Your search query is processed in a number of
steps, as shown in the image below.
What exactly happens in each
step?
1. The search query is
converted
When you enter a question, the AI
assistant sends it to a language model,
namely GPT-4.1 mini. This Large
Language
Model (LLM) This Large Language
Model (LLM) enriches your search query
with synonyms, related terms,
translations, and relevant background
knowledge on the subject. Based on this,
the model rewrites your question into a
search query that Limo can interpret
correctly. Read
here how this happens.
⚠️ Please note:
- The LLM does not have access to the most recent eventsGPT-4.1 mini does not have information published after June 1st, 2024. If you have questions about recent research topics, you will still receive an answer, but the model may be less able to enrich your question with current terms or insights.
- Content filteringA filter automatically blocks content that is considered offensive or harmful. If you ask a question that is detected as such, you may:
- o receive no answer,
or;
- o receive a notification
that the quality of the
answer may be
limited.
You can avoid this problem by formulating your search query in a concrete and detailed manner and using scientific terms. This helps the model understand that your question is academic in nature. - o receive no answer,
or;
2. The results are retrieved
The extensive search query is
executed in a sub-collection of Limo, and the
results are retrieved. Read here which
sources the tool uses .
3. The results are re-sorted
The 30 most relevant results are
re-analyzed. Based on semantic similarity
(embeddings), the tool selects five
sources that best match your search
query.
💡Good to know: You will not always see exactly the same five sources when you re-enter the same search query. See also the the info about "Try again".
4. An overview is generated
The abstracts of these five relevant sources are sent back to the
language model with the instruction to
create a brief overview, including
references to the sources that were
scanned.
5. The assistant formulates an answer
You will see a concise, informative
answer, along with the five sources that
were consulted and suggestions for related
search queries.
Last updated: 11/09/2025