FAQ about Primo AI
The Central Discovery Index (CDI) of Primo is searched by Primo AI. This includes the data of electronic resources of different publishers and providers.
Not included, among other things:
- Print media in the University Library’s locations
- Some types of publication such as press content
- Content not approved for search by providers
- Content with insufficient bibliographic data
- Withdrawn content
- Providers whose content is not represented in the CDI
The natural language request is converted into a normal search query by a Large Language Model (LLM).
This search query is used to search for applicable sources in the Central Discovery Index (CDI).
The CDI is a database containing billions of bibliographic records of electronic resources. Not all content is licensed for the FU and not all publishers provide their data for AI search (for more information see "What data is being searched?").
From the results classified as appropriate, the first 30 results are re-ranked. For this purpose, the metadata, keywords and abstracts are evaluated, but not full texts.
Up to five sources are selected from these results.
From the abstracts of these five sources, a summary with footnotes to the corresponding sources is created.
The aim is to generate a short text with as few hallucination as possible and with existing sources from the CDI.
Note: Despite existing sources, the summary must always be compared with content of the cited sources.
No. Printed sources are currently not included in the search. The CDI lists only electronic literature.
Exception: If Primo AI selects an electronic book as a result which we only have in print at the FU, it will be displayed.
The CDI also includes resources that the FU has not licensed. To display only licensed sources, the checkbox "available within the university network" below the search bar in Primo AI can be used.
- In Primo, you can deactivate the slider "Display resources without direct access" on the right above the filters.
- Resources that we have not licensed can be found through interlibrary loan or in other libraries in Berlin/ Brandenburg. You can also submit purchase suggestions for books and e-books.
The AI-generated summary may not be quoted (directly or indirectly), because it cannot be accessed permanently and because it is only based on the abstracts and metadata of the sources. Therefore, it cannot be used as a basis for scientific texts. If you want to paraphrase content, use the original sources on which the summary is based. According to the FU's Guidelines for the Use of AI-Based Systems and Tools in Teaching and Learning, you are responsible for your texts, including AI-generated content and suggested sources.
The AI-generated summary of Primo AI always needs to be critically reviewed. It is inherent in the functioning of large language models that they can produce erroneous statements. The sources given are existing sources. However, you must always check whether the proposed sources meet subject-specific quality criteria, e.g.
- Topic relevance
- Currency
- Worth citing
In addition, check whether there are other relevant publications on your topic, e.g. by searching the library portal Primo or in subject-specific databases.
According to the FU's Guidelines for the Use of AI-Based Systems and Tools in Teaching and Learning, you are responsible for your texts, including AI-generated content and suggested sources.
- No, your questions will not be used to train the large language model. You have the option to rate the results with thumbs up/down. This is voluntary and is used by the providers of Primo (Clarivate/Ex Libris) to further develop Primo AI.
- Your user data from Primo will not be passed on to the provider or third parties.
- Note: Nevertheless, do not enter any personal data of you or third parties.
- Your user data from Primo will not be passed on to the provider or third parties.
- Note: Nevertheless, do not enter any personal data of you or third parties.
- The providers of Primo AI (Clarivate/Ex Libris) have published Guardrails for Responsible AI.
A comparison of the resource consumption of a classic search and an AI-supported search is not specifically available for Primo. However, it is generally known that the training of a language model consumes the most energy and a request to an AI-supported system requires significantly more energy than a conventional database query.
Some developers of large language models specify filters, meaning that summaries cannot be generated from the sources for certain topics. Therefore, it is important to always critically question the results. View the suggested sources directly, rephrase your request, or use the option to see more results in Primo.
- If you have any further questions, please contact: auskunft@ub.fu-berlin.de
- Also note our services for working with AI
- KI@FU (in German)
Keywords
- Primo AI


