How Search Works in Keenious

When you search in Keenious, you’re not typing keywords into a database in the traditional sense. Instead, you describe what you need in natural language—just as you might explain your research question to a librarian. From that point forward, the system interprets your intent and forms a structured academic search on your behalf.

We aim to be transparent about how this works so you can trust the results, steer the process, and understand why certain papers appear higher than others—without having to learn complex search syntax.

When you describe your information need—whether it’s a full question, a short description, or a highlighted passage—the system analyzes it with advanced language understanding (Google’s Gemini). Rather than lifting keywords, it focuses on intent and scope.

It considers the main topic or theme; whether you’re exploring broadly or seeking something specific; any implied preferences such as recency, open access, or peer review. It also draws on the broader context of your session so results reflect what you’re actually working on. That context can include what you typed in your prompt and follow-ups, what’s visible in the content frame (a paper or a result list), any text you’ve highlighted, your document in Word or Google Doc if you are using the addon, and PDFs you upload.

Once your information need is understood, the AI turns it into a structured representation that our search engine can execute. If your description contains essential terms—like a specific location, instrument, or software—the system can treat those as required conditions so results are both conceptually relevant and appropriately precise.

Keenious searches metadata sourced from OpenAlex. This includes titles, abstracts, citation information and publication data, and identifiers; we do not search full text.

Two complementary methods are used to identify relevant research:

This works like a traditional academic database. It looks for exact or closely related wording in article metadata. Matches in titles are strong relevance signals, followed by matches in the abstract if it is available.

Semantic search identifies research based on meaning rather than exact wording. Your information need and millions of articles are represented in a shared conceptual space in the form of embeddings, allowing the system to retrieve work that discusses the same ideas even when terminology differs. For example, a search for “climate anxiety in youth” may also surface work referring to “eco-anxiety” or “environmental stress responses in adolescents.”

You can continue writing in the language you’re most comfortable with while asking Keenious to search for articles in another language. When you do this, the system retrieves in the target language you specify and returns results in that language, while explanations and summaries remain in your chat language. Searches run one language at a time to keep results interpretable; if you want to broaden across languages, simply ask Keenious to search for articles in that language and it will perform a search in your requested language, while keeping the conversation language the same.

Keyword evidence and semantic evidence are blended into a single ranking. When your description is short or broad, the system leans more on semantic similarity. When your description is specific or technical, it places greater emphasis on exact textual matches. Papers that are strong on both axes—matching your wording and your underlying idea—tend to rise to the top.

After the initial relevance ranking, Keenious applies clear scholarly cues to help you notice research that is both relevant and credible:

These adjustments are modest and do not override relevance; they fine-tune ordering so meaningful, trustworthy research is easier to spot.

The results are presented as a ranked list with clear metadata so you can quickly assess relevance and open individual records for details. From there, you can guide the system in natural language—such as asking it to narrow to time period, language, or broaden to adjacent areas—or you can adjust visible filters directly in the interface. Active filters are always shown so it’s clear what is shaping your results, and retracted items are excluded automatically.

After each search, Keenious also proposes next steps. These may be focused refinements (for example, narrowing by method or period) or suggestions to explore nearby topics that commonly co-occur with your subject. The aim is to make searching an active, engaging process in which you learn the landscape while discovering sources.

There is no need to rebuild the query each time. Because the system keeps track of the structured representation it used, you can simply state what should change, and it will update the search accordingly.

You explain your information need in plain language. The system interprets it and forms a structured academic search. Keenious then queries OpenAlex metadata using both keyword-based and semantic search to find work aligned with your topic and intent. Results are ranked primarily by relevance and refined with clear scholarly cues. You refine as you go—by speaking naturally, adjusting filters, or following suggestions—to reach useful, trustworthy sources with minimal friction.