript.
As I mentioned before, if a dataset must be deposited in a repository, require that upfront. The same applies to required images, disclosures, and statements about AI use.
The goal is to identify potential blockers early, rather than investing time in a manuscript only to discover later that it doesn’t meet a requirement for publication.
ML: You’ve previously cautioned that research integrity tools aren’t plug-and-play and implementations can sometimes unintentionally create work through ambiguous results or false positives. What advice would you give editors assessing screening tools today, and how does it compare to a year ago?
BC: I’ve spoken with different publishers who have found it helpful to pilot or trial these tools before fully implementing them, for two reasons.
One is that these tools usually need to be configured for a journal’s specific content, discipline, and publishing workflow.
The second is that teams need to determine what they’ll do with the information the tools provide. Most research integrity tools don’t simply give you a thumbs up or a thumbs down result. Instead, they provide one or more metrics or signals, which might include confidence scores, percentages, tiers, or lists of potential problems.
If you integrate a tool into your workflow and it provides five different indicators, how will your team process that information? What steps will you take in response? These are important questions because the answers can vary significantly between publishers.
Looking back a year or a year and a half ago, I remember speaking with a publisher that was trialing a tool to detect AI-generated text in articles. During the pilot, the team began receiving signals and had to determine how to use them.
For example, if a tool flagged two paragraphs as potentially AI-generated, would the team email the author, conduct an internal review, document the finding, or take another approach? But the results weren’t definitive; they indicated a likelihood.
Think about a tool reporting a 67% chance that four sentences in a paper were AI-generated. What should the team do with that information?
The publisher worked through these questions and developed potential processes. By the time that work was complete, about a year later, the team was questioning whether it wanted to use the tool at all.
And, I’ll note, some of these pilots can be expensive because they rely on server resources to process text, with costs associated with the amount of content being analyzed.
Ultimately, editors should evaluate not only whether a tool can identify potential problems but also whether they have the processes and resources to act on its results.
ML: As discussed during the ISMTE session, “Ready for Takeoff: Putting AI safeguards (and slop guards) in place,” many are questioning the viability of using AI-writing detectors in editorial workflows. What are the limitations of relying on these tools to identify AI-generated or low-quality submissions, and what approaches should editors consider instead?
BC: One challenge with tools that identify AI-generated text, image manipulation, or issues with data is that it’s a little bit of an arms race. Meaning, even as the tool is trying to evaluate if something did come from AI, AI models themselves are changing, so it’s a moving target.
Also, people can give AI systems specific instructions to avoid recognizable patterns. Two years ago, for example, people may have pointed to em dashes, lists of three, or certain phrases as signs of AI-generated text. Today, it’s very easy to give instructions to AI and say, “Here’s an article, don’t use any of these 50 things when you’re generating text for me.”
That makes detection difficult.
When evaluating these tools, I think teams should consider where AI can be helpful rather than approaching AI use as inherently negative. There has been a lot of fear surrounding AI, but I’ve heard examples of journals taking a more nuanced approach.
For instance, some journals ask authors who use AI for grammar cleanup, copyediting, or language assistance to provide the manuscript before that editing takes place. The goal is to assess the underlying work and ensure that the editing process hasn’t affected its meaning, particularly when authors rely on AI for language support in an area where they may not be experts.
I recently spoke with a managing editor who said that their team was finding that AI-assisted grammar editing was saving time. They were receiving manuscripts with stronger grammar and copyediting, including some papers that editors thought had good science but might previously have been borderline candidates for peer review because of the amount of editing it would require upfront.
Those manuscripts were arriving closer to the journal’s standards, making the editorial process more efficient.
If editors focus only on whether AI was used, they might miss that benefit. If you simply think AI equals bad, that could be a false negative.
ML: Looking ahead, where do you see the greatest opportunities for software to reduce editorial strain without encroaching on editorial judgment? What types of automation do you think have the most potential to help editors manage growing submission volumes?
BC: I think there’s a lot of opportunity for areas around technical checks.
I recently spoke with a managing editor whose team had to verify that tables were included as separate files rather than in the main article, and that images and figures were also submitted as separate attachments, in accordance with the journal’s requirements.
That is something a computer can do really well. It can look in the DOCX, which is XML under the hood, look for a table or something table-like, and flag content that may not meet the journal’s requirements.
That is just a small example, but I think there are many technical checks that could be automated. Anything that involves digital content, generally, computers can be very good at. I think that’s an area where there’s a huge opportunity to continue to help people save time.
Reviewer suggestions are another area with potential. There’s a lot of time spent even deciding who you want to invite, and for a lot of people, they have a mental set of rules they’re following. Software could help suggest reviewers based on those workflows.
You can still evaluate the final list, but think about cutting out the clicking, typing, searching, creating a list, and then evaluating. It might help you jump forward to be able to make that final decision about whether it’s a good list.
I am very optimistic about the number of small clicks that can be saved that we do every hour. Hundreds a day, thousands, or millions a year, that add up to real bandwidth.
Years ago, I heard someone in tech in Chicago discuss how even a 10% improvement can have a major impact over a couple of years. I think about that when considering editorial workflows: What if you could find a way to save 2% or 3% of your time this quarter? Those improvements could add up to meaningful efficiency gains over time.
I think that’s going to be very possible for a lot of us, as we have more computer processing power and we have some AI-augmented tools to help us evaluate where we are.
When we’ve been doing something the same way for a long time, we should keep asking ourselves: Are there more efficient ways to do this?
Have thoughts or questions about managing more submissions in the age of AI? Let us know in the blog comments or on social media. You can find Scholastica on LinkedIn and Bluesky!
Don’t forget to follow the Scholastica blog for our latest scholarly publishing recommendations and product updates.
