For me, medical writing has always been about more than words. I think we might all agree that its purpose is to make science understandable, communicate evidence accurately and help people make better decisions. But as clinical research becomes increasingly structured, automated and machine-readable, an uncomfortable question emerges: are we gradually replacing the craft of writing with the assembly of information?
The latest developments in regulatory documentation, particularly the International Council for Harmonisation's M11guideline for clinical study protocols, suggest that the answer may be more complicated than we think. Standardisation offers genuine benefits, but it also raises questions about professional judgement, creativity and the value of the individual writer in the process of communication.
In my previous article, Finding Your Voice as a Medical Writer, I explored why medical writers need to develop a distinctive voice to establish their purpose and value in modern pharmaceutical development. That argument becomes particularly relevant if the industry increasingly expects writers to follow predefined structures, populate templates and assemble reusable content. And that raises the question: If every document is becoming a kit of standardised components,what happens to the person putting it together?
I certainly don’t want a future in which every medical writing task arrives like a LEGO® set. The box specifies the age range, the number of pieces, the assembly instructions and the expected finished product. The writer's task, simply tofollow the instructions, fit the components together and check that nothing is missing. There might even be a warning onthe box: Professional judgement required only where explicitly indicated.
It is a horrifying prospect. It is also worth taking seriously.
From prose to components
The elephant in the room, medical writing has been becoming more structured for decades. The process did not beginwith large language models or generative artificial intelligence, nor is standardisation necessarily a threat to good writing.
For example, the introduction, methods, results and discussion structure (IMRAD) structure of the traditional manuscript is now so familiar that we rarely question it. But the format emerged gradually. A historical analysis of leading medical journals found that IMRAD only became increasingly prevalent from the 1940s. By the 1970s it made up approx. 80% of original articles and was effectively universal by the 1980s [1]. Structured abstracts subsequently made key information easier to identify, and research examined their quality against that of conventional abstracts [2].
Reporting guidelines took the process further. CONSORT established explicit expectations for reporting randomised controlled trials, with subsequent revisions reflecting developments in methodology and transparency [3]. These initiatives were designed to improve reporting, not to eliminate the writer. Indeed, a checklist can tell us that an outcome must be reported, but it cannot guarantee that the outcome is interpreted correctly or explained meaningfully.
Nevertheless, something important happened along the way. Information became easier to locate, compare and classify. A structure introduced to help the human reader could also help a computer identify the same information. The consequences are becoming increasingly apparent in pharmaceutical development. Research review described lean authoring as a framework for producing clearer, more targeted clinical study reports while preparing content for reuse and automation [4]. Instead of repeatedly recreating the same information, organisations can develop validated components that are reused across documents. An endpoint, for example, might appear in a protocol, statistical analysis plan, clinical study report, regulatory submission and publication. Traditionally, writers have aimed to maintain consistency across these separate appearances. In a component-based environment, the endpoint might be defined once, with its meaning and relationships explicitly maintained, and then incorporated across multiple outputs.
The attraction is obvious: less duplication, better traceability and potentially fewer inconsistencies. But the underlying change is profound. The document is no longer the fundamental unit of medical writing. The information components become the data blocks to be incorporated into multiple outputs.
M11: writing for people, or structuring for machines?
The final ICH M11 guideline, Clinical Electronic Structured Harmonised Protocol (CeSHarP), provides a particularly interesting development. Published as a harmonised guideline, template and technical specification in December 2025, it establishes a common structure for clinical study protocols and supports their electronic exchange between organisations and regulatory authorities [5]. The intention is not simply to make all protocols look alike. Harmonised headings, defined information elements and technical specifications create the conditions for more consistent interpretation and electronic processing.
Yet consider the practical consequences for the medical writer. A protocol has always needed to explain the scientific rationale, justify the design, describe the procedures and communicate the study's requirements to investigators, ethics committees, regulators and operational teams. Under a structured approach, those same requirements must also fit a predefined architecture. For me, this is where the distinction between standardisation and understanding matters.
A template can specify where eligibility criteria belong. It cannot decide whether those criteria are scientifically justified.A controlled field can capture an endpoint consistently. It cannot determine whether the endpoint adequately answers a specific research question. A machine-readable intervention description can facilitate exchange between systems. It cannot, by itself, establish whether the intervention is appropriate for the population.
M11 does not explicitly declare that machines matter more than humans. Its purpose is harmonisation and electronic exchange. Nevertheless, a reasonable inference is that machine processing becomes an increasingly important design consideration when information must conform to a structured, interoperable format.
The danger is not that the protocol becomes structured. It is that we begin to confuse successful completion of the structure with successful communication of the science. And the emerging documents do not make easy reading for humans.
The benefits of building with standardised pieces
There is a strong case for the transformation. Pharmaceutical development involves large volumes of interconnected data/information, and inconsistencies between documents can have consequences beyond poor presentation. Reusable, validated content improves traceability, reduces avoidable duplication and makes changes easier to control.
The FAIR principles formalised the importance of making research outputs Findable, Accessible, Interoperable and Reusable. Importantly, they explicitly recognise the needs of machines as well as human researchers [6]. The same thinking has reached scientific publishing. Stocker and colleagues proposed producing machine-readable expressions ofresearch findings during knowledge creation rather than extracting information from published articles afterwards [7]. Thework suggests that structured scientific knowledge can improve the accuracy, richness and reproducibility of subsequentreuse.
Imagine a clinical study in which the population, endpoints, intervention, statistical methods and results exist asconnected, traceable information. The protocol, registry entry, clinical study report and publication could all be generated,at least in part, from the same underlying data package. This doesn’t necessarily mean that the same paragraphs appears everywhere. Different audiences need different explanations. Rather, the underlying facts and their relationships would be maintained while the presentation might change according to purpose.
That is a potentially valuable distinction. It could release medical writers from repetitive assembly work and allow them to spend more time interpreting evidence, explaining uncertainty and helping development teams communicate their findings. But there is a catch.
When the instructions become the objective
Standardisation is useful when it makes key information more reliable and accessible. It becomes problematic when compliance with the structure is mistaken for quality of the content. Psychological research offers a useful warning. Areview of cognitive offloading describes how people use external tools to reduce the mental demands of a task [8].Offloading can be highly effective. Checklists, templates and automated systems allow us to manage complexity without having to remember every detail. However, reducing cognitive effort is not the same as improving judgement.
Research into automation bias, whereby people become overly reliant on automated decision support, introduces the potential for users to failed to recognise errors introduced by automated systems, with workload, task complexity and time pressure being relevant factors [9].
The direct evidence concerns healthcare decision support rather than medical writing, so applying it to document production represent an informed extrapolation, not a demonstrated outcome of M11. Nevertheless, the underlying question is pertinent: if a document appears complete because every required field has been populated, will reviewers be less inclined to challenge its scientific logic?
A perfectly formatted protocol can still contain a poorly justified endpoint. A consistent eligibility criterion can still exclude the wrong patients. An automated summary can accurately reproduce an analysis while presenting its clinical significance misleadingly. Worse, reusable components can propagate errors efficiently. If an incorrect definition is embedded in an authoritative source, automation may distribute that mistake across every dependent document. Efficiency magnifies the consequences of whatever enters the system.
I don’t raise the point because I believe we should reject automation. Rather, it is to ensure that validation addresses meaning and scientific validity, not merely formatting, completeness and consistency.
So where does the medical writer fit?
This brings us back to the question of voice. A distinctive writing style is not simply a matter of choosing interesting words or varying sentence length. In scientific communication, voice also emerges from judgement: deciding what matters, explaining why it matters, recognising what the evidence cannot establish and making uncertainty intelligible without obscuring it.
A template can organise an argument, but it cannot guarantee that the argument is convincing. A language model can generate plausible explanations, but plausibility is not proof. A structured data field can preserve a numerical result, but it cannot independently determine the appropriate interpretation of that result.
Future medical writers may therefore need to become both better communicators and more sophisticated information specialists. It seems apparent that understanding content models, metadata, provenance, controlled terminology and automated validation will increasingly matter alongside pharmacology, clinical methodology and regulatory requirements. This is not necessarily a loss of professional identity. It could be an opportunity to move beyond repetitive document production towards (I hope) a more influential role in clinical development. The distinction is between assembling information and sharing the understanding regards to what that information means.
The future: fewer writers, or more valuable writers?
Perhaps the most interesting possibility is that medical writing will divide into two increasingly distinct activities. One involving the routine assembly, transformation and checking of structured content. The other scientific interpretation,argument, contextualisation and communication. The first is becoming increasingly automated. The second will remain dependent on expertise, although AI will undoubtedly influence how it is performed.
That distinction also challenges anyone commissioning medical writing. If they purchase documents primarily by pagecount, template compliance or speed of delivery, they will encourage the very commoditisation they fear. In contrast, if they value scientific judgement, critical thinking and the ability to communicate complex evidence to different audiences, they create space for a more valuable professional contribution. The LEGO analogy is therefore useful, but incomplete. Standardised pieces can produce remarkable structures. Yet the instructions cannot anticipate every problem, and thefinished model is not necessarily an adequate representation of the world.
Medical writing may indeed be becoming data formatting. But currently this isn’t where it ends.
The challenge is to automate the assembly without automating away the thinking. If we succeed, structured content should give medical writers more time to exercise their judgement, not less reason to do so. And perhaps that is the real test of the profession's future: not whether we can build a document from a thousand standardised pieces, but whether we can explain how those pieces belong together, what the completed structure tells us, and what it still leaves unanswered.
References
- Sollaci LB, Pereira MG. The introduction, methods, results, and discussion (IMRAD) structure: a fifty-year survey. J Med Libr Assoc. 2004;92(3):364-367.
- Taddio A, et al. Quality of nonstructured and structured abstracts of original research articles in the British Medical Journal, the Canadian Medical Association Journal and the Journal of the American Medical.
- Hopewell S, Chan AW, Collins GS, Hróbjartsson A, Moher D, Schulz KF, et al. CONSORT 2025 statement: updated guideline for reporting randomised trials. Lancet. 2025;405.
- Geary L, et al. Optimizing clinical study report development through lean authoring: a framework for meeting regulatory expectations and preparing for automation. Ther Innov Regul Sci. 2026;60(5):1232-1246.
- International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. ICH M11 guideline: clinical electronic structured harmonised protocol (CeSHarP), clinical study protocol template and technical specifications. Step 5. 20.
- Wilkinson MD, et al. The FAIR guiding principles for scientific data management and stewardship. Sci Data. 2016;3:160018.
- Stocker M, et al. Rethinking the production and publication of machine-readable expressions of research findings. Sci Data. 2025;12(1):677.
- Risko EF, Gilbert SJ. Cognitive offloading. Trends Cogn Sci. 2016;20(9):676-688.
- Goddard K, Roudsari A, Wyatt JC. Automation bias: a systematic review of frequency, effect mediators, and mitigators. J Am Med Inform Assoc. 2012;19(1):121-127.
