Writing high-quality queries in generative AI: 7 principles from information science
Generative artificial intelligence (AI) seems to have turned every user into a search expert. However, simply typing a question into an interface is not a search strategy, nor does it guarantee a conclusive result. Prompt engineering is the gateway to relevant and targeted responses in generative artificial intelligence. In other words, a good answer depends on the quality of the question and the context in which it is asked, a concept often referred to as context engineering. Information science provides keys to support us in this interactive exercise and aligns perfectly with the emergent field of generative AI. Here are seven fundamental tenets of information science that can be used to formulate effective queries and yield more relevant results.
1. The principle of the reference interview: contextualizing the need
In the field of information science, librarians explore the real needs of requesters through a reference interview. Behind the initial question lie all the components that need to be exploited during the search.
Application to writing a query — Rather than submitting a raw query, provide rich context: your role or the role you assign to the tool, the target audience you are aiming for, the objective you are pursuing, the level of depth you are looking for.
Example — Rather than "Explain information monitoring to me," write: "I am (or you are) an analyst in the pharmaceutical sector. I need to prepare a summary note for my executive committee on emerging trends in information monitoring applied to research and development. What recent methodological approaches should I cover?"
2. The principle of controlled vocabulary: choose your terms carefully
Thesauri, keyword lists, and controlled vocabularies exist for a reason: synonymy and polysemy are the enemies of documentary research. The same word can refer to very different realities depending on the field, while the same reality can be translated into several words in free vocabulary.
Application to writing a query — Favor specialized vocabulary from your field to provide both a semantic and ontological framework. For example, if you work in monitoring, clearly distinguish between "information monitoring", "economic intelligence" and "competitive intelligence". Each term directs the tool toward a different response register. Also specify the terminology and language level to be used in the response if you have preferences (French, English, technical jargon, or popularized language).
3. The principle of facets: structuring the query into dimensions
Mathematician and librarian Shiyali Ramamrita Ranganathan, father of faceted classification, taught us to break down a subject according to its fundamental dimensions: Personality, Matter, Energy, Space, Time (PMEST). This multidimensional approach remains a powerful intellectual tool for structuring any information request. Much like constructing a bibliographic reference, it involves specifying metadata.
Application to writing a query — Break down your query into explicit facets. Specify the what (subject), the who (target audience), the how (format, tone, structure), the where (geographical or sectoral context), and the when (time period). A faceted query leaves little room for ambiguity and reduces irrelevant responses. This is also the principle of the 5 Ws in journalism.
Example — "SUBJECT: automated monitoring tools. AUDIENCE: university library managers in Quebec. FORMAT: comparative table. CRITERIA TO BE COVERED: cost, features, integration with library management systems, learning curve. PERIOD: tools available in 2026."
4. The principle of search strategy: iterate and refine
An effective search strategy is rarely linear. You broaden, narrow, combine concepts with Boolean operators, and rephrase. Searching is an iterative process that builds on the results obtained.
Application to writing a query — Take a conversational and iterative approach with the AI tool. Start with an exploratory query, analyze the response, then refine. Ask the tool to clarify, expand on a point, rephrase from a different angle, or exclude certain aspects.
Interacting with a generative AI tool is a dialogue, not a one-time query. The best results are obtained after three, four, or even five successive exchanges, just as you would refine a search string in a database.
5. The principle of relevance: defining your evaluation criteria
In information science, recall and precision are the two classic measures of a tracking system's performance. It is not usually possible to maximize both simultaneously: a choice must be made.
Applying this to writing a query — Explicitly tell the generative AI tool whether you are looking for comprehensiveness (a complete overview of a topic, even if it includes marginal elements) or accuracy (only the most relevant and reliable elements). This distinction, which comes naturally to information professionals, is very useful for calibrating the response. You can also specify filtering conditions: "Only include academic sources" or "Focus on the three most cited approaches in the literature".
6. The principle of source evaluation: requiring traceability
Information literacy teaches us to evaluate sources according to established criteria: authority, timeliness, objectivity, coverage, and accuracy. This critical reflex is essential when dealing with generative AI tools, as they can produce plausible but inaccurate content, known as "hallucinations".
Application to writing a query—Incorporate traceability requirements directly into your query. Ask the tool to cite its sources, distinguish between established facts and assumptions, and point out areas of debate in the literature. Generative AI is a starting point for research, not a primary source. Your expertise in evaluating sources is your asset and also allows you to put algorithmic literacy into practice, i.e., "being aware of the use of algorithms [...], understanding how they work, being able to critically evaluate algorithmic decision-making, and possessing the skills required to manage and even influence algorithmic operations". [2]
7. The principle of information mediation: specifying the output format
The role of the information professional is not limited to finding information: they format it, repackage it, and adapt it to the needs of its recipient. The same content can take the form of an annotated bibliography, a monitoring bulletin, an executive summary, or a concept map, depending on the target audience and intended use.
Application to writing a query — Always specify the expected output format. Do you want a five-point summary? A comparative table? A classification plan? A narrative summary with recommendations? Generative AI is flexible in its ability to structure information, provided you tell it how. Also indicate the desired length, tone (formal, popular, analytical), and structure (with or without subheadings, with or without recommendations).
Conclusion
Long before the emergence of generative AI, information science gave us the intellectual tools we need to understand context, formulate effective queries, and evaluate the responses we receive. Large language models are changing our interlocutor, but not the nature of information work. And this interlocutor, as powerful as it may be, remains entirely dependent on the quality of the inputs we provide, the questions we ask, and the intellectual work we do downstream, using its outputs.
At Cogniges, we are convinced that the fundamental principles of information science are the key to transforming these tools into levers and preventing certain pitfalls. The art of querying is a natural extension of our practices, which support you in an ethical approach to research and monitoring.
See also
- Seven principles for clearly stating your information needs
- Information sources: selection criteria for documentary research or evidence-based monitoring
- Artificial intelligence in the service of information science: points of convergence and limitations
References
[1] Blanquet, Marie-France. (2012). Un visionnaire venu des Indes : Shiyali Ramamrita Ranganathan. Bulletin des bibliothèques de France (BBF), 1, 12-17.
[2] Gaudet. Marie-Claude, Parent-Rocheleau, Xavier et Pasquier, Vincent. (2023, mai). La littératie permettrait-elle une meilleure gestion algorithmique? Revue Gestion.