The future of scientific publishing: choices an editorial team can make now

1. Separate what exists from what your journal can operate

A standard can describe an exchange format; a provider can document a feature; a journal can propose using it. None of those statements establishes that the journal has the staff, permissions or working configuration to deliver it. Start every proposal with the reader problem and the current route through the editorial office.

For example, a reader may need to discover whether an article has been corrected. That problem calls for maintained update records and clear links. A new writing assistant is relevant only if it helps the team perform that work without introducing a new source of error. Select the tool after the responsibility and output are clear.

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

2. Give AI a job whose result a person can assess

Define an assistive task in terms of an input and an output: suggesting clearer wording for an approved paragraph, identifying a missing heading, or drafting a lay summary for editorial review. Separate those tasks from deciding whether a study is valid, whether misconduct occurred, or whether a manuscript should be accepted. The latter decisions require the journal’s qualified people and procedures.

For medical journals, ICMJE calls for clear AI-use policies, disclosure of AI used in production, and editorial responsibility for the accuracy of ancillary content. It cautions against uploading confidential manuscripts where confidentiality cannot be assured without authors’ explicit permission. Its reviewer guidance also requires attention to journal policy, confidentiality and disclosure.

A proposal should state what material the service receives, which terms and retention arrangements were checked, and what remains unknown. If those arrangements cannot support the intended confidential input, use permitted synthetic or public material for the rehearsal. A fluent sample is not permission to send the next manuscript.

Keep the approved source beside the suggestion. The person reviewing it should be able to reject a changed number, missing qualification or new claim before it enters a manuscript, letter or public summary. Record who accepted the final text and what the assistance actually did.

Reference: ICMJE: editors’ role in responsible AI use · ICMJE: AI use by reviewers

3. Connect the article to the work needed to understand it

A reader may need a protocol, an analysis script, a data dictionary or a correction alongside the narrative. Make each object’s role explicit. A link should identify what it points to, its version, who maintains it and any access conditions. An empty repository or a planned identifier is not a substitute for the object itself.

JATS provides XML elements and attributes for representing journal articles. That supports structured exchange; it does not verify the underlying research. CRediT describes contributions through a taxonomy of 14 roles that complements authorship. Assigning a role should not bypass the journal’s authorship criteria or resolve a contributor dispute automatically.

For a small journal, begin with one representative article package. Check whether the title, table, figure caption, references and Arabic text survive the actual export. Let the research team check the scientific relationships between files. The production team can confirm that a script is present without claiming to have reproduced its results.

Reference: NLM: Journal Article Tag Suite · NISO: Contributor Role Taxonomy (CRediT)

4. Make openness a usable access arrangement

The UNESCO Recommendation on Open Science supports broad access while recognizing proportionate, justified restrictions, including privacy and confidentiality. It also recognizes shared infrastructures and community needs. This is a policy framework, not a determination that a particular dataset may be released or that a journal meets local obligations.

Describe the arrangement for each research object. An article might be public while participant-level data require a reviewed request under an authorized process. Give readers an accurate availability statement: what exists, what can be accessed, where a request goes, and what conditions apply. Do not promise access when no responsible custodian or usable procedure exists.

Openness also needs readable formats, useful descriptions and maintenance. Budget for correcting links, updating access information and supporting readers in the journal’s languages. A free-to-read page can still be hard to use if its tables are inaccessible or its Arabic explanation contradicts the English result.

Reference: UNESCO: Recommendation on Open Science

5. Plan for the article after publication

An article can acquire a correction, an expression of concern, a retraction or a related later version. Decide who receives those matters, who makes the editorial decision, and who updates each public format. A future publishing service needs that continuing responsibility, not just a fast route to an initial page.

Crossmark can help readers inspect updates where a publisher implements and maintains it. Crossref makes clear that the button is not a guarantee; participating publishers have responsibilities to maintain content and register updates. Consider the operational commitment before adding the display element.

Rehearse a correction through the files a reader actually receives. If a PDF was downloaded before the correction, what route helps that reader find the current status? Preserve the relationship between the original article and the notice. Removing an uncomfortable record without an appropriate editorial process does not create a trustworthy publication history.

Reference: Crossref: Crossmark, updates and publisher responsibilities

6. Three possible operating models, with reasons to defer each

The following are planning scenarios, not predictions about adoption or delivery dates. A team can combine parts of them or decide that its current process is more appropriate. The reader benefit should justify the extra work and the obligations that continue after launch.

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

7. Worked example: a fluent summary is returned for correction

This is an invented editorial meeting, using an invented public abstract and no participant data. A journal is considering an AI-assisted lay-summary service. The source states that the study did not establish whether the intervention reduces admissions. In this fictional rehearsal, the English draft preserves that limit; the Arabic draft says that the intervention reduces admissions. Neither summary has been released.

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

8. Fund the continuing work before expanding the service

Start with the smallest change that addresses a real reader or editor problem. Estimate staff review, accessibility work, support, storage, service costs and the effort of leaving the tool. A society subsidy, institutional support or author charge has different implications for the budget; none should determine an individual editorial decision. Use the team’s own figures rather than an invented return-on-investment percentage.

If measuring time, define the start and end of the task and include human correction time. If measuring quality, specify the errors and who judges them. A comparison needs comparable work and conditions. Keep a service modest until the team has enough relevant evidence and sustained capacity to accept the responsibility.

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

Questions for a practical final check

• Documented capabilities, local observations and future scenarios are labelled separately.
• AI assistance has permitted inputs, a bounded task and a responsible reviewer.
• Each research object has a real version, custodian and accurate access statement.
• Corrections and bilingual meaning checks remain part of the operating plan.
• Expansion depends on relevant evidence and a funded maintenance responsibility.