AI-Assisted Writing Policy

The journal’s position

Generative AI tools are in ordinary use, and a policy that pretended otherwise would only make disclosure less honest. BORJSS neither bans these tools nor treats their use as unremarkable. Three principles govern how the journal handles them.

1. Accountability cannot be delegated. A tool cannot take responsibility for what it produces, so the responsibility remains entirely with the human authors.
2. Substantive use must be visible. Readers and reviewers are entitled to know how a manuscript was produced where that bears on how they should read it.
3. Confidentiality is absolute. Nobody may put another person’s unpublished work into a third-party service.

“Generative AI” here means any tool that produces text, images, code, data or analysis from a prompt — large language model assistants, AI writing and paraphrasing tools, and image generators.

AI tools cannot be authors

No AI tool may be listed as an author or co-author, and none may be named in the acknowledgements as a contributor.

The reason is not symbolic. Authorship under the authorship policy requires approving the final version and being accountable for the integrity of the work — including answering for it after publication. A tool can do neither. Nor can it hold copyright, declare a competing interest, consent to publication, or respond to a query about its own output.

Use that must be disclosed

The following uses are permitted, and must be declared in the manuscript. Declaring them has no bearing on the editorial decision.

Where AI was used as part of the research method itself — for coding interview transcripts, say, or classifying a corpus — disclosure in the declaration is not enough. It belongs in the methods section, with the tool, the version, the prompts or classification criteria, and how the output was validated against human judgement, as any other instrument would be reported under the research integrity policy.

Use that needs no disclosure

Routine assistive tools that correct rather than generate need not be declared:

• Spelling and grammar checkers;
• Reference managers and citation formatters;
• Sentence-level suggestions of the kind a word processor offers;
• Search tools used to find literature, where you then read it;
• Statistical software used in the ordinary way.

The line is between correcting text you wrote and generating text you did not. If you are unsure which side a use falls on, declare it — over-disclosure costs nothing.

Use that is not permitted

• Fabricating data or results. Generated data presented as observed, or generated images presented as evidence, is fabrication, and is treated as such under the publication ethics policy.
• Generating references. AI tools invent plausible citations. Every reference must be one the authors have located and read. Fabricated references are the most common way undisclosed AI use surfaces, and they are treated as a serious integrity failure whether or not the tool was declared.
• Generating the analysis or the argument. The interpretation, the claims and the conclusions must be the authors’ own reasoning.
• Paraphrasing to evade similarity screening. Running another’s text through a paraphrasing tool is plagiarism, and is handled under the plagiarism policy.
• Entering confidential material into a tool. Applies to unpublished data belonging to others, participant data, and any manuscript you are reviewing.

How to disclose

Disclosure goes in a short statement near the end of the manuscript, before the references, headed “Use of AI tools”. It is published with the article. It should name the tool and version, say what it was used for, and state that the authors reviewed and take responsibility for the output. For example:

Where no generative AI was used, say so: “No generative AI tools were used in the preparation of this manuscript.” A missing statement is not read as an absence of use, and the manuscript is returned for one — as with the competing-interests statement under the conflict of interest policy.

Accountability

Authors are fully responsible for every part of a manuscript, whatever produced it. “The tool generated it” is not a defence to a fabricated reference, an inaccurate summary, an unattributed passage or a false claim.

Before submitting, authors should have verified that every reference exists and says what it is cited for; that every factual claim is supported by a source they have read; that no passage reproduces someone else’s text without attribution; and that the argument is one they can defend in their own words.

Reviewers

Reviewers must not enter any part of a manuscript, or their report on it, into a generative AI tool. A manuscript under review is confidential unpublished work, and uploading it transmits it to a third party outside the authors’ control.

This holds regardless of a tool’s claims about data retention, and regardless of the reviewer’s intent. It applies to summarisers and translation services as much as to chat assistants. A review must also be the reviewer’s own specialist judgement; a generated report is of no use to an editor. See the peer reviewer ethics policy.

Editors

The same confidentiality rule binds editors: no part of a submitted manuscript may be entered into a generative AI service at any stage of handling.

Editorial decisions are made by people. AI tools are not used to screen, assess, rank or decide on submissions, and no decision is taken on the output of an automated system.

Detection

The journal does not use AI-detection software, and does not accept detector output as evidence. These tools produce false positives at rates that make them unusable, and they flag the writing of second-language authors disproportionately — which in this journal’s author community would be an unfair instrument.

Concerns are instead pursued through what can be checked: whether the references exist and support what they are cited for, whether the described method could have produced the reported results, and whether the authors can explain their own analysis when asked. Those checks establish something; a detector score does not.

Undisclosed use

Where undisclosed use comes to light, the authors are shown the specific concern and given a fair opportunity to respond, following the procedure in the publication ethics policy. The response is proportionate:

• Undisclosed but otherwise legitimate use — the statement is added, before publication or by correction afterwards. The journal treats a first instance of this as a reporting failure rather than misconduct.
• Fabricated references or content — the manuscript is rejected, or, if published, corrected where confined to specific references and retracted where the substance is affected.
• Generated data or images presented as real — treated as fabrication: rejection or retraction, and the author’s institution informed.

This policy will be revised as these tools and the norms around them change. Questions about a specific use are welcome at ceoborjss@gmail.com before submission — asking is never held against an author.