Writer’s Block (2026)

After writing more than 80 scientific and reflective blogs over the last 12 months, writer’s block stops feeling like an occasional inconvenience and starts resembling a feature of everyday life. It is no longer an absence of ideas but friction at the boundary between accumulated expertise and exploration of my own thoughts. The paradox is that the deeper one’s knowledge base becomes, spanning decades of observation in a field that has itself evolved continuously, the more pathways exist for any single article to take shape, and the harder it becomes to choose the most productive narrative to explore [1].

In the last 18 months, another consideration has entered the room: large language models (LLMs) [2]. For many seasoned writers, AI tools appear as the ultimate cheat (and not in a good way), an always-available surrogate for initiation, structure, and even stylistic fluency. Yet this raises a deeper tension: not merely whether AI can solve writer’s block, but whether relying on represents a deliberate surrender of craft, judgement, and intellectual lineage.

Writer’s block as a cognitive and professional constraint

From a cognitive perspective, writing remains a multi-component process involving planning, text generation, and revision processes governed by limited working memory and attentional resources [3]. When your experienced writer faces persistent initiation failure, it is often due to overactivation of evaluative processes: the internalised voice of peer review, mentors, editors, and past standards [1].

Research on academic and professional writers suggests that perfectionism, excessive self-monitoring, and fear of evaluation frequently contribute to writing avoidance behaviours [4][5][6]. In experienced authors, the obstacle is often not a shortage of ideas but an inability to tolerate producing imperfect early drafts while those ideas are still developing. Ironically, expertise itself can become inhibitory: the more you know about how good writing should look, the more difficult it becomes to allow a rough first version to exist.

Studies of academic writers have consistently shown that productivity is not primarily a function of inspiration, but of behavioural regularity and low-stakes engagement with writing [4]. The blocked writer tends to wait for readiness; the productive writer writes irrespective of readiness. Over time, this produces a paradoxical effect: the more one understands writing norms, the greater their potential to inhibit spontaneous output.

Research on creativity reinforces this observation. Idea generation benefits from incubation and reduced cognitive pressure, with meta-analytic evidence demonstrating improved problem-solving after periods of temporary disengagement from a task [7]. As discussed in the PERL guideline, deliberate periods of reflection can often be more productive than prolonged attempts to force progress [8].

Contemporary neuroscience provides a plausible explanation. Creative insight appears to emerge through interactions between executive control systems and the brain’s default mode network, allowing associative processing and novel connections to develop outside conscious effort [9][10]. Some of our most productive thinking therefore occurs when we temporarily stop trying to think about the problem directly.

Writer’s block, therefore, is rarely an absence of cognition. More often, it reflects interference between competing cognitive modes: generation versus evaluation, exploration versus judgement, creativity versus quality control.

The lure of artificial fluency

Large language models enter the equation as a highly unusual cognitive artefact: they collapse the mental burden of initiation. A blank page becomes a filled page in seconds. The familiar struggle of “just getting something down” evaporates.

From a purely functional perspective, this addresses what cognitive psychology recognises as one of the most difficult phases of writing: initiation. With apparent ease, LLMs generate provisional drafts that give body to thoughts which can then be managed externally and revised incrementally.

AI may also function as a form of cognitive off-loading. By externalising initial structure and language, it reduces immediate working-memory demands and allows attention to be redirected towards evaluation and refinement [11]. Whether this ultimately improves thinking or merely disguises unfinished concept resolution remains an open question.

However, this apparent solution introduces a second-order problem: displacement of the author’s engagement. The writer is no longer struggling to express thought but deciding whether thought is required at all.

The “ultimate surrender” framing

For a seasoned writer, particularly one who has built an intellectual identity through decades of disciplinary engagement, the temptation of reliance on AI can feel like a form of surrender. Not merely to convenience, but to a different epistemic regime: one in which language is produced without corresponding lived cognitive effort.

This concern is understandable. Most professional writers are acutely aware of their intellectual journeys, involving teachers, supervisors, mentors, reviewers, editors, and colleagues whose guidance shaped their development. To outsource substantial portions of writing to AI can feel uncomfortably close to bypassing the apprenticeship through which expertise was acquired.

There is also a professional anxiety. If writing becomes cheap and instantaneous, the perceived value of human authorship itself shifts from production towards curation. Take that thought to its obvious end and professional writing becomes increasingly commoditised, possibly unnecessary. Whether that future fully materialises remains uncertain, but the concern is not irrational.

AI is not a solution without epistemic cost

That LLMs are not truth-preserving systems remains an important consideration. They generate statistically coherent language patterns rather than genuine understanding. As highlighted in foundational critiques, these systems are perhaps best viewed as “stochastic parrots”: models capable of recombining linguistic patterns without possessing semantic comprehension [12]. This creates a fundamental risk: fluency without accountability [13].

Empirical and conceptual analyses of AI-assisted writing in academia have highlighted persistent concerns regarding hallucinated content, fabricated references, embedded bias, and subtle semantic distortions, particularly when users over-trust generated outputs [14][15]. Within scientific and medical writing, these issues are not cosmetic; they threaten the integrity of the evidential record itself.

You better be on your toes when AI is used as your writing assistant. Every generated sentence should be treated as provisional until independently verified. Citations require checking. Claims require tracing back to primary sources. Logical consistency must often be reconstructed manually. Medical writers are used to providing highlighted materials with their submissions to clients. The need to identify exactly where a model obtained, or appears to have obtained, its information, read and understand it, can reverse any promised efficiency gain. AI may accelerate drafting, but it frequently increases verification workload [16].

The paradox of assisted authorship

Writing externalises thought; AI externalises writing. When both processes are layered together, the writer risks losing visibility of and engagement with their own reasoning pathway. This becomes particularly problematic in scientific and professional domains where an argument integrity is at least as important as linguistic expression.

The result is a new form of cognitive debt. Drafts become easier to produce but harder to fully own. The writer increasingly becomes editor of machine-generated structure rather than originator of conceptual architecture. Box tickers!

This concern aligns with broader findings in human–AI interaction research. Automation bias describes the tendency to accept machine outputs even when they are incorrect [15]. Human factors research has also demonstrated that prolonged reliance on automation can contribute to skill degradation, complacency, and reduced engagement with critical decision-making processes [17][18]. Habitual dependence on AI-assisted drafting may therefore weaken some of the cognitive habits that expert writing traditionally develops.

When desperation meets deadlines

Despite these concerns, we live in a pragmatic reality: deadlines exist. Journals, blogs, policy outputs, client deliverables, and professional commitments do not pause for cognitive fatigue or creative blockage.

In circumstances where time pressure is acute and output is genuinely unavoidable, AI can serve a legitimate purpose. I would view it primarily as a structural assistant rather than an author: generating outlines, suggesting organisational alternatives, summarising source material, or providing prompts when semantic inertia becomes overwhelming.

To be fair, versions of this workflow are already widespread across our industry (and others). The technological genie is unlikely to return to the bottle.

However, current LLMs remain insufficiently reliable to function as autonomous writing systems in professional scientific contexts. Nothing should be accepted without verification against domain knowledge, primary literature, and logical consistency. The model’s output is not evidence; it is suggestion.

A defensible workflow may therefore resemble the following:

• Human thought generates scope and conceptual structure (the prompt).
• AI generates provisional language and organisational options.
• Human authors verify sources and factual accuracy.
• Human authors refine meaning and assume responsibility for the final text [16].

Returning to first principles

Even with AI available, the original problem of writer’s block does not disappear; it relocates. The challenge is no longer simply “How do I write the first sentence?” but “How do I ensure that the first sentence accurately reflects my thinking and the underlying evidence?”

This observation ultimately returns us to established writing science. Boice’s work still holds: frequent, low-pressure writing sessions remain among the most reliable antidotes to blockage [4]. Incubation still matters: temporarily stepping away from a project often improves subsequent thinking [7][9]. Working-memory limitations and evaluative load continue to influence initiation and productivity [3].

AI does not remove these constraints; it simply gives them a different form.

Conclusion

The temptation to view AI as the perfect solution to writer’s block is understandable, particularly for writers facing sustained production demands, deadline pressure, or cognitive fatigue. For some, it may even appear to offer an escape from the uncertainty and frustration that often accompany the writing process.

Yet that framing remains incomplete. AI does not eliminate writer’s block; it redistributes its costs, from initiation to verification, from creation to curation, and from generation to oversight. Used indiscriminately, it risks reducing the writer’s role to verification and curation rather than original conceptual development.

Used carefully, however, AI can function as a contingency tool: valuable for overcoming inertia, exploring alternative structures, and reducing some of the mechanical burdens of drafting. Its value lies not in replacing cognition but in supporting it.

The seasoned writer’s dilemma is therefore not whether to reject or embrace AI outright. Rather, it is how to exploit its advantages without allowing convenience to replace thought. The blank page still matters, not because it is empty, but because it is the last place where thought is undeniably your own.

References

1. Hardman TC (2012). The Blank Page Problem: How Experienced Medical Writers Beat Writer's Block.
2. Hardman TC (2026). Medical Writing 2026: Adapting to AI and Rising Complexity.
3. Hayes, J. R., & Flower, L. (1980). Identifying the Organization of Writing Processes. In L. W. Gregg, & E. R. Steinberg (Eds.), Cognitive Processes in Writing: An Interdisciplinary Approach (pp. 3-30). Hillsdale, NJ: Lawrence Erlbaum.
4. Boice R. Professors as Writers: A Self-Help Guide to Productive Writing. New Forums Press; 1990.
5. Kellogg RT. The Psychology of Writing. New York: Oxford University Press; 1994.
6. Rose M. Writer’s Block: The Cognitive Dimension. Carbondale (IL): Southern Illinois University Press; 1984.
7. Sio UN, Ormerod TC. Does incubation enhance problem solving? A meta-analytic review. Psychological Bulletin. 2009;135(1):94–120.
8. Hardman TC. (2020). You will be judged.
9. Beaty RE, Benedek M, Silvia PJ, Schacter DL. Creative cognition and brain network dynamics. Trends Cogn Sci. 2016;20(2):87–95.
10. Beaty RE, Silvia PJ, Nusbaum EC, Jauk E, Benedek M. The roles of associative and executive processes in creative cognition. Mem Cognit. 2014;42(7):1186–1197.
11. Risko EF, Gilbert SJ. Cognitive offloading. Trends Cogn Sci. 2016;20(9):676–688.
12. Bender EM, Gebru T, McMillan-Major A, Shmitchell S. On the dangers of stochastic parrots. FAccT 2021.
13. Hardman TC (2025). Medical Writing vs Artificial Intelligence: Threat, Tool, or False Debate?
14. Hardman TC. (2026). From Hand-Written Bibliographies to AI-Hallucinated Citations
15. Dwivedi YK, et al. So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI. International Journal of Information Management. 2023;71:102642.
16. Hardman TC. (2026). AI: Solving a Problem That Doesn’t Exist.
17. Parasuraman R, Riley V. Humans and automation: Use, misuse, disuse, abuse. Hum Factors. 1997;39(2):230–253.
18. Mosier KL, Skitka LJ. Human decision makers and automated decision aids: Made for each other? In: Parasuraman R, Mouloua M, editors. Automation and Human Performance. Mahwah (NJ): Lawrence Erlbaum Associates; 1996. p. 201–220.