Open Access and Artificial Intelligence: a practical agenda for publishers – part 2

Please note: This is part 2 of an AI-generated summary of a session organised by OIPA on AI in publishing, held on 5th June 2025, at which five speakers presented their views. These speakers were George Walkley, Andrew Gray, Rupert Gatti, Jane Secker and Tanya Kant. It has been edited by staff to ensure it’s an accurate reflection of the session, and the speakers have all given their permission to be quoted. The blog reflects AI-related challenges, discussions and practices in circulation at the time of the event. However the fast pace of change in technical, legislative and workplace contexts mean that some described processes and practices may have significantly evolved. If you missed it, read part 1.

Continued…

Controlling AI crawlers

Tanya Kant’s contribution introduced another practical question: should open access publishers block AI web crawlers?

A website’s robots.txt file can indicate which automated systems should not collect its pages. In some cases, instructions can be applied selectively to particular crawlers or areas of a website.

However, this is not a complete solution. A robots.txt instruction is a preference signal, not necessarily a technical barrier, and its effectiveness depends on crawler operators choosing to respect it. Blocking crawlers may also affect discovery and legitimate research uses.

Publishers should nevertheless understand their options. Decisions might take account of:

• whether the crawler identifies itself transparently;
• whether it respects licensing and attribution;
• whether it causes excessive demand on the publisher’s infrastructure;
• whether the collected content will be used for commercial model training;
• whether the service directs readers back to the original publication;
• whether access can be managed by page, content type or purpose;
• whether blocking the crawler could reduce legitimate discovery.

A selective approach may be more consistent with open access values than either unrestricted collection or blanket refusal. Collective monitoring by OIPA members could also help smaller publishers understand crawler behaviour without each organisation having to develop its own expertise independently.

AI literacy is more than prompt writing

AI training is often framed as a lesson in writing better prompts. Tanya Kant argued for a broader understanding of AI literacy, encompassing awareness, practical skills, critical evaluation and ethical judgement.

For publishers, AI literacy should include:

• understanding how different tools use submitted data;
• checking whether prompts and files may be retained or used for training;
• recognising hallucinated information, fabricated citations and biased outputs;
• understanding copyright, licensing and confidentiality risks;
• evaluating the environmental and infrastructure costs of AI;
• examining the commercial interests behind different platforms;
• knowing when an AI tool is inappropriate;
• keeping appropriate records of AI-assisted work;
• applying meaningful human review.

Different tools also require different forms of scrutiny. A system that cites sources and directs users to publishers’ websites presents different opportunities and risks from one that provides an unattributed synthesis. Staff should evaluate tools individually rather than treating every large language model as interchangeable.

A practical checklist for OIPA members

University-based and open access publishers can begin with a manageable set of actions.

1. Map current use
   Ask staff where and how they are already using AI. A supportive audit is more likely to produce an accurate picture than a punitive exercise.
   
2. Establish clear boundaries
   Identify acceptable, restricted and prohibited uses. Pay particular attention to unpublished manuscripts, personal data, peer-review reports, contracts and third-party copyrighted material.
   
3. Introduce proportionate disclosure
   Develop clear questions for authors, reviewers, editors and suppliers. Explain what must be disclosed and why.
   
4. Review contracts and guidance
   Ensure that author agreements, reviewer guidance, privacy information and supplier contracts address AI where necessary.
   
5. Check tool settings
   Review whether organisational accounts protect submitted information from reuse in model training. Free consumer tools may provide different safeguards from paid or institutional versions.
   
6. Keep humans accountable
   Assign responsibility for checking factual claims, references, permissions, translations, metadata and accessibility information.
   
7. Examine licences and crawler policies
   Understand how Creative Commons licences interact with AI uses and review the publisher’s robots.txt settings, server traffic and infrastructure costs.
   
8. Run bounded pilots
   Test AI on low-risk, repetitive tasks and record the time saved, correction rate and quality of the final output. A pilot should be assessed against evidence, not novelty.
   
9. Train staff critically
   Training should cover copyright, bias, confidentiality, environmental impact and source verification as well as practical tool use.
   
10. Share learnings
    OIPA members can reduce duplicated effort by sharing policy language, workflow experiments, tool assessments and evidence of crawler activity.

Openness requires transparency

The webinar did not offer a simple answer to the challenges posed by AI: the technology is developing quickly, its legal status remains contested and its effects vary considerably across disciplines and publishing contexts. A consistent theme nevertheless emerged: transparency must be central to the response.

Open access publishers are well placed to lead this work. Openness is not simply a technical licence or an absence of a paywall. It is a commitment to accountable processes, equitable access and the responsible stewardship of knowledge.

That commitment requires publishers to be honest about where AI is useful, clear about where it is not, and willing to explain how decisions are being made. By combining careful experimentation with collective scrutiny, OIPA members may be able to use AI to improve access and efficiency without surrendering the values on which open publishing depends.

Related articles

“Not Plausible, But Verifiable: Scholarly Metadata in the Age of AI”

AI in publishing with George Walkley (podcast episode)

AI for the Modern Editor