Ever optimistic, I didn’t think finishing this revise-and-resubmit would take this long. It’s late, a mug of sweet cherry tea on my desk and an even sweeter dog at my feet, and I’m checking the instructions of the journal’s submission portal, double-checking the abstract is less than 250 words, and triple-checking that I’ve uploaded the correct version of my manuscript.
Hitting submit, I conjure the image of my manuscript finding its place in the knowledge path. I’ve cited the prior evidence and hope that my manuscript will be cited in future work, perhaps in a meta-analysis or a proposal to fund future research. Once published, this manuscript will find its place between what is known now and what is yet to be known. Some sections of the knowledge path are broadened by ample publications and well-established facts; some are narrow and in need of more research; and some are tangled with conflicting evidence and controversy.
Increasingly, for better or worse, generative AI is in the knowledge path. As researchers, we're consuming communications drafted by generative AI every day: the research summary posted on social media by a professional society, the final report sent out after the conclusion of a clinical trial, the descriptions of drugs and treatments posted on the websites of primary manufacturers, and even that article published in a peer-reviewed journal—the list goes on and on. So, when we sit down to do our own scientific writing, are we using sources and scientific writing drafted by generative AI? We can’t know. So, the burden of quality-checking what we read increases.
Whether you personally adopt AI into your writing process or not, the knowledge path that informs your scientific writing is littered with generative AI content. As researchers who care about the accuracy of our knowledge path, our burden to check for accuracy is undeniably increasing.
The Separation of Quality Checking and Writing in the New Era of Generative AI
With generative AI in the knowledge path, the burden of quality-checking scientific communications is changing. When writing a scholarly and/or scientific piece without AI, checking for accuracy and quality comes naturally in a manual writing process. We check our facts and figures, confirm our sources, re-confirm our sources because we can’t remember the authors’ last names, re-read as we write to make sure we’re not contradicting ourselves, sentence-by-sentence, paragraph-by-paragraph, before moving on to the next. To know what to write, we are critically examining our sources and statements as we write.
Compare this process to writing a first draft utilizing generative AI: We write a quick prompt and hit submit. In less time than it took to type the prompt—voilà!—a seemingly accurate first draft. This efficiency is impressive. But when we hit “enter” on that prompt, the link between writing and fact-checking splits apart.
When generative AI is used to produce the initial writing, important quality-checking steps are skipped during the drafting stage. Ideally, they would be completed as a secondary stage: a quality-checking stage that requires more rigor for generative AI drafts than for manual writing drafts. In other words, in the new era of AI, the work of quality-checking evidence sources, key terms, etc., is separating itself from the writing. This separation is creating a need for better quality checking by writers and readers, as well as better tools to support it.
Maybe work is like matter, and it can’t be created or destroyed; it can only be transformed. Any work saved by using generative AI to create the first draft of a scientific communication is lost to quality checking, either by the “writer” or the reader.
To uphold an accurate knowledge path, there is an immediate need to better understand, study, and implement those steps in a post-drafting stage. It was always good practice to check for accuracy and quality, but our practice assumed manually created writing. That has changed. We can no longer safely make that assumption. We need new practices and new tools.
Tools for Quality Checking
As a scientific writer, I have long relied on tools to execute good work: I use EndNote to organize my references, I use PubMed to download citations, I use Word for character counts, spell check, and formatting, and I use an app to lock myself out of my own phone because my endlessly amusing friends are funny texters. To keep myself organized and accurate as I write, I use a smorgasbord of tools. Now, as a writer and reader who cares deeply about inaccuracies in the knowledge path, I need tools to support quality checking that is increasingly necessary as AI is used more widely in the field.
This is a call for more research into what we lost when scientific writing was no longer exclusively drafted manually, and what challenges we gained when generative AI entered the knowledge path. We need more research to develop the tools we need to quality-check scientific writing.
This is also a call to action for other scientists, researchers, and scholars to shoulder the burden of quality-checking the evidence in your knowledge path. Annoyed by the extra work, I’m asking you to shoulder? That burden is characteristic of this new era of AI: system-level challenges that benefit few and burden individuals.
System-Level Challenge, Individual Level Burden
Reluctance to adopt AI is multi-fold and well-founded. To name a few: AI Data Centers are beyond environmentally destructive and disrupt communities with sound and light pollution, the billionaires profiting from AI won’t live anywhere near these centers but reap their benefit, targeting the Midwest (and cold Great Lakes water) for their construction, plus generative AI in the learning path is hurting critical thinking skills and is disrupting learning paths for profit. Big picture? AI, as an innovation, has an inequitable and unethical distribution of benefits and burdens. However, the fear of falling behind motivates researchers and scientific writers to adopt AI. The individual-level burden is high. The individual-level benefit is negligible. The decision to adopt generative AI in scientific writing may seem like an individual choice. But it’s not. It’s a system-level challenge.
______________
Assistant Professor Cynthia Drake Morrow, PhD, is a health services researcher with a background in bioethics and teaches public health within the Charles Stewart Mott Department of Public Health. Dr. Morrow is also the founder of Knowledge Resolution, a startup developing tools and consulting services for evidence verification.
April 20, 2026