The Reality: A Bifurcated Industry
A clear divide is emerging in the scholarly publishing landscape. On one side are "AI-native" workflows that leverage automation to compress timelines. On the other are legacy processes that scale linearly with human effort. This bifurcation is not merely about using new tools; it is about survival economics.
The Speed of Science
The primary differentiator today is time. In a manual workflow, literature review and manuscript triage are bottlenecks that consume weeks of high-value staff time. AI tools have dismantled these constraints. Early testing of AI-driven literature review tools suggests they can reduce the time spent on research tasks by approximately 50% 11. This allows researchers and editors to reach the hypothesis generation or decision phase weeks earlier than previously possible.
The acceleration extends to quality control. Frontiers’ AIRA system, for example, performs over 40 distinct quality checks—ranging from language quality to image integrity—in under two minutes 28. A human editor performing the same depth of analysis would require hours. When an organization can validate a submission in minutes while a competitor takes days, the competitive advantage shifts decisively to the former.
The Economic Disadvantage of Manual Labor
The most critical risk for legacy workflows is economic. Manual processes operate on linear cost curves: processing 1,000 manuscripts costs roughly ten times as much as processing 100. As submission volumes grow, costs balloon and backlogs form.
AI-driven workflows, conversely, offer scalable efficiency. Once the infrastructure is established, the marginal cost of processing additional manuscripts decreases. Costs for automated research generation are dropping precipitously; experimental systems like the "AI Scientist" can now generate a full scientific paper for approximately $15 30. While this specific example represents an extreme edge case, the trend is undeniable. Organizations relying on manual labor for routine tasks cannot compete with the unit economics of automation.
Why This Happens: The "Blob" Problem vs. Structured Data
If the benefits of AI are so clear, why are many organizations struggling to adapt? The answer often lies not in their software, but in their file formats. The industry's historic reliance on the PDF is becoming its greatest liability.
The PDF Trap
To a human reader, a PDF is a polished, final document. To an AI model, a PDF is often a "messy blob" of unstructured data. Without semantic tagging, an algorithm struggles to distinguish a figure caption from a reference, or a methodology section from a funding statement. This lack of structure introduces noise. When AI models attempt to extract data from these "blobs," error rates increase, and the risk of "hallucinations"—where the AI invents plausible but false information—rises.
The XML Imperative
The organizations winning the race for efficiency have adopted an "XML-first" mindset. JATS (Journal Article Tag Suite) XML serves as the connective tissue that makes AI possible 25. Unlike a flat text file, JATS XML wraps every element of content in semantic tags. It tells the machine definitively: "This is an author affiliation," "This is a p-value," and "This is a clinical trial ID."
This structure is the prerequisite for "Agentic AI"—systems that can autonomously plan and execute tasks. When content is structured, AI agents can reliably extract metadata, verify citations, and reformat references without human intervention 6. Without this design infrastructure, AI tools are forced to guess, rendering them unreliable for high-stakes scholarly publishing.
Garbage In, Garbage Out
This technical reality dictates a strategic truth: you cannot automate what you have not structured. Investing in expensive AI tools while maintaining a PDF-based workflow is a recipe for failure. The models will lack the context required to function accurately 46. The shift to AI requires a foundational redesign of how information is architected from the moment of submission.
What Leading Organizations Do Differently
Successful adaptation to this new era does not mean replacing humans with machines. Instead, leading organizations are aggressively reallocating human capital. They recognize that paying a PhD-level editor to check reference formatting is a misuse of talent.
From Administration to Strategy
Forward-thinking publishers are shifting their teams from administrative drudgery to strategic oversight. By automating the 80% of tasks that are routine—formatting checks, metadata verification, basic language polishing—staff are freed to focus on the 20% that requires deep expertise: assessing scientific novelty and developing journal strategy 9.
This shift also addresses the "burnout" crisis in peer review. By using AI to filter out technically flawed manuscripts before they reach reviewers, publishers reduce the burden on the academic community. Tools like Paperpal Preflight have been shown to lower rejection rates by 15.5% by catching errors early, ensuring that reviewers only see manuscripts that are ready for serious scientific scrutiny 14.
Integrity as a Service
Perhaps the most powerful application of this new workflow is in research integrity. Humans are notoriously poor at detecting manipulated images or subtle statistical anomalies across thousands of papers. AI excels at this pattern recognition. The Journal of Clinical Investigation (JCI) implemented Proofig AI to screen for image issues and saw their rejection rate for data integrity issues triple from 1% to 3% 20.
Crucially, these systems operate with a "Human-in-the-Loop" model. The AI acts as a scout, flagging potential duplications or manipulations, but a human editor makes the final judgment 21. This preserves the role of the editor as the guardian of the scientific record while equipping them with tools that match the sophistication of modern fraud.
Key Takeaways
- Structure First: Automation capabilities are capped by data quality. Transitioning from unstructured PDFs to semantic JATS XML is the necessary first step for AI readiness.
- Speed is Strategic: In an open-access world, timeline compression is a competitive advantage. Reducing administrative drag attracts higher-quality submissions.
- Augment, Don't Replace: The goal of research workflow automation is to remove administrative burden, allowing human experts to focus on scientific validity and strategy.
- Integrity requires Scale: As submission volumes rise, only AI-assisted workflows can maintain rigorous checks on image and data integrity.
Subscribe to our insights to stay ahead of the evolving landscape of research publishing and design strategy.
Footnotes
1 wiley.com, "Source from wiley.com," 2025. Link Confidence: Medium
6 anara.com, "Agentic AI for literature reviews: The complete guide," 2025. Link Confidence: Medium
9 integranxt.com, "A Beginner’s Guide to Optimizing AI in Your Publishing Workflow," 2023. Link Confidence: Medium
11 rdworldonline.com, "Can ScienceDirect AI cut literature review time by 50%? Early testing with 30,000+ users shows strong potential," 2025. Link Confidence: Medium
14 penelope.ai, "Case Study - BMJ Open — Penelope.ai," 2018. Link Confidence: Medium
20 proofig.com, "How Academic Fraud is Undermining Research Integrity," 2025. Link Confidence: Medium
21 proofig.com, "The Journal of Clinical Investigation talks about Strengthening Scientific Integrity with Data Accuracy via Proofig AI," 2025. Link Confidence: Medium
25 sciflow.net, "JATS-XML: The Emerging Standard in Academic Publishing for Everyone?," 2025. Link Confidence: Medium
28 youtube.com, "Meet AIRA, Frontiers’ proprietary AI-powered assistant - YouTube," 2025. Link Confidence: Medium
30 sakana.ai, "The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery," 2024. Link Confidence: Medium
46 nih.gov, "From Valid XML to Valuable XML: When “Good” Matters More Than “Valid” - Journal Article Tag Suite Conference (JATS-Con) Proceedings 2025 - NCBI Bookshelf," 2025. Link Confidence: Medium