CRediT, ICMJE, and the authorship landscape

Last updated: 2026-10-20

Synthesis

Authorship in modern science is increasingly contested, increasingly formalised, and increasingly machine-readable. Two complementary artefacts dominate the landscape: ICMJE’s four authorship criteria (substantial contribution, drafting, approval, accountability), which define who can be an author; and the Contributor Roles Taxonomy (CRediT), which describes what each contributor actually did, mapping contributions onto 14 standardised roles. The empirical literature on ghost and gift authorship shows the older “author list” model has been quietly failing for decades: roughly one in five high-impact biomedical papers in 2008 reported either honorary or ghost authorship [3]. A 2024 meta-analysis of 19 health-science surveys puts the figure higher when contributions are checked against ICMJE criteria rather than self-assessed: 51% of publications had honorary authorship on that measure [8]. The combination of CRediT metadata plus explicit ICMJE compliance is the field’s response.

The 2023 ICMJE update making LLMs ineligible for authorship [5] is relevant for scriptorium beyond the obvious “don’t list ChatGPT as a co-author” — it implies that any AI-assisted scholarly tool must be disclosable, attributable, and ultimately reducible to a human’s accountability. The same May 2023 update asked authors to disclose AI use in the cover letter and in the submitted work [5]. A January 2024 amendment added that authors who used AI to conduct the study should give the tool, version and prompts in the Methods [10]. The January 2026 update did not introduce these rules; it consolidated ICMJE’s AI guidance into a dedicated top-level Section V covering authors, reviewers and editors [5]. The current text also warns that nondisclosure of AI use “may be construed as misconduct in some circumstances” [5]; when that sentence was first added could not be established. Scriptorium’s design is consistent with this: skills are explicitly invoked, outputs are reviewed by humans, and the underlying manuscript remains a human-accountable artefact.

For the build itself: this is a schema-level addition, not a new skill. MANUSCRIPT_STATE.yaml should grow a contributors: field aligned to CRediT roles. Several v0.x skills can then consume it without a dedicated authorship skill being justified. (Superseded in practice. Scriptorium shipped a dedicated author-contribution-audit skill that reads the manuscript’s Author Contributions prose, and the schema deliberately omits contributors:. See Implementation priority.)

Evidence and frameworks

CRediT: the 14 roles

The Contributor Roles Taxonomy was developed through a workshop process beginning in 2012, formalised by Brand et al. in 2015 [1], and adopted by CASRAI and (later) NISO. It became an ANSI/NISO standard, Z39.104-2022, approved by ANSI on 14 January 2022 [4]. CRediT is widely referenced in publisher submission workflows (the earlier “more than 150 publishers” figure could not be verified and has been dropped). In 2026 Crossref added CRediT support to its metadata schema (Schema 5.5), so roles can now travel with the DOI record [7].

The 14 roles are:

1. Conceptualization
2. Data curation
3. Formal analysis
4. Funding acquisition
5. Investigation
6. Methodology
7. Project administration
8. Resources
9. Software
10. Supervision
11. Validation
12. Visualization
13. Writing — original draft
14. Writing — review & editing

Each contributor can hold multiple roles, and roles can be marked “lead”, “equal”, or “supporting”. The point is to disaggregate “author” into “what did this person actually contribute”, which is where the older author-list model breaks down for large collaborations.

The 2015 Brand et al. paper [1] frames CRediT as “moving beyond authorship” toward an “attribution” model — useful background for why the taxonomy looks the way it does. Allen, O’Connell and Kiermer [6] describe how publishers began implementing it in the shift from authorship to contributorship.

The role list itself has not changed. The CRediT Standing Committee says that “current resource constraints make further revisions of CRediT a challenge at this time”. Meanwhile it is publishing illustrative examples of the roles and prioritising implementation of the existing version in scholarly metadata schemas such as Crossref’s [4, 7]. Crossref’s Schema 5.5 allows multiple roles per contributor and a flag for the corresponding author. Before 5.5, a contributor could carry only one Crossref contributor role [7].

ICMJE: the four criteria

The International Committee of Medical Journal Editors specifies four conditions, all of which must be met, for someone to be listed as an author:

1. Substantial contributions to the conception or design of the work, or the acquisition, analysis, or interpretation of data.
2. Drafting the work or reviewing it critically for important intellectual content.
3. Final approval of the version to be published.
4. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Contributors who do not meet all four criteria should be listed in the acknowledgements with their specific contributions described. The current ICMJE text gives, as “examples of activities that alone (without other contributions) do not qualify a contributor for authorship”, “acquisition of funding; general supervision of a research group or general administrative support; and writing assistance, technical editing, language editing, and proofreading” [2]. These are a common source of dispute. (Corrected in the 2026-10 refresh: the earlier version listed “data acquisition” here. ICMJE’s list does not, because acquisition of data counts toward criterion 1. Data collection alone still does not make someone an author: ICMJE’s own examples of acknowledgement wording include contributors who “collected data”, since they do not meet the other three criteria [2].)

Ghost, gift, and honorary authorship

Wislar et al.’s 2011 BMJ cross-sectional survey [3] is the canonical empirical study. Across six leading general medical journals in 2008:

• Honorary authorship (named author who did not meet ICMJE criteria): 17.6%.
• Ghost authorship (substantive contributor not named as author): 7.9%.
• Either or both: 21.0%.

This was an improvement over the 1996 baseline (29.2%), but the absolute rate remains substantial. The literature distinguishes:

• Gift authorship — author added as a favour (mentor, senior colleague, collaborator on a different project).
• Honorary authorship — author added in deference to position or to confer credibility (department chair, funding-source representative).
• Ghost authorship — substantive contributor (often a professional medical writer, industry employee, or student) omitted from the byline.

Meursinge Reynders et al.’s 2024 systematic review and meta-analysis [8] pooled 19 surveys in the health sciences, all rated as low methodological quality. Prevalence depended on how researchers were asked:

• 26% perceived honorary co-authors when given no authorship criteria.
• 18% did so when referred to the ICMJE criteria.
• 51% of publications had honorary authorship when researchers declared their co-authors’ contributions and these were then compared with ICMJE criteria.

In the same surveys, 10% of researchers reported being approached to add an honorary author, and 16% admitted adding one [8]. The gap between 18% and 51% is the operational point (researcher inference). Authors under-recognise honorary authorship until their contributions are written down and checked against the criteria. That is the check a contributions audit performs.

The CRediT taxonomy is partly a response: explicit role assignment makes it harder to add an honorary author with no plausible contribution, and harder to omit a ghost author who did the heavy lifting.

Authorship order by field

Order conventions vary substantially across disciplines (see discipline-conventions):

• Biomedicine and life sciences: first author = lead contributor (often graduate student/postdoc); last author = senior responsible PI / corresponding author. Middle authors ordered approximately by contribution. Equal-contribution markers (†, *, “these authors contributed equally”) are common for first-author shared positions.
• Mathematics, theoretical physics, economics: alphabetical by surname. Deviation from alphabetical is sometimes meaningful and sometimes a journal style.
• Computer science / ML: contribution-based, no consistent field convention; lead author is typically first, advisor often last but not always.
• Particle physics (“hyperauthorship”): hundreds-to-thousands of authors listed alphabetically by collaboration convention; byline is unreadable as a contribution signal.

A scriptorium audit that assumes “last author = senior PI” silently mislabels mathematics and economics submissions.

AI and authorship policy

ICMJE’s 2023 update [5] is explicit: chatbots and LLMs cannot be listed as authors because they cannot satisfy the accountability criterion. Nature, Science, JAMA, and the Cell Press journals have issued aligned statements; the consensus is robust. The January 2026 ICMJE update moves the AI guidance into a new top-level Section V, “Use of Artificial Intelligence in Publishing”, with subsections for authors, reviewers and editors [5]. Most of the disclosure rules below predate it; the date each first appeared is given where it could be established. Required disclosure now reads:

• AI used for writing, editing, proofreading → describe in acknowledgements.
• AI used for data collection, analysis, or figure generation → describe in methods. Since a January 2024 amendment, authors who used AI to conduct the study should give enough detail to allow replication, “including the tool used, version, and prompts where applicable” [10]. The January 2026 Section V keeps this wording [5].
• Cover letter must mention AI use, in addition to the disclosure in the manuscript itself. This dates from the May 2023 update [5].
• AI as author or source. Authors should not list AI as an author or co-author, nor cite AI as an author. “Referencing AI-generated material as the primary source is not acceptable” [5].
• Consequences. “Nondisclosure of AI use may require corrective action and may be construed as misconduct in some circumstances” [5]. This sentence is in the current Section V text; the date it was first added could not be verified, because icmje.org blocked automated fetching.
• Peer-reviewer confidentiality: reviewers “must adhere to the journal’s stated policy on the use of AI or request permission from the journal” before using AI in a review. Confidentiality “may prohibit the uploading of the manuscript to software or other AI technologies where confidentiality cannot be assured unless such use of AI is explicitly permitted by journals” [5]. (Reworded in the 2026-10 refresh. The earlier text said uploading was acceptable if the LLM was “contractually non-retaining”, which is not ICMJE’s test. The test is the journal’s policy and its explicit permission.)

This last constraint is operationally relevant for reviewer-simulation: a user running the skill on their own unsubmitted manuscript is fine; a user running it on a manuscript they are peer-reviewing is not.

How this informs scriptorium

1. Add contributors: to MANUSCRIPT_STATE.yaml. Structure: a list of {name, orcid?, roles: [CRediT role string]} objects. This unblocks several downstream skills without requiring a new authorship-specific skill in v0.1. (Not adopted. See Implementation priority. If it is revisited, align it with Crossref Schema 5.5, which already models multiple CRediT roles and a corresponding-author flag [7].)

2. Skills that consume contributors:

   • reviewer-simulation can flag authorship-policy issues (missing contribution statement, last author shown as non-contributor, AI listed as author).
   • A future submission-readiness audit can check ICMJE compliance pre-submission.
   • revision-summary can attribute changes to specific contributors when journals require it.

3. AI-disclosure language as a skill output. Scriptorium itself is an AI-assisted tool. Outputs of transformation skills should be invoke-traceable enough that users can write honest acknowledgements (“Scriptorium’s argumentative-flow skill was used to review the discussion section; all edits were reviewed by the authors”). This is a MANUSCRIPT_STATE /skill-log discipline, not a new skill. Under the ICMJE text (cover-letter disclosure since May 2023 [5]) the same record should also support a cover-letter disclosure. Where AI was used to conduct the study, as opposed to writing assistance, the Methods should name the tool, version and prompts (since January 2024 [10]; retained in the January 2026 Section V [5]).

4. Discipline-specific authorship-order norms. A scope warning in the schema documentation: contribution-order interpretation is biomedical-default; in math/economics, alphabetical order is not informative.

Implementation priority for scriptorium

Verdict: Yes (v0.1 schema augmentation) and No (no dedicated authorship skill).

Reconsidered 2026-10-20 — superseded by what shipped. Scriptorium now has a dedicated author-contribution-audit skill. It audits the manuscript’s Author Contributions section against the ICMJE criteria and CRediT roles, and it deliberately does not add contributors: to MANUSCRIPT_STATE.yaml. Declared prose stays in the manuscript. The current verdict is therefore Yes — dedicated skill (shipped); No — schema field. The original reasoning below is kept for the record. The refreshed evidence supports the dedicated skill. Self-assessed honorary authorship (18%) understates the criteria-checked rate (51%) [8], which is exactly the gap a contributions audit closes. Three changes are pending for the skills pass:

• Non-author activities. The skill’s list of activities that alone do not justify authorship should match the current ICMJE list (funding acquisition; general supervision or administrative support; writing assistance, technical and language editing, proofreading), which does not name data acquisition. A contributor who only collected data still fails criteria 2–4 [2].
• AI disclosure. The skill’s AI checks should cover the ICMJE AI rules as now consolidated in Section V (January 2026) [5]: the cover-letter plus in-manuscript disclosure (May 2023 [5]); tool, version and prompts in the Methods (January 2024 [10]); no AI cited as an author or as a primary source; and the misconduct warning for nondisclosure (in the current text; date added not verified). The skill should present these as established ICMJE requirements, not as new in 2026.
• CRediT metadata. It may note that CRediT roles can now be deposited with Crossref metadata (Schema 5.5) [7].

Yes — schema: Add contributors: to MANUSCRIPT_STATE.yaml in v0.1 or v0.2. Minimal effort, high downstream leverage. Required data: a list of {name, orcid?, roles: [string]} items; roles drawn from the 14 CRediT categories; optional equal_contribution and corresponding booleans. Document the field in the schema, with a note that biomedical authorship-order conventions inform roles but other disciplines may use the field differently.

No — dedicated skill: A standalone authorship-audit skill is not justified in v0.1–v0.3. The cost of building it is disproportionate to how often a user will invoke it. Embed ICMJE-compliance checks in reviewer-simulation and (future) submission-readiness instead.

If Maybe later: if a non-trivial fraction of users start submitting to high-stakes biomedical venues where authorship disputes are a known failure mode, promote to a authorship-and-contribution-audit skill. Trigger: a user reports that an authorship issue was caught at submission that scriptorium could have flagged.

Open questions / weak evidence

• CRediT adoption is uneven: many adopting publishers require role metadata at submission but don’t publish it visibly, so its audit value depends on whether downstream readers can see it. A 2025 scoping review of 45 papers found that CRediT promotes transparency. It also found barriers: limited applicability across research types, ethical concerns, conflicts among contributors, low awareness, inconsistent implementation and ambiguous role definitions [9]. Crossref Schema 5.5 [7] addresses visibility, but uptake by publishers is not yet measured.
• The Wislar et al. data are 15+ years old. More recent surveys exist but are heterogeneous. The 2024 meta-analysis [8] pools them but rates every included survey as low methodological quality, so the field-wide trend is still unclear.
• AI authorship policy is evolving fast. The ICMJE position is robust, but specifics (do you need to disclose grammar checkers? citation managers? Paperpal?) are inconsistent across journals. The January 2026 ICMJE text [5] sets the floor for ICMJE-following journals, but it does not settle the grammar-checker question.
• The dating of individual ICMJE AI rules rests on secondary and search-index copies, because icmje.org blocked automated fetching: the May 2023 cover-letter rule on search-index copies of ICMJE’s May 2023 update, the January 2024 tool/version/prompts sentence on a dated journal blog post [10]. When the misconduct sentence was added is unknown.
• The claim that Nature, Science, JAMA and Cell Press issued aligned AI-authorship statements was not re-checked journal by journal in the 2026-10 refresh.
• ORCID integration with CRediT is improving but is not yet universal; some publishers require ORCIDs for all authors, others only the corresponding author.

References

[1] Brand, A., Allen, L., Altman, M., Hlava, M., & Scott, J. (2015). Beyond authorship: attribution, contribution, collaboration, and credit. Learned Publishing, 28(2), 151–155. DOI: 10.1087/20150211.

[2] International Committee of Medical Journal Editors. Defining the Role of Authors and Contributors. https://www.icmje.org/recommendations/browse/roles-and-responsibilities/defining-the-role-of-authors-and-contributors.html.

[3] Wislar, J. S., Flanagin, A., Fontanarosa, P. B., & DeAngelis, C. D. (2011). Honorary and ghost authorship in high impact biomedical journals: a cross sectional survey. BMJ, 343, d6128. DOI: 10.1136/bmj.d6128. PMID: 22028479.

[4] CRediT — Contributor Roles Taxonomy. https://credit.niso.org/. (Hosted by NISO since 2022.) Standard: ANSI/NISO Z39.104-2022, approved by ANSI on 14 January 2022 (https://www.niso.org/standards-committees/credit). Standing Committee updates: https://credit.niso.org/category/updates/.

[5] International Committee of Medical Journal Editors. (2023). ICMJE Recommendations, updated May 2023. AI / chatbot authorship guidance. https://www.icmje.org/news-and-editorials/updated_recommendations_may2023.html and https://www.icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-authors.html. Superseded in part by the January 2026 update, which adds Section V, “Use of Artificial Intelligence in Publishing” (A. Use of AI by authors; B. Use of AI by reviewers; C. Editors’ role): https://www.icmje.org/news-and-editorials/updated_recommendations_jan2026.html and https://www.icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-reviewers.html. (The icmje.org pages blocked automated fetching during the 2026-10 refresh. The quotations were checked against search-index copies of the January 2026 Recommendations, including ICMJE’s annotated PDF. Corrected in the 2026-10 audit: the earlier wording presented the cover-letter rule and the tool/version/prompts sentence as new in January 2026. They date from May 2023 and January 2024 [10] respectively; January 2026 consolidated them into Section V.)

[6] Allen, L., O’Connell, A., & Kiermer, V. (2019). How can we ensure visibility and diversity in research contributions? How the Contributor Role Taxonomy (CRediT) is helping the shift from authorship to contributorship. Learned Publishing, 32(1), 71–74. DOI: 10.1002/leap.1210.

[7] Crossref. (2026). Schema 5.5 now available: adding CRediT, new record types for blogs and posters, and more. Crossref blog. https://www.crossref.org/blog/schema-5.5-now-available-adding-credit-new-record-types-for-blogs-and-posters-and-more/. See also the CRediT site post “Crossref Schema 5.5 now available: includes support for CRediT”: https://credit.niso.org/updates/crossref-schema-5-5-now-available-includes-support-for-credit/.

[8] Meursinge Reynders, R. A., ter Riet, G., Di Girolamo, N., Cavagnetto, D., & Malički, M. (2024). Honorary authorship is highly prevalent in health sciences: systematic review and meta-analysis of surveys. Scientific Reports, 14(1), 4385. DOI: 10.1038/s41598-024-54909-w. PMID: 38388672.

[9] Godskesen, T., Helgesson, G., & Eriksson, S. (2026). Implementation, barriers, and improvement strategies for CRediT: a scoping review. Accountability in Research, 33(4), 2528953. DOI: 10.1080/08989621.2025.2528953. PMID: 40658084. (Published online 14 July 2025.)

[10] Ochsner Journal. (2024, 25 January). Additional AI recommendations from the International Committee of Medical Journal Editors. Ochsner Journal blog. https://ochsnerjournal.wordpress.com/2024/01/25/additional-ai-recommendations-from-the-international-committee-of-medical-journal-editors/. (Dated secondary record of the January 2024 ICMJE amendment, quoting it verbatim: “Authors who used AI technology to conduct the study should describe its use in the methods section in sufficient detail to enable replication of the approach, including the tool used, version, and prompts where applicable.” The same post records the January 2024 addition “Reviewers must request permission from the journal prior to using AI technology to facilitate their review.” Cited because icmje.org blocked automated fetching.)