Plain language and lay summaries in scientific writing

Last updated: 2026-10-20

Synthesis

Plain-language writing — the practice of communicating with non- expert audiences using familiar vocabulary, short sentences, and clear structure — has migrated from a fringe communication concern to a policy and funding requirement. The Plain Writing Act of 2010 (US Public Law 111-274) [1] mandates plain writing in US federal agency public communication. The EU Clinical Trials Regulation 536/2014 [2] requires lay summaries for every clinical trial conducted in EU member states. The Wellcome Trust, MRC, NIHR, and other major funders require lay summaries on grant applications [3]. Cochrane requires a plain-language summary in every systematic review, written to a standardised template [4, 9]. PLOS Computational Biology asks for a non-technical Author Summary in every research article [5], and PLOS Medicine asks for a bulleted Author Summary when research articles are revised after peer review [12]. The NIH Clear Communication initiative is the dominant US-side infrastructure for plain-language public-facing research communication [6].

The relevant point for scriptorium is that lay-summary writing is a distinct genre with codified expectations, not a “simpler abstract.” Cochrane plain-language summaries follow a standardised template (introduced in 2022) with a “Key messages” section, fixed question-style headings, and plain-language translation of GRADE certainty statements [9]. EU CTR lay summaries follow regulator-specified content checklists. Funders score lay summaries explicitly. A skill that purports to “simplify” a manuscript without respecting these genre conventions produces something that looks like a plain-language summary but is not — it is a stylistically simplified abstract, which is a different artefact.

Whether scriptorium should build a lay-summary-pass skill depends on demand. The skill is technically tractable (LLMs do register-shift comparatively well), and the genre conventions are well-codified. But author-side demand is concentrated in regulated genres (clinical trials, Cochrane reviews, certain funded grant schemes); a general-purpose research paper does not typically need a lay summary at submission. The right call for v0.1/v0.2 is to note the skill as a maybe-later candidate, watch for funder-requirement signals, and design MANUSCRIPT_STATE to accommodate the skill without requiring it.

Evidence and frameworks

Policy mandates

Plain Writing Act of 2010 (US Public Law 111-274). [1] Signed 13 October 2010. Requires federal agencies to use “plain writing” — clear, concise, well-organised, audience-appropriate — in every covered document. Agencies must designate plain-writing leads, maintain plain-writing sections on their websites, and report annually on compliance. The Act applies to federal communication; its indirect effect on US-based researchers is through the agencies that fund them (NIH, NSF, CDC, FDA) and that increasingly require plain-language elements in their applications and outputs.

EU Clinical Trials Regulation 536/2014. [2] Applies to all clinical trials conducted in EU/EEA member states (applicable since 31 January 2022, when the Clinical Trials Information System went live; new trial applications have had to go through CTIS since 31 January 2023; since 31 January 2025, trials approved under the earlier Clinical Trials Directive that are still running must also comply [10]). Sponsors must submit, within one year of trial completion (six months for paediatric trials), a lay summary written so as to be understandable to non-specialists. Content is regulator-prescribed: identification, sponsor, scientific information, population, IMPs, adverse reactions, overall results, interpretation, planned follow-up, where to find more information. Summaries must be translated into the languages of the participating member states. This is the most operationally consequential lay-summary requirement currently in force; non-compliance has regulatory consequences.

NIH Clear Communication and plain-language guidance. [6] The NIH Office of Communications and Public Liaison maintains plain- language guidance for NIH-produced public-facing materials. The guidance is internal-facing but is widely cited as the de facto standard for NIH-funded researcher communication: average 20 words per sentence, one idea per sentence, one topic per paragraph, key message first. National Cancer Institute’s Clear & Simple materials extend the guidance to low-literacy audiences.

Funder requirements. [3] The Wellcome Trust requires a plain- language summary on grant applications (typically 200–300 words). NIHR (UK), MRC (UK), CIHR (Canada), NHMRC (Australia), and the European Research Council have similar requirements with varying length and content specifications. The Wellcome public-engagement funding scheme is a notable extension: lay summaries are not just required, they are evaluated by lay-reviewer panels.

Practitioner standards

Cochrane plain-language summaries. [4, 9, 11] Cochrane has required a PLS for every systematic review for many years. The earlier standards, PLEACS (Plain Language Expectations for Authors of Cochrane Summaries), were replaced in 2022 by new guidance and a template, incorporated into the online Handbook as a supplement to the Reporting chapter [9]. The template’s standardised headings are: plain language summary title; Key messages; a tailored heading (e.g., “What is epilepsy?”); an optional tailored heading; What did we want to find out?; What did we do?; What did we find?; an optional Main results heading; What are the limitations of the evidence?; How up to date is this evidence? [9]. The guidance also describes how to incorporate GRADE statements when reporting the results [9]. The Handbook describes the PLS as a stand-alone summary whose audience “may include people with a health condition, carers, healthcare workers, or policy makers”, some of whom may not have English as a first language [11]. (Length guidance is set in the template itself; the specific word range was not re-verified in the 2026-10 refresh, so it is not quoted here.) The Cochrane standards are the most rigorously documented genre conventions for plain-language summaries; they are the natural reference if scriptorium ever builds a lay-summary skill.

PLOS Author Summaries. [5, 12] PLOS Computational Biology asks all authors of research articles to include “a 150-200 word non-technical summary of the work” immediately after the abstract. The summary should be written in the first person, be distinct from the scientific abstract, and make the findings “accessible to a wide audience that includes both scientists and non-scientists”, placing the work in a broader context and avoiding acronyms and complex terminology [5]. PLOS Medicine uses a different format and timing: its submission guidelines do not ask for an Author Summary, but authors of research articles revised after peer review are asked for one that will immediately follow the Abstract, written as “2-3 single sentence bullet points” under three questions — Why Was This Study Done?, What Did the Researchers Do and Find?, and What Do These Findings Mean? — including the headline numbers and the study’s limitations [12]. Formats differ by journal, so the requirement should be confirmed per venue.

Science-communication research

Bubela et al. (2009), “Science communication reconsidered.” [7] The interdisciplinary consensus paper that consolidated a generation of science-communication thinking. The paper articulates a post-deficit-model view: the public is not an empty vessel into which science can be poured, and communication that treats it that way fails. Effective science communication accounts for the audience’s existing schemas, values, and contexts.

Brossard and Lewenstein (2010). [8] The canonical critique of the deficit model of public understanding of science. Identifies four models: deficit (give them facts), contextual (situate the facts in audience-relevant frames), lay-expertise (recognise that non-specialists have relevant knowledge), and public participation (engage the audience as participants, not recipients). The implication for plain-language writing: a lay summary that simply strips jargon and shortens sentences from a research paper is a deficit-model artifact. A good lay summary translates content into the audience’s frame — addressing their actual questions, explaining the trade-offs, situating the work in their decision context.

Tools and metrics

Readability indices (Flesch Reading Ease, Flesch-Kincaid Grade Level, SMOG, Gunning Fog, Coleman-Liau) are widely used as automated plain-language proxies. They are well-validated for target audiences (originally military training materials and public health pamphlets) but fail systematically for scientific text. The reasons are detailed in quantitative-quality-measures: the formulas inflate difficulty for correctly used technical vocabulary; they are insensitive to genuine cognitive complexity beyond syllables-and-sentence-length. They are useful as outlier flags (“this passage is at grade 18, suspiciously high even for your audience”) but not as quality scores.

Hemingway Editor is the most visible practitioner tool in the plain-language space. Its core heuristic — flag long sentences, passive voice, adverbs, complex words — is a useful triage layer but, like the indices, it is genre-naive and produces false positives on legitimate scientific prose.

How this informs scriptorium

A lay-summary-pass skill, if built, would need to do four things the existing transformation skills do not do:

1. Genre-conform to a target standard (Cochrane PLS, EU CTR, PLOS Author Summary, Wellcome lay summary). The target is a MANUSCRIPT_STATE field, e.g. lay_summary_target: cochrane-pls | eu-ctr | plos-author-summary | wellcome | nih-public | custom.
2. Register-shift without removing technical content the lay reader needs (effect direction, certainty language, what was measured). This is not the same as compression; it is closer to translation. Skopos-theoretic framing in semantic-preservation is the right intellectual model: purpose-relative preservation.
3. Preserve numeric findings with appropriate framing. Effect sizes should be presented in audience-meaningful units (natural frequencies, absolute risk reductions, NNTs where appropriate).
4. Cite, but for non-specialists. A lay summary does not have a formal bibliography, but it must not invent claims and must point to where the technical findings can be verified.

The preservation contract for a lay-summary skill is different from the contract for compression/flow skills. Section-boundary preservation is irrelevant (the lay summary is a separate artefact). Statistic-string preservation needs adaptation rules (“p<0.05” → “the result was statistically significant”; numeric effect sizes preserved with appropriate denominators). Citation preservation becomes “all claims must be supported by the source paper”, not “all citation keys must remain intact.” The skill needs its own preservation report variant.

The risks of getting a lay-summary skill wrong are non-trivial:

• Overclaim drift. Plain-language phrasing tends to drop hedges (“may improve” → “improves”). This is exactly the citation-overreach failure mode (citation-overreach-research) that scriptorium’s avoid_hype constraint is meant to prevent.
• Deficit-model artefact. A skill that mechanically strips jargon and shortens sentences without addressing the audience’s frame produces simplified scientific prose, not a lay summary. Brossard and Lewenstein [8] is the warning.
• Misleading numeric framing. Relative risk reductions sound larger than absolute risk reductions; NNT framings are more honest. A lay-summary skill must default to honest framing.

Implementation priority for scriptorium

Verdict: Maybe later — a dedicated lay-summary-pass skill is plausible but not v0.1.

Reviewed 2026-10-20: none of the flip conditions below has been met. The end of the EU CTR transition period (31 January 2025) [10] widens the set of trials that need lay summaries, but that demand falls on trial sponsors rather than on manuscript authors, so the verdict stands.

Condition that would flip to Yes:

• Funder mandates expand and authors begin asking for explicit lay-summary tooling. Evidence: NIH adds a lay-summary section to the standard application; NSF adopts something similar; UK funders tighten enforcement of existing requirements.
• A Cochrane-affiliated or systematic-review-focused user community surfaces as a scriptorium audience. Cochrane PLS is the most rigorously specified target and the most natural first build.
• Evidence accumulates that LLM-generated lay summaries systematically inflate certainty (per citation-overreach-research reasoning), establishing a clear problem the skill must solve rather than create.

If Yes (anticipated scope):

• Skill name: lay-summary-pass
• Phase: v0.3 (after compression and argumentative-flow mature; depends on Skopos-style purpose-relative preservation being well-established in the codebase).
• Scope: Generate a target-genre-conformant lay summary from the manuscript abstract + key-findings sections. Target genre selected via MANUSCRIPT_STATE field lay_summary_target. Output is a separate document, not an in-place transformation. Preservation report tracks numeric findings, hedging strength, and source-text attribution.
• Required data:
  • MANUSCRIPT_STATE.target_audience (already plausible field).
  • MANUSCRIPT_STATE.lay_summary_target (new).
  • Genre templates (Cochrane PLS, EU CTR, PLOS, Wellcome) as scriptorium-bundled reference resources.
  • Hedging vocabulary (Hyland-style hedge markers; see also semantic-preservation open question on Hyland’s hedging list) for the certainty-preservation check.

Why this is not v0.1: Author demand is concentrated in specific regulated genres, not in general research papers. Building a skill without a clear user community produces theatre. The lay-summary genre is also a place where overclaim drift can cause real harm — inflated certainty in plain language reaches a wider audience than inflated certainty in a technical paper. The skill should not ship before scriptorium’s hedging-preservation and overclaim-detection machinery is mature.

Open questions / weak evidence

• Empirical evidence on whether LLM-generated lay summaries are systematically worse than human ones is thin. Plausible failure modes (overclaim, deficit-model affect, numeric framing errors) are documented in the science-communication literature but not benchmarked against LLM output specifically. (This note is a tier-1 note; the 2026-10 refresh verified identifiers and policy facts but did not sweep the post-2025 literature on LLM-generated lay summaries.)
• The Cochrane PLS template is mature and detailed; the EU CTR template is regulator-defined but less didactically documented. Funder lay-summary norms vary widely; there is no single template.
• Readability indices remain the de facto plain-language metric in many funder contexts despite their well-known failures on scientific text. Scriptorium should not rely on them as quality scores (see quantitative-quality-measures) but may need to report them because funders/reviewers do.

References

1. United States. Plain Writing Act of 2010, Public Law 111-274 (124 Stat. 2861), 13 October 2010. https://www.govinfo.gov/app/details/PLAW-111publ274.
2. Regulation (EU) No 536/2014 of the European Parliament and of the Council of 16 April 2014 on clinical trials on medicinal products for human use. Official Journal of the European Union. 2014;L158:1–76. https://eur-lex.europa.eu/eli/reg/2014/536/oj.
3. Wellcome Trust. Grant conditions for grantholders, including lay-summary expectations. https://wellcome.org/research-funding/guidance/policies-grant-conditions (accessed 2026-05-17).
4. Cochrane. Guidance for writing a Cochrane Plain Language Summary. Cochrane Handbook for Systematic Reviews of Interventions. Chapter III, Section 2 supplementary material. https://training.cochrane.org/handbook/current/chapter-iii-s2-supplementary-material (accessed 2026-05-17).
5. Public Library of Science. Submission Guidelines: PLOS Computational Biology (Author Summary). https://journals.plos.org/ploscompbiol/s/submission-guidelines (accessed 2026-10-20). (Replaces the earlier PLOS Medicine submission-guidelines link. PLOS Medicine’s submission guidelines do not describe an Author Summary; the journal requests one at revision instead, see [12].)
6. National Institutes of Health. Plain Language at NIH (Clear Communication). https://www.nih.gov/institutes-nih/nih-office-director/office-communications-public-liaison/clear-communication/plain-language-nih (accessed 2026-05-17).
7. Bubela T, Nisbet MC, Borchelt R, et al. Science communication reconsidered. Nature Biotechnology. 2009;27(6):514–518. doi:10.1038/nbt0609-514. PMID: 19513051.
8. Brossard D, Lewenstein BV. A critical appraisal of models of public understanding of science: Using practice to inform theory. In: Kahlor LA, Stout PA, eds. Communicating Science: New Agendas in Communication. Routledge; 2010:11–39.
9. Cochrane. ME Support Digest, Issue 3, March 2022: “New guidance and template for writing Cochrane Plain language summaries is now available” (replaces PLEACS; lists the template headings). https://pages.cochrane.org/ME-Support-Digest-March-2022 (accessed 2026-10-20).
10. European Medicines Agency. Clinical Trials Regulation. https://www.ema.europa.eu/en/human-regulatory-overview/research-development/clinical-trials-human-medicines/clinical-trials-regulation (accessed 2026-10-20). (CTIS live from 31 January 2022; mandatory for new applications from 31 January 2023; transition for Directive-approved trials ended 31 January 2025.)
11. Cumpston M, Lasserson T, Flemyng E, Page MJ. Chapter III: Reporting the review [last updated July 2026]. In: Higgins JPT, Thomas J, Chandler J, et al., eds. Cochrane Handbook for Systematic Reviews of Interventions version 6.5. Cochrane; 2026. https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-iii (accessed 2026-10-20).
12. Public Library of Science. Revising Your Manuscript: PLOS Medicine (Author Summary). https://journals.plos.org/plosmedicine/s/revising-your-manuscript (accessed 2026-10-20).