← Glean-ia-acs · Day 2 of 56
Keyword vs. Natural-Language Questions
Choosing how to ask
You'll be able to
- Decide when to search with keywords vs a full question
- Phrase a question so the assistant can answer it directly
Read first
Day 2 Lab
- Write one need as three keywords.
- Write the same need as a full question.
- Run both; compare what comes back — a list of docs vs a synthesized answer.
- Note which mode fit the need better, and why.
Working Example: Same Need, Two Modes
Glean accepts queries as keywords or as questions; the mode changes the shape of the result. Verified against search_basics.
Need: "What is our parental leave policy?"
Keyword mode: parental leave policy
→ a ranked list of documents (you open the benefits doc yourself)
Question mode: What is our parental leave policy?
→ a synthesized, cited answer drawn from the benefits docs
Rule of thumb:
Know the exact document → keywords (faster to the source)
Want the answer itself → a question (Glean assembles and cites it)Check for understanding
Give one need better served by keywords and one better served by a question.
Check yourself
1. What does Glean's Enterprise Graph primarily connect?
The graph models high-value entities and the network of signals — docs, tickets, messages — around each.
2. When is a natural-language question better than keywords?
Questions let Glean assemble and cite an answer; keywords are faster when you just need to open a known doc.
3. Your top result is the wrong TYPE of thing entirely. That is a...
Wrong type means the query expressed the wrong intent; fix it by reframing, not by tuning ranking.
4. What does a result's snippet tell you?
The snippet shows the matched text — why this result surfaced. Position signals rank confidence; metadata shows owner.
5. What does `type:document onboarding` do?
type: narrows by document kind; here, only documents (not messages or tickets) about onboarding.
6. `app:slack quarterly roadmap` returns...
app: scopes the query to one connected system — here, Slack.
7. In one sentence, what does `from:me` do?
from: finds documents associated with a person; from:me is the special case for yourself.
In 10 seconds
“Glean takes keywords or a full question. Keywords win when you know the exact doc; a question wins when you want a synthesized, cited answer. Match the mode to your need.”
Glean Foundations and Search Intent