The Future of Research Integrity: Building a Seal of Authenticity for Trusted Scholarship

Take Home Points

• Stop asking, "Was AI used?" Start asking, "Why should I trust this research?"

• The scholarly record needs visible accountability, not invisible assumptions of good faith.

• The strongest trust signal is not an AI-free badge—it is a transparent record of authorship, evidence, review, and correction.

The most dangerous thing about AI-generated research content is not that it looks fake. Rather, it is that it can look entirely plausible.

A manuscript can now arrive with fluent prose, credible structure, polished figures, confident claims, and references that appear legitimate. Some of it may be perfectly valid, while other sections may be carelessly assembled, or even deliberately fraudulent. The problem for readers, editors, reviewers, and publishers is that a casual reading of the article tells us less than it used to. 

Given the current state, the next phase of research integrity should not be focused on whether AI was used. It needs to ask a more nuanced question: "Can I see why this work should be trusted?" 

Scholarly publishing has always depended on trust. Traditionally, our tools and processes have assumed that we are mostly dealing with individuals acting in good faith. We trust that authors conducted the work they describe, and that peer review is being performed with appropriate independence and expertise. We trust that data, images, methods, ethics approvals, citations, and conflicts of interest have not been fabricated, manipulated, or concealed. And we largely take authors, reviewers and editors at their word when they make such declarations.   

Trust in the Age of AI 

In the age of AI, readers need more than the traditional signals of credibility. A journal name, a DOI, and a statement that an article was peer reviewed are no longer enough on their own. They remain important, but they do not answer the growing set of questions readers now bring to published research: Who contributed to this work? Were AI tools used? If so, precisely how was it used? Is the data available? Were the images checked? Were the references verified? Has the article been corrected or flagged since publication? 

This is where scholarly publishing should be heading: toward a visible, verifiable seal of authenticity for research outputs. But we need to be clear about what that seal should — and should not — mean. 

It should not mean "no AI was used." That is the wrong standard and an unrealistic one. AI is already embedded in the research and publishing workflow, from language editing and translation support to coding assistance, literature discovery, image analysis, and editorial triage. The issue is not the mere presence of AI. The issue is the absence of accountability. 

AI detection will not solve this problem. Detection tools are too limited, too easy to misunderstand, and too narrow in what they measure. Even if they worked perfectly, they would only tell us something about the likely origin of the text.

A meaningful seal of authenticity should not then tell readers that an article is AI-free. It should tell them that the article has passed through a transparent integrity process. 

A Seal of Authenticity 

A credible seal of authenticity would function less like a sticker and more like a passport. It would not simply announce that an article is "trusted." It would show the basis for that trust. It would make visible the chain of responsibility, disclosure, verification, and correction that sits behind the final publication. 

At minimum, that should include five elements: 

1. First, verified human responsibility. Readers should be able to see that the authors are identifiable, that their affiliations are credible, that their contributions are described, and that conflicts of interest and funding sources are disclosed. ORCID, CRediT, institutional identifiers, and funder metadata are not just administrative conveniences. Used properly, they are part of the trust infrastructure. 
2. Second, meaningful AI disclosure. A clear statement of whether AI tools were used, which tools were used, and for what purpose. Language polishing is different from generating text. Generating text is different from analyzing data. An AI-assisted figure is different from an AI-assisted literature search. The disclosure should be specific enough to help editors and readers understand the nature of the contribution. 
3. Third, evidence availability. Where possible, the article should connect readers to the underlying data, code, protocols, materials, preregistration, ethics approvals, trial registrations, or image originals. Not every discipline can make everything open. There are legitimate privacy, consent, legal, and commercial constraints. But when evidence cannot be shared, the reason should be clear. "Available on reasonable request" should not be allowed to function as an acceptable default. 
4. Fourth, integrity checks. Journals and publishers should be transparent about which checks have been completed: plagiarism screening, image screening, citation checks, reviewer identity validation, conflict-of-interest review, ethics statement review, and data availability checks where relevant. The point is not guaranteed perfection. The point is to show that the article did not simply pass from submission to publication on trust alone. 
5. Fifth, a living article status. Trust does not end at publication. If an article is corrected, updated, subject to an expression of concern, retracted, or removed, that status must travel with the article wherever it appears. Crossref's Crossmark service demonstrates how this can work in practice, enabling readers to check the status of content across platforms, including in PDFs downloaded before a later change. NISO's work on communicating retractions, removals, and expressions of concern is important because the scholarly record needs correction signals that are visible, standardized, and machine-readable. 

This last point is critical. A seal of authenticity cannot be static. If it remains unchanged after a serious correction or expression of concern, it becomes misleading. Trust signals must be dynamic because the scholarly record is dynamic. 

Provenance 

Publishers have a particular responsibility here. Journals can set policies and apply checks, but publishers control much of the infrastructure that determines whether trust signals are visible, persistent, and interoperable. If an authenticity seal appears only as a graphic on a PDF, it will fail. Readers increasingly encounter research through indexing services, repositories, discovery platforms, systematic review tools, institutional systems, and AI-powered search environments. The trust signal must travel with the content. 

Provenance standards may become part of this future. C2PA, for example, offers an open standard for digital content provenance through Content Credentials, which are often described as a kind of "nutrition label" for digital content because they can show information about origin and edits. For scholarly publishing, this kind of approach could be especially useful for images, figures, datasets, supplementary files, and potentially article versions. The Scholarly Kitchen has also discussed the relevance of content provenance and C2PA-style approaches to research integrity and image manipulation.  

A better model may be a layered "Research Integrity Credential" rather than a single seal. One layer could show author attestations and disclosures. Another could show journal-level checks. Another could show links to data, code, images, protocols, and funding. Another could show post-publication status. Readers could then see not just whether an article carries a trust mark, but what that trust mark is based on. 

This approach would also shift the conversation away from AI panic and toward research integrity more broadly. The same infrastructure that helps disclose AI use can also help address paper mills, fake reviewers, manipulated images, unverifiable data, undisclosed conflicts, and poor correction practices. 

The Future Promise 

AI may be forcing scholarly publishing to do something it should have done more aggressively already: make accountability easier to see. 

The industry should not promise readers that every article carrying a seal is error-free or unquestionably valid. That would be impossible and irresponsible. It should promise something narrower but more useful: that defined checks were performed, defined disclosures were made, defined evidence is available, and defined accountability sits behind the publication. 

Building such a system would require coordinated action across the scholarly communication ecosystem. Publishers, journals, scholarly societies, universities, research funders, repositories, indexing and discovery services, libraries, standards bodies, and infrastructure providers would all have roles to play. This kind of cooperation is ambitious, but it is not unprecedented. The DOI system has shown that shared, persistent infrastructure can operate across publishers and platforms, while Crossref's Crossmark service already allows publishers to communicate corrections, retractions, and other updates through metadata that can be accessed across platforms and from downloaded PDFs. NISO's CREC Recommended Practice offers another model: publishers, aggregators, full-text hosts, libraries, researchers, and other stakeholders worked together to establish common practices for creating, transferring, and displaying retraction-related metadata. 

A practical first step could be for a neutral standards or infrastructure organization, such as NISO or Crossref, to bring together publishers, repositories, research institutions, funders, and technology providers to define the minimum evidence, governance, and interoperability requirements for the seal. Initial development could be supported jointly by publishers and research funders, which have a direct interest in preserving confidence in the literature, with universities, societies, libraries, and infrastructure providers contributing expertise, testing, and governance. The eventual signal would need to work not only on a publisher's website, but also wherever an article is discovered, stored, cited, or reused. 

Research integrity in the age of AI will not be protected by detectors alone. It will be protected by transparent authors, rigorous journals, responsible publishers, interoperable metadata, durable provenance, and correction systems that follow the article into the places where readers find it. 

The future trust signal for scholarship should not say, "No AI touched this." It should say, "Here is the evidence behind this work. Here is who is responsible for it. Here is how it was checked. Here is what has changed since publication."

That is the seal of authenticity scholarly publishing needs: not a badge that asks readers to trust us, but a record that shows them why they can. 

KnowledgeWorks Global Ltd. (KGL) is the industry leader in editorial, production, eLearning, online hosting, and transformative services for every stage of the content lifecycle. We are your source for learning solutions, intelligent automation, research integrity, digital delivery, and more. Email us at info@kwglobal.com.