Artificial intelligence has moved from the periphery of research information management to its centre. Legacy enterprise RIMS platforms are repositioning around AI-assisted profile curation and AI-powered research intelligence; funders and policy bodies are debating where AI belongs in research evaluation; and 81% of researchers report active LLM use in their daily work. The question for a research office in 2026 is no longer whether AI belongs in your RIMS, but where — and where it does not.
This guide is a practical reference for VPs of research, research office directors, and CIOs deciding how to adopt AI inside a RIMS responsibly. It covers what AI can do well today, what it cannot, what good governance looks like, and why none of it works without the data discipline that has always been the foundation of credible research information.
Key takeaways
- AI in a RIMS is mostly disambiguation, summarisation, and pattern detection. It is rarely a replacement for human evaluation.
- Author name disambiguation is the highest-value, lowest-risk application of AI in a RIMS — but only when paired with ORCID and other persistent identifiers.
- LLM-generated summaries are useful for discovery and discoverability, dangerous for evaluation. Treat them as descriptions, not judgements.
- AI evaluation tools must align with DORA and CoARA principles — research is evaluated on its own merits, not on opaque AI-derived scores.
- Data quality is the prerequisite. An AI layer on a poorly reconciled dataset accelerates wrong answers; it does not fix them.
Where AI helps inside a RIMS
The strongest current use cases share a pattern: AI accelerates work humans already validate, on data that has already been reconciled. Author disambiguation uses machine learning to match name variants, transliterations, and affiliation changes to a single researcher record — work that previously required manual review at scale. See AI-Assisted Author Disambiguation for the detail. Publication summaries use LLMs to generate plain-language abstracts of a researcher's recent outputs, surfaced on public profiles. Topic clustering surfaces emerging research themes from output metadata, useful for strategic planning. Anomaly detection flags duplicate records, missing metadata, and ingestion errors faster than periodic manual review.
Where AI is risky inside a RIMS
Equally important is where AI should not lead. Research quality evaluation via LLM-derived scores is opaque, biased in ways not yet understood, and difficult to audit — a problem covered in detail in LLM for Bibliometrics and Research Summaries. Hiring and promotion decisions driven by AI-generated rankings of researchers fall foul of the DORA principle that research is evaluated on its own evidence. Generated content as primary record — AI-summarised abstracts treated as authoritative — risks the publication record drifting from what the researcher actually published. The pattern: AI is helpful as a layer; it is dangerous as a verdict. We unpack the policy framing in Responsible AI in Research Evaluation.
The data-quality prerequisite
No AI feature reaches its potential on poorly reconciled data. An author-disambiguation model fed inconsistent affiliations produces confidently wrong matches. An LLM summarising publications missing from the institutional record summarises a fiction. A topic-trend dashboard built on incomplete metadata identifies trends that do not exist. The first investment when planning AI in a RIMS is not the AI — it is the data foundation underneath. That is the argument in AI is Only as Good as the Data Beneath It, and the foundation we describe in Building a Single Source of Truth for Research Data.
Governance: who is accountable for the AI?
Responsible AI adoption inside a RIMS needs the same governance as any other consequential decision support. Document which models are in use, on which data, with which review process. Make AI-generated content visibly labelled when surfaced (profile summaries, suggested affiliations). Provide a manual override path for researchers and administrators. Audit periodically. None of this is novel; it is what any institutional system handling researcher-affecting decisions already requires, applied to AI features.
What this looks like in production
Discover RIMS operates in production at Universitas Hasanuddin across 2,500+ researchers, 15,300+ publications, and 18 faculties and research units. The foundation that makes any future AI layer credible — reconciled author identity across ORCID, Scopus, OpenAlex, Crossref, and Scimago — is in place today, before adding generative features that depend on it.
Frequently asked questions
Should we wait until AI in RIMS is more mature before adopting? Adopt the data foundation first. AI features can be added incrementally once that foundation is in place; without it, no future AI integration will produce reliable results.
How does AI affect DORA and CoARA compliance? AI-derived scores are exactly the kind of opaque, non-contextual proxy DORA and CoARA caution against. Use AI for discovery and discoverability, not for individual evaluation.
Will AI replace research office staff? No. It changes their work toward higher-judgement tasks — reviewing AI outputs, governing the models, narrating strategy — and away from manual reconciliation.
Where to start
Begin with reconciliation, not generation. Discover RIMS unifies five global sources — Scopus, OpenAlex, ORCID, Crossref, and Scimago — into one continuously reconciled record. That is the foundation every credible AI feature in a RIMS, today or later, depends on.