Jan SlifkaArtificial intelligence has quickly become part of the research workflow. Many researchers now use large language models to summarize publications, improve manuscripts, write code, or prepare grant applications. Data Management Plans (DMPs) are no exception. It is becoming increasingly common to upload a project proposal to an AI assistant and ask it to generate the first draft of a DMP.
This approach makes sense. Most of the information needed for a Data Management Plan already exists before anyone opens a DMP template. Project proposals, work package descriptions, data collection protocols, software management plans, dissemination plans and infrastructure documentation all contain pieces of information that eventually need to be reflected in the final plan. Asking AI to extract and organize that information can save a considerable amount of time.
At the same time, using a general-purpose AI assistant raises practical questions. The generated content is separate from the rest of the research data management workflow, making it harder to keep project documentation, collaboration, version history and DMP preparation in one place. Research organisations may also have policies governing which AI services can be used with project documentation.
The Project Materials Prefill Plugin brings AI-assisted drafting directly into FAIR Wizard, allowing researchers to work with the documents they already have while keeping the entire DMP workflow in one place.
Large language models are very good at producing coherent text, but creating a Data Management Plan involves more than writing paragraphs. A DMP is a structured document that follows the requirements of a funding programme or institution. Researchers often revisit it throughout the lifetime of a project, collaborate with multiple contributors, and generate updated versions as plans change.
When AI is used as a separate application, the workflow typically involves several manual steps. Project documents are uploaded to the AI assistant, the generated responses are copied into the DMP questionnaire, and the finished document is reviewed and shared using other tools. While this process may still be faster than starting from an empty page, much of the work happens outside the system where the DMP is ultimately managed.
The Project Materials Prefill Plugin was developed to integrate AI into this existing workflow. Instead of generating a standalone document, it analyses project materials and suggests answers directly within the FAIR Wizard questionnaire. The suggestions become part of the project where they can be reviewed, edited, shared with collaborators, and included in the final Data Management Plan.
Whenever project documentation is processed by an AI service, privacy and data protection become important considerations. Research proposals and supporting documents may contain confidential information, intellectual property, or personal data, so organisations need to understand how that information is handled.
The Project Materials Prefill Plugin uses Anthropic Claude through Amazon Bedrock. Documents processed by the plugin are not shared with the model provider and are not used to train the underlying foundation models, while FAIR Wizard continues to manage the rest of the workflow.
The plugin also follows a defined data retention policy. Uploaded documents, extracted text, prompts and generated outputs are retained for up to 30 days before being automatically deleted. This temporary retention allows us to investigate technical issues, troubleshoot processing problems and provide support when users need assistance. Technical and security logs are retained for the same period to help monitor the service and resolve operational issues.
The plugin is designed to provide drafting assistance rather than automated decision-making. Every suggested answer should be reviewed and, where necessary, corrected before it is accepted as part of the project.
The workflow is designed to fit naturally into the existing FAIR Wizard experience.
The process starts by selecting the documents that already describe your research project. These may include the project proposal, grant application, work package descriptions, software management plans, infrastructure plans, data collection protocols, or other relevant documents. Multiple files can be uploaded together so the AI has a broader understanding of the project context. The upload takes place directly from the FAIR Wizard project.
Once the documents have been submitted, the plugin extracts relevant information and analyses it together with the context of the FAIR Wizard questionnaire. Instead of producing a separate report, it generates suggested answers for individual questions within the questionnaire. Because the analysis may take several minutes, FAIR Wizard sends an email notification when the prefilled questionnaire is ready for review.
The generated answers appear directly in the questionnaire, where they can be edited, expanded or replaced before they are accepted. This review step remains an important part of the process because AI-generated content may be incomplete or may not fully reflect the specific details of the project. The plugin provides a starting point that researchers can refine, rather than a completed DMP.
After reviewing the suggested answers, the project continues exactly like any other FAIR Wizard project. Researchers can collaborate with colleagues, share the project with data stewards, make further revisions, and generate the final Data Management Plan using the existing FAIR Wizard features. The AI-assisted drafting becomes one step within the broader research data management workflow rather than a separate activity.
Many researchers have already discovered that AI can help produce the first draft of a Data Management Plan. The Project Materials Prefill Plugin builds on that idea by integrating AI directly into FAIR Wizard, where the rest of the DMP process already takes place.
Instead of copying information between multiple applications, researchers can begin with the documents they have already prepared, receive suggested answers within the questionnaire, and continue working in the same environment where the project is managed, reviewed and shared. The result is a workflow that reduces repetitive work while keeping researchers in control of the final content.
