Authors and reviewers should not have to interpret a vague sentence about using AI “responsibly.” They need to know what they may do, what they must disclose, and what they must never place in an outside tool.
This guide helps conference chairs turn those questions into a clear working policy. It focuses on the decisions that matter most: language assistance, substantial writing or research support, reviewer confidentiality, human responsibility, and the conference’s own use of AI-assisted review tools.
The downloadable pack is deliberately short. It includes a five-decision planning table, copy-ready policy wording, suggested fields for Dryfta submission and review forms, and a final launch checklist. Use the Word file to edit the wording. Use the PDF when you need a clean reference copy for colleagues or committee review.
Download the conference AI policy template: Editable Word template | PDF reference copy
The template is a practical starting point, not legal advice. Check the requirements of your proceedings publisher, institution, sponsors, professional body, privacy team, and applicable law before adopting it.
Why a conference needs a specific AI policy
Generative AI now appears inside writing assistants, search tools, coding environments, translation services, image software, and meeting platforms. A policy that only mentions one chatbot will become outdated quickly. At the same time, a policy that tries to cover every possible tool often becomes too technical for authors and reviewers to use.
The better approach is to focus on actions and information. What may a person ask a tool to do? What material may enter it? What must remain a human decision? What must be disclosed?
A useful conference policy gives clear answers at three moments:
- Before submission: Authors can tell whether a proposed use is allowed and what they need to report.
- Before review: Reviewers understand that unpublished work and review material are confidential, even when a tool appears private or convenient.
- When the conference uses AI: Participants know whether AI assists screening or review, how human judgment is protected, and whom they can contact with a concern.
Without these answers, people make different assumptions. One author may treat translation as routine editing while another discloses it. One reviewer may regard summarization as harmless while another recognises it as disclosure of confidential material. A short but precise policy prevents that confusion.
For the wider operational context, see Dryfta’s guide to the risks of AI in event management. The page you are reading has a narrower role: it helps a conference set practical rules for submissions and peer review.
Decide these five points before writing the policy
You do not need a long technical framework. Most conferences can build a sensible first policy by settling five questions.
| Decision | Recommended starting point | Why it matters |
|---|---|---|
| Spelling and grammar | Allow correction of an author’s own words without disclosure, provided the author checks every change. | This keeps the rule proportionate and avoids collecting disclosures that add little value. |
| Translation and substantial rewriting | Require disclosure. Ask the author to name the tool, purpose, and affected part of the submission. | These uses can change meaning, voice, or technical accuracy. |
| Generated text, code, analysis, data, or images | Require a clear disclosure and, when the use affects the research, a methods description. | Readers and reviewers need enough information to assess how the work was produced and checked. |
| Reviewer use of outside AI tools | Prohibit uploading or describing confidential material. Do not allow AI to replace evaluation, scoring, or recommendations. | Confidentiality and independent judgment are central to peer review. |
| AI assistance operated by the conference | Explain the purpose, limits, and human role. Keep final decisions with accountable people. | The conference should meet the same standard of transparency it expects from participants. |
This starting point will not suit every field. A conference handling sensitive clinical data, security research, unpublished intellectual property, or tightly controlled proceedings may need a stricter author rule. A conference with an approved institutional environment may permit a narrow reviewer use, such as improving the grammar of reviewer-authored text that contains no submission details. The final policy should state one clear choice rather than presenting participants with several models.
Draw a simple line between correction and contribution
Routine spelling and grammar correction generally changes how a sentence is presented, not the intellectual contribution behind it. Substantial rewriting, interpretation, analysis, code generation, synthetic images, or research design can shape the content itself. That distinction is easier to understand than a long list of brands or features.
The line will not always be perfect. Translation, for example, can preserve meaning in one case and materially alter it in another. That is why disclosure is a useful middle ground. It does not automatically make the use unacceptable. It gives the committee and reviewers the context they need.
Protect confidential review material
A manuscript, abstract, figure, dataset, reviewer report, score, rebuttal, committee discussion, and a hidden author or reviewer identity can all be confidential. Reviewers should not paste any of this information into a public or unapproved AI service.
This rule should also cover indirect disclosure. A detailed summary, a screenshot, a distinctive method, or a description of unpublished findings may reveal the substance of a submission even when the title and author names are removed.
Keep academic judgment human
A reviewer may use software to manage their work, but the review itself requires subject knowledge, context, and accountable judgment. The reviewer must read the submission, assess it against the published criteria, write the report, and choose the score or recommendation.
The same principle applies to chairs. A generated summary or recommendation should not quietly become the basis for acceptance, rejection, or a misconduct finding. If the conference uses AI-assisted review, the policy should say exactly what the system does and which decisions people retain.
Ask for disclosures people can actually complete
“Did you use AI? Yes or No” rarely gives enough information. It treats a spelling correction, a translated paragraph, generated code, and an AI-assisted data analysis as if they were the same activity.
A useful disclosure asks four things:
- Which tool or service was used? Include the version when it is known and relevant.
- What did it do? Describe the task rather than using a general phrase such as “assistance.”
- What part of the work did it affect? Name the section, code, data, figure, or other material.
- How did the authors check the result? Mention source checking, testing, validation, editing, or another human review step.
Explain the conference’s own use of AI
Participants may reasonably want to know whether an organizer-operated system assists reviewer matching, screens submissions, drafts summaries, or produces an additional review. The policy should state the purpose in plain language and explain that a person remains responsible for the consequential decision.
Dryfta’s AI Reviewers can work alongside human reviewers using the conference’s existing review questions and scoring scale. A conference can add up to three Virtual Reviewers and compare their ratings with human assessments. If you choose this route, tell authors before review begins, define how the output will be used, and keep the final decision with the program committee.
What current scholarly policies have in common
Policies differ across publishers, funders, and conferences, but several principles appear repeatedly.
The International Committee of Medical Journal Editors says an AI tool should not be listed as an author, humans remain responsible for accuracy and attribution, and manuscripts should not enter an AI system when confidentiality cannot be assured without the authors’ permission. Its guidance also calls for transparency about the tool and purpose.
IEEE’s submission and peer-review guidance requires disclosure when generated content appears in an article, while generally excluding editing and grammar assistance from that requirement. IEEE also tells reviewers not to process manuscript information through a public AI platform to generate review content.
Some organisations take an even stricter view of reviewer use. NIH prohibits generative AI in its peer-review process, with a route for pre-approved accessibility needs. Other conferences allow limited use under stated privacy and responsibility conditions. ICML 2026, for example, describes both a conservative reviewer policy and a more permissive option, while keeping responsibility for the review with the reviewer.
Author rules also vary. ACM CCS 2026, CoRL 2026, and VehicleSec 2026 use different wording and levels of detail, but each expects human responsibility and some form of transparency for substantive use.
These examples were checked on 19 October 2026. They do not create one universal standard. They show why a conference should publish its own position before submissions open, while respecting any stricter external requirement.
Copy-ready conference AI policy
The following wording covers the central rules without turning the policy into a technical manual. Replace every bracketed item and remove any option that does not apply.
Purpose and scope
This policy explains how generative AI may be used in submissions and peer review for [conference name]. It applies to authors, reviewers, chairs, committee members, and administrators involved in [submission types and stages]. If a publisher, institution, sponsor, professional body, or law sets a stricter requirement, the stricter requirement applies.
Rules for authors
Authors may use generative AI for spelling, grammar, and formatting corrections to their own text without disclosure, provided they review every change.
Authors must disclose the use of generative AI for translation, substantial rewriting, drafting, summarization, literature assistance, code, data processing, analysis, interpretation, or generated images and media. The disclosure must name the tool or service, explain the purpose, identify the affected material, and describe how the result was checked.
When generative AI forms part of the research method or affects the findings, authors must provide enough detail in the methods or another appropriate section for reviewers to understand and assess the use.
A generative AI system cannot be listed as an author. The named authors remain responsible for the accuracy, originality, citations, permissions, methods, and conclusions of the submitted work. A disclosure does not excuse fabricated references, false claims, plagiarism, copyright infringement, or another integrity failure.
Author disclosure statement
During the preparation of this submission, the authors used [tool or service, and version if known] for [specific purpose]. The use affected [sections, code, data, figures, or other material]. The authors [describe checking, editing, testing, source verification, or validation] and take responsibility for the final submission.
If the conference exempts only spelling and grammar correction, say so directly beside the disclosure field. Authors should not have to search a separate document to learn whether an everyday use needs reporting.
Rules for reviewers and chairs
Submissions, supplementary files, author or reviewer identities, reviews, scores, rebuttals, and committee discussions are confidential. Reviewers and chairs must not upload, paste, describe, or otherwise expose this material to a public or unapproved generative AI service.
Reviewers must personally read and evaluate the assigned work. They must not use generative AI to write the review, assign a score, judge novelty or validity, rank submissions, infer identity, produce a recommendation, or replace their independent judgment.
If [conference name] approves a limited use, it will identify the permitted service and task in writing. No confidential submission details may be included, and the reviewer remains responsible for the final report.
Reviewer declaration
I confirm that I did not submit confidential conference material to a generative AI service and did not use generative AI to replace my evaluation, scoring, recommendation, or review. I personally completed this review and take responsibility for its content.
If a rule may have been broken
Concerns should be sent to [contact] with the submission or review identifier and the relevant information. [Role or committee] will check which policy version applies, review the available evidence, and give the affected person a reasonable opportunity to respond before making an adverse finding.
A detector score, writing style, or intuition alone will not prove a breach. The conference will consider the specific rule, reliable evidence, any correction offered, the effect on confidentiality or review fairness, and the seriousness of the conduct.
Possible responses include guidance, a corrected disclosure, replacement of affected material, re-review, a warning, rejection or withdrawal, removal from a reviewing role, or referral under another applicable policy. The response will be proportionate to the evidence and impact.
If your organisation already has a research integrity or complaints process, link to it instead of creating a second system. The policy only needs to tell people where to report a concern and what basic safeguards apply.
Set up the policy in Dryfta
A policy works best when its questions appear inside the submission and review workflow. Dryfta lets organisers create separate submission types and custom fields, build customised review forms and scoring, assign reviewers manually or automatically, set reviewer limits, and use single-, double-, or triple-blind review.
Add focused fields to the submission form
Start with a required choice:
Did generative AI contribute to this submission beyond spelling, grammar, or formatting correction of the authors’ own text?
- No
- Yes
- Unsure. Please contact [role or email] before the deadline.
Show the following fields when an author selects Yes:
- Tool or service used
- Purpose of use
- Part of the submission or research affected
- Checks, testing, or validation completed by the authors
- Location of a fuller methods description, if the use affected the research
Add one required confirmation:
I confirm that this disclosure is accurate on behalf of all authors and will be updated if the use changes during revision, rebuttal, or final submission.
Do not ask authors to paste confidential prompts, copyrighted material, personal information, or large interaction logs into the form. The purpose is to understand the use, not to collect every keystroke.
Add one clear reviewer declaration
Use the reviewer declaration above as a required confirmation beside the review form rather than repeating it in a separate terms page. If a reviewer cannot confirm it, direct them to the chair before submission. A private note field can help the chair resolve a mistake or an approved exception without exposing the issue to other reviewers.
Match the visibility to the review model
In double-blind or triple-blind review, the AI disclosure must not reveal author identity. Use an anonymised disclosure in the submission when reviewers need it to assess the method. Keep identity-revealing details in a chair-only field.
Dryfta supports single-, double-, and triple-blind review. Before opening review, test the form with a sample author, reviewer, and chair account. Confirm exactly which fields each role can see.
Build the review form around human judgment
Use review questions that ask reviewers to assess the work directly: relevance, quality, originality, methodological soundness, clarity, and any field-specific criteria. Add scoring anchors so that two reviewers understand a number in a similar way.
If your conference uses Dryfta Virtual Reviewers, use the same review form and scoring scale for the AI-assisted assessment. Comparing like with like makes differences easier to spot. It does not make the scores interchangeable. Chairs should investigate large disagreements and read the underlying comments before deciding.
For more detail on assignment controls, see Dryfta’s guide to assigning abstracts to reviewers. For the full submission-to-publication context, see the conference workflow guide.
Common mistakes that make an AI policy hard to use
Using broad language without examples
Words such as “ethical,” “appropriate,” and “responsible” express an intention, but they do not tell an author whether translation needs disclosure or tell a reviewer whether summarization is allowed. Add concrete examples for the uses most likely to occur.
Treating every use as equally serious
A corrected comma is not the same as generated analysis or an uploaded confidential manuscript. If every use triggers the same disclosure and response, people may ignore the policy or over-report minor actions. Separate routine correction, disclosed contribution, prohibited disclosure, and research-integrity concerns.
Asking for a yes-or-no disclosure only
A Yes answer without context creates more work for the committee. A No answer may also be unreliable when authors are unsure what counts. Ask for the tool, purpose, affected material, and human checks, and offer an Unsure route before the deadline.
Naming a product instead of describing the rule
Tool names and features change. A rule based on actions and information remains understandable when a new service appears or an existing writing tool adds a generative feature.
Setting rules for authors but not for reviewers or organisers
A balanced policy covers everyone who handles the work. Authors need disclosure rules. Reviewers need confidentiality and judgment rules. Organisers need a transparent notice when they use AI-assisted review or decision support.
Relying on an AI detector
Detector output may be wrong and cannot explain whether a use was permitted or disclosed. It can prompt a closer human review, but it should not trigger automatic rejection or serve as the only evidence of a breach.
Publishing the policy after submissions open
Participants should see the rule before they decide how to prepare or review work. Publish it with the call for papers, link it from the submission form, repeat the reviewer rule in the invitation or dashboard, and include any final-material requirement in the acceptance instructions.
A practical launch checklist
Before opening submissions, confirm that:
- The policy names an owner, contact, version, and effective date.
- Authors can distinguish exempt correction from uses that need disclosure.
- The disclosure fields ask what the tool did and how the result was checked.
- Reviewer confidentiality covers submissions, identities, reviews, scores, rebuttals, and committee discussions.
- Reviewers know that evaluation, scoring, and recommendations remain their responsibility.
- Any Dryfta Virtual Reviewer use has a clear participant notice and a defined human role.
- Blind-review settings have been tested with sample accounts.
- The publisher and institutional requirements have been checked.
- A person or committee is ready to answer questions and handle concerns.
- The same current policy appears in author, reviewer, chair, and final-submission guidance.
Set a review date after the conference. Note the questions people asked, where the wording caused confusion, and whether a form field collected useful information. Update the next version based on that evidence rather than making the current rule longer by default.
Frequently asked questions
Can authors use generative AI in a conference submission?
That depends on the conference’s published rule and any stricter publisher or institutional requirement. A practical policy may exempt narrow spelling and grammar correction, require disclosure for translation or substantial rewriting, and require more detail when AI affects code, data, analysis, images, or research findings. The human authors remain responsible in every case.
Must authors disclose every AI-assisted correction?
Not necessarily. Many policies treat routine spelling or grammar correction differently from generated content. State the exemption precisely. If a tool rewrites arguments, adds material, translates technical content, or affects the research, disclosure is safer and more informative.
Can an AI tool be listed as an author?
No under this template. A tool cannot approve the final submission, declare conflicts, answer questions about the work, hold rights, or accept responsibility. Those duties belong to the named human authors.
Can reviewers use ChatGPT or another generative AI service?
Reviewers must follow the conference’s published rule and confidentiality agreement. The recommended starting point prohibits placing submission or review material in public or unapproved tools and prohibits delegating evaluation, scoring, or recommendations. A narrow approved use should be named in writing and must not expose confidential details.
Where should a disclosure appear during double-blind review?
Use an anonymised statement in the submission when reviewers need the information to assess the method. Keep identity-revealing details in a chair-only field. Tell authors before submission which fields reviewers can see and what must appear in the final version.
What should happen when an author did not disclose a use?
First establish what happened. Was the use permitted? Was the omission minor or material? Did it affect the research, confidentiality, or fairness of the review? Give the author a chance to respond and choose a proportionate outcome. A corrected disclosure may be enough in one case, while fabricated evidence or deliberate concealment may require a stronger response.
Can a detector prove that the policy was broken?
No. Detector output, writing style, or intuition alone should not establish a breach. The committee should examine specific evidence, the policy version, the disclosure, relevant source or revision records, and the person’s response.
How should a conference explain Dryfta Virtual Reviewers?
State that the feature supplies an additional AI-assisted assessment using the conference’s review questions and scoring scale. Explain how chairs will use the output, confirm that human reviewers and the program committee remain responsible, and provide a contact for questions.
Make the policy usable before submissions open
A good policy is not the longest document. It is the one an author, reviewer, or chair can understand before taking action.
Set the five important decisions, adapt the copy-ready wording, place focused questions inside the submission and review forms, and test what each role can see. Keep the rules proportionate, protect confidential work, and make human responsibility visible wherever AI assists the process.
The editable Word template gives your committee a five-page working document. The PDF reference copy is ready to circulate for review. Both include the decision table, policy clauses, Dryfta field wording, review settings, and launch checklist described in this guide.
Dryfta can help your team collect clear submission information, configure blind review, build scoring forms, manage reviewer assignments, and compare human assessments with Virtual Reviewer output. Request a Dryfta demo to discuss your conference workflow.
