CRediT’s 14 contributor roles describe what a person did on a research output. They were not designed with AI tools in mind, and — as of 2026 — no publisher, journal, or the NISO CRediT Standing Committee itself has added a 15th role for AI. So when a contributor used an AI tool as part of doing their own work — running an AI-assisted analysis under Formal analysis, generating draft code under Software, tightening prose under Writing – review & editing — how should that be recorded? Not as a new role for the AI. This guide covers the practical convention for noting AI assistance within a human contributor’s existing role attribution, role by role.
This is a narrower question than two others worth distinguishing up front:
- Can AI itself be an author or hold a CRediT role? No, under every major publisher and editorial-body position checked for this guide — see the AI as author and AI co-authorship rejection (ICMJE 2023) Dictionary entries for that question specifically.
- How do I write a general AI-use disclosure statement? That’s covered on CASRAI’s AI disclosure guidance for authors — the wording, required elements, and journal-by-journal placement of the standalone declaration.
This guide sits between those two: given that AI can’t hold a role, and given that a separate disclosure statement exists, what — if anything — changes about how a human fills in their own CRediT role when AI was part of doing it?
The one fixed rule: AI does not hold a CRediT role
Every publisher policy checked for this guide agrees on this point, even though their AI-disclosure mechanics differ. Elsevier’s author instructions state plainly: “Authors should not list AI Tools as an author or co-author, nor cite AI Tools as an author,” because “authorship implies responsibilities and tasks that can only be attributed to and performed by humans.” ICMJE’s current recommendations are equally direct: AI-assisted tools “should not be listed as authors” because they cannot take responsibility for the accuracy, integrity, or originality of the work. This is the same accountability logic behind ICMJE’s four-part authorship criteria generally — a CRediT role, like authorship itself, carries an implicit claim of responsibility that a tool cannot make.
Practically: whatever AI assistance went into a piece of work, the CRediT role for that work is still assigned to the human who directed the tool, reviewed its output, and takes responsibility for it — never to the tool itself, and never split out as a separate “AI” line in the contributor statement.
Two disclosures, not one merged statement
The publishers with the clearest AI policies keep the CRediT statement and the AI-use disclosure as two separate, adjacent statements rather than merging them into one:
- Elsevier requires a standalone “Declaration of Generative AI and AI-assisted technologies in the writing process,” inserted at the end of the manuscript, immediately above the references — a different location and a different required wording (“During the preparation of this work the author(s) used [tool] in order to [reason]”) than the CRediT contributor-role block, which stays a plain per-author role list.
- PLOS requires AI contributions to be “clearly reported in a dedicated section of the Methods, or in the Acknowledgements section for article types lacking a Methods section” — again, structurally separate from the author-contributions statement.
- ICMJE requires authors to disclose AI use “in both the cover letter and the submitted work in the appropriate section if applicable,” with routine language polishing typically landing in the acknowledgments.
None of these mechanisms modify the CRediT statement itself. That’s most of why “does the AI get a CRediT role” is a recurring point of confusion: the two disclosures usually sit close together in the manuscript, but they’re answering different questions — CRediT says who is accountable for what; the AI declaration says what tool did what, under whose direction. See Generative-AI disclosure statement and AI tool disclosure for the Dictionary-level definitions of the second statement.
So what does “recording AI assistance per role” actually mean?
Since the CRediT taxonomy itself has no field for “tool used” and no publisher’s structured CRediT-collection form (Editorial Manager’s role picker, for example) has an AI-specific input, “AI contribution disclosure in CRediT statements” in practice means something narrower than editing the taxonomy: connecting the two existing statements so a reader can tell which role’s work involved AI assistance, without inventing a new role or blending the two statements into one.
The clearest real-world model for making that connection is APA’s generative-AI disclosure guidance, which — while it doesn’t use CRediT’s own role labels — explicitly draws the analogy for at least one role. APA’s guidance states that AI-assisted copyediting or translation should be disclosed “similar to the Writing – Review and Editing CRediT role, only generative AI may not be listed as an author.” The same guidance table extends the logic, implicitly, to other roles:
| Generative-AI use | Nearest CRediT role | Where APA says to disclose it |
|---|---|---|
| Create or refine code | Software | Method section: tool used and number of iterations |
| Analyze, refine, format, or visualize data | Formal analysis / Visualization | Method section: tool used and number of iterations |
| Generate a resource list to inform a literature review | Investigation | Introduction: prompts, keywords, and tool used |
| Translate text or implement extensive copyediting | Writing – review & editing | Author note: tool used (“similar to the Writing – Review and Editing CRediT role, only generative AI may not be listed as an author”) |
| Draft manuscript content, literature summaries, or outlines | Writing – original draft | Introduction: prompts and tool used |
| Create or edit figures, tables, or images | Visualization | The section where the item appears (e.g. Results) |
Treat this as a practical mapping to apply consistently, not an official CRediT extension — CASRAI’s own credit.niso.org reference carries no AI-specific guidance as of this writing, and no publisher’s structured CRediT form has an AI field. The mapping is useful precisely because it lets an author’s AI-use declaration point back at a specific role line (“the code generation noted under Software used…”) instead of reading as a vague, disconnected paragraph.
Role-by-role notes
Conceptualization
AI tools are sometimes used to brainstorm research questions or framing. The role still belongs to whoever evaluated, selected, and takes ownership of the resulting idea — disclose the tool’s use in the AI-use statement (typically the Introduction, per APA’s “resource list” mapping above), not as a change to who holds Conceptualization.
Investigation
Where AI assisted a literature search or evidence-gathering process (see AI in literature search), disclose the tool, prompts, and keywords used — APA recommends preserving the actual output in case of a later inquiry. The human who ran and verified the search still holds Investigation.
Formal analysis
AI-assisted statistical or qualitative analysis (see AI in qualitative coding) is one of the clearer analogues in APA’s table: disclose the tool and the number of iterations in the Methods section, the same way a statistical package like R or SPSS is named. The contributor who ran, checked, and takes responsibility for the analysis holds Formal analysis; the tool is not a co-analyst.
Software
AI-assisted code generation is probably the single most common case in practice. APA’s guidance is explicit: disclose the tool and number of iterations in the Methods section. The Software role stays with the person who directed the generation, reviewed the code, and is accountable for it running correctly — not the tool that produced a first draft of it.
Data curation
Where AI assisted with cleaning, tagging, or organizing a dataset, treat it the same way as Formal analysis above: name the tool and describe what it did in the Methods or data-availability materials, and keep the role attributed to the human curator responsible for the dataset’s integrity.
Visualization
AI-generated or AI-edited figures, plots, or images (see AI image generation) should be disclosed in the section where the item appears — typically Results — including the tool and prompt used. Several journals (PLOS among them) treat AI-generated images as needing the same scrutiny as any other result, given the risk of fabricated or misleading visual content; disclosure here is not optional filler.
Writing – original draft
AI-assisted drafting (see AI-assisted writing) is disclosed in the AI-use statement, generally noting the tool and how it was used (outline generation, drafting sections, etc.). The author who reviewed, edited, and takes responsibility for the final text holds Writing – original draft.
Writing – review & editing
This is the one role APA names explicitly: AI-assisted copyediting or translation (see AI translation and AI summarisation) is “similar to the Writing – Review and Editing CRediT role, only generative AI may not be listed as an author.” Basic grammar/spelling checks are generally treated as not requiring disclosure at all by most publishers (Elsevier explicitly exempts them); more extensive AI-assisted editing should be named.
Funding acquisition, Project administration, Supervision, Resources, Methodology, Validation
AI assistance is less commonly reported for these roles, and none of the publisher policies reviewed for this guide give them a specific disclosure pattern. If an AI tool was used incidentally within one of these roles — drafting part of a grant narrative, for example — disclose it under the AI-use statement and, if it more specifically involved drafting text, consider whether it’s better described under Writing – original draft instead of stretching one of these six roles to cover it. Don’t invent a disclosure convention these roles don’t have real guidance for.
A sample two-part disclosure
The following is an illustrative example only — a format template, not a real manuscript, no institution or author implied to be real:
Author contributions (CRediT):
Author A: Conceptualization, Writing – original draft.
Author B: Formal analysis, Software, Writing – review & editing.Declaration of generative AI and AI-assisted technologies:
During the preparation of this work, Author B used [Tool Name] to assist with generating an initial draft of the data-processing scripts described under Software, and to assist with copyediting the manuscript draft under Writing – review & editing. All AI-generated output was reviewed, tested, and edited by Author B, who takes full responsibility for the accuracy and integrity of the final content.
Note what the sample does and doesn’t do: it never lists the tool as a contributor or gives it a role; it keeps the two statements textually separate; and it deliberately cross-references the specific role lines the AI assistance relates to, which is the practical mechanism this guide is describing — not a change to CRediT, just a clearer link between two things that already exist side by side in most manuscripts.
What’s still unsettled
This is a fast-moving area, and CASRAI’s position here is interpretive guidance applying the existing 14-role taxonomy and current publisher practice — not a new normative standard. As of this writing, credit.niso.org (the NISO CRediT Standing Committee’s own reference page) contains no AI-specific role or field, and no major publisher’s structured CRediT-collection form has one either. A joint initiative from the International Science Council, the Committee on Publication Ethics, STM, the World Conferences on Research Integrity Foundation, and the Global Young Academy is working toward a cross-organizational Global Reporting Standard for AI disclosure in research through 2026, which may eventually touch on contribution-taxonomy questions like this one. Until then, check the specific instructions of the journal or funder you’re submitting to — AI-disclosure policies (unlike the CRediT taxonomy itself) are still changing fast and are not uniform across publishers.
Frequently asked
Can I list an AI tool under the Software role instead of a person?
No. Every publisher policy reviewed for this guide — Elsevier, ICMJE, PLOS among them — states that AI tools cannot be listed as authors or contributors. The Software role, like every CRediT role, is assigned to the human accountable for that work, even when a tool assisted with it.
Where does the AI-use statement go relative to the CRediT statement?
It varies by publisher, but the two are kept separate. Elsevier places its “Declaration of Generative AI and AI-assisted technologies” at the end of the manuscript, immediately above the references. PLOS places AI disclosure in a dedicated Methods (or Acknowledgements) section. Neither folds it into the CRediT contributor-role block itself.
If AI helped write my code, does that change who gets the Software role?
No. The role stays with the person who directed the tool, reviewed and tested the resulting code, and is accountable for it. What changes is that the AI-use disclosure statement should name the tool and describe what it did — separately from the role attribution.
Is there an official CRediT extension or role for AI contributions?
Not as of this writing. The NISO CRediT Standing Committee’s own reference (credit.niso.org) has no AI-specific role or field, and no major publisher’s structured CRediT form has added one. A multi-organization Global Reporting Standard effort for AI disclosure is under development through 2026 and may eventually address this; until it lands, the two-statement convention described in this guide is the practical approach, not an official standard.
Do basic grammar or spell-check tools need to be disclosed at all?
Generally no. Elsevier’s policy, for example, explicitly exempts “basic checks of grammar, spelling and punctuation” from requiring a declaration. More substantive AI-assisted editing — restructuring passages, translation, generating text — does need disclosure.
