IEEE authorship has long been governed by a single test: a three-part, conjunctive checklist covering significant intellectual contribution, drafting or revision, and final approval (see CASRAI’s IEEE Authorship Guidelines: The 3-Part Test for the full breakdown). That test was written for human contributors and says nothing explicitly about generative AI tools such as ChatGPT, Copilot, or Gemini. As those tools became common in manuscript preparation, IEEE added a separate, more recent layer of guidance addressing generative AI specifically — disseminated through the IEEE Author Center and echoed in parallel guidance from individual IEEE technical societies. This guide explains exactly what that AI-specific layer adds on top of the pre-existing 3-part test, where the two rules interact, and what it means in practice for an author preparing an IEEE submission.
The 3-part test, briefly
IEEE’s authorship test requires all three of the following, not just one or two:
- Significant intellectual contribution to theoretical development, system or experimental design, prototype development, or data analysis and interpretation.
- Drafting or revision of the manuscript for intellectual content.
- Final approval of the version accepted for publication.
Anyone who does not meet all three should be named in an acknowledgments section instead of the byline — the same structural logic ICMJE uses for its own four-part authorship criteria. Full detail, including how the test handles contested cases, is in the dedicated 3-Part Test guide.
Why the 3-part test alone doesn’t settle the AI question
Read literally, the 3-part test is a test of what was contributed, not who or what is doing the contributing. A generative AI tool can, in a narrow technical sense, produce draft text, suggest analysis, or restructure a section — activities that sound adjacent to criteria 1 and 2. Nothing in the original three-criterion wording says a tool categorically cannot satisfy them. That ambiguity is exactly the gap IEEE’s generative-AI guidance was written to close: rather than asking editors and program committees to reason out, case by case, whether an AI’s output amounts to a “significant intellectual contribution,” IEEE added an explicit, categorical rule that operates upstream of the 3-part test entirely.
What the AI-exclusion policy actually adds
Four things, none of which are present in the original 3-part test:
1. A categorical bar, independent of contribution level
An AI system cannot be listed as an author or co-author on an IEEE paper, full stop. This isn’t a finding that AI tools typically fail the 3-part test on the facts — it’s a rule that forecloses the question before it gets to that test. No amount of drafting or analytical output from a generative AI tool converts it into an eligible contributor.
2. An accountability rationale the original test doesn’t spell out
IEEE’s stated reasoning is that authorship carries an accountability obligation a tool cannot bear: authors must be able to stand behind the accuracy and integrity of the work, defend it under scrutiny, and be held responsible if something is wrong. A generative AI system cannot do any of that. This makes explicit something the 3-part test’s final-approval and drafting criteria only implied.
3. A disclosure requirement with no equivalent in the 3-part test
The 3-part test says nothing about disclosing how a manuscript was produced — it only defines who qualifies for the byline. IEEE’s AI guidance adds an affirmative transparency obligation on top: authors who use generative AI in preparing a manuscript are expected to disclose that use, typically in the acknowledgments section, describing what the tool was used for. This is a new documentation requirement, not a restatement of the authorship threshold.
4. An explicit link to research-integrity enforcement
IEEE’s guidance treats using generative AI to fabricate or manipulate data, figures, or code as a research-integrity violation, not merely an authorship-eligibility problem. That connects the AI question to IEEE’s existing plagiarism and misconduct machinery (see IEEE’s plagiarism policy and Prohibited Authors List) in a way the 3-part authorship test, which is only about who belongs on the byline, was never designed to do.
Where authors still have to use judgment
The categorical AI-authorship bar is unambiguous, but the disclosure obligation still requires interpretation in edge cases:
- Language-polishing tools (grammar and phrasing checkers) sit closer to conventional copyediting and are generally treated differently from generative drafting or idea-generation tools, though authors should check the specific venue or technical society’s current guidance rather than assume an exemption, since wording varies by outlet.
- AI-assisted literature search or summarization used to inform an author’s own writing, rather than generate submitted text directly, is a judgment call for disclosure — the safer default is to disclose whenever a tool’s output materially shaped the submitted content.
- Co-author consistency matters on multi-author papers: one disclosure covering all AI use across the manuscript, agreed among co-authors before submission, avoids inconsistent or missing disclosures from different sections written by different people.
For the full mechanics of AI-disclosure wording, placement, and which tools are commonly treated as exempt across publishers generally (not just IEEE), see CASRAI’s IEEE generative AI policy dictionary entry and the broader Journal and Publisher Policies on Generative AI in Manuscripts guide.
How IEEE’s approach compares to the wider publisher landscape
On the core rule — no AI co-authorship, with disclosure required for AI-assisted drafting — IEEE lines up with the near-universal position across scholarly publishing, including ICMJE’s Generative AI Policy (added to the ICMJE Recommendations in the May 2023 update) and Nature Portfolio’s AI policy. Where IEEE differs is governance structure: rather than a single, centrally branded policy document, IEEE’s generative-AI rules are disseminated through the IEEE Author Center and echoed in parallel guidance from individual IEEE technical societies (among them AESS, CAS, RAS, and PES). The substance is consistent across societies, but exact wording can vary by venue — a practical reason to check the specific journal or conference’s own author guidelines before submission rather than relying on a single IEEE-wide policy page.
Practical takeaway for authors
The 3-part test still answers “who on this team qualifies as an author.” IEEE’s generative-AI guidance answers a question the 3-part test was never built to address: “can a tool be one of those authors, and what do human authors owe in disclosure when a tool was involved.” Treat them as complementary, not competing, rules — an AI tool is excluded from the byline before the 3-part test is ever applied to it, and every human author who does qualify under the 3-part test remains fully accountable for the entire manuscript regardless of how much AI assistance was used in preparing it.
Frequently asked questions
Can I list ChatGPT, Copilot, or another AI tool as a co-author on an IEEE submission?
No. IEEE does not permit an AI system to be listed as an author or co-author under any circumstances, regardless of how much of the drafting or analysis it performed.
Does IEEE’s AI policy replace the 3-part authorship test?
No, it supplements it. The 3-part test still determines which human contributors qualify as authors. The AI-specific guidance adds a categorical exclusion for AI tools plus a disclosure obligation that the original test doesn’t address.
Do I need to disclose using an AI tool just to check grammar on my IEEE manuscript?
Basic language-polishing tools are generally treated differently from generative drafting tools, but exact treatment can vary by IEEE venue, so check the specific journal or conference’s current author guidance rather than assuming an automatic exemption.
What happens if I don’t disclose generative AI use in an IEEE submission?
Undisclosed AI use that later comes to light can be treated as a transparency and integrity problem under IEEE’s broader research-integrity framework, separate from, and in addition to, any authorship-eligibility question.
Is IEEE’s generative-AI policy the same across every IEEE journal and conference?
The substance is consistent — no AI co-authorship, disclosure required for AI-assisted drafting — but the rules are disseminated through the IEEE Author Center plus individual technical societies rather than one single centrally branded document, so exact wording can vary by venue.
