HQ is only as good as what you ask it. It reads your message and decides what to search for or what plan to propose. A specific request gets a specific result; a vague one makes HQ guess. This page is how to phrase a prompt so it gets it right the first time.
Specific beats vague
The single biggest lever. A vague prompt forces HQ to guess your criteria, and it guesses generically.
| Weak | Better |
|---|---|
| "Find good candidates." | "Find candidates with 5+ years of Python who applied to the Backend Engineer role." |
| "Show me the pipeline." | "How many candidates are in the interview stage across all open jobs?" |
| "Update this candidate." | "Set Priya Nair's source to 'Referral' and her position to 'Staff Engineer'." |
| "Schedule the interview." | "Create a scheduling request for the Backend Engineer first-round, for the three candidates in the screening stage." |
"Good" and "the interview" mean something to you and nothing to HQ. The fix is always the same: name the criteria, the role, the stage, the people.
Give it the criteria, not the conclusion
When you want a search, hand HQ the filters — not the verdict you've already reached. "Strong candidates" is a conclusion. "Candidates with a match rating above 80% who've had at least one Hire on their feedback" is criteria HQ can act on. Tell it what makes a candidate a match for this search, and it can find them.
Multi-step requests are fine
You don't have to break a task into one instruction per message. HQ handles a request with several steps and, for changes, rolls them into a single plan you approve once.
A reasonable multi-step ask:
"Find candidates for the Frontend role with React and TypeScript, add the top three to that job, and move them to the screening stage."
HQ runs the search itself, then proposes a plan: add three named candidates, move three stages. You see all six steps before anything happens. One prompt, one review.
Name records when you can
If you already know the candidate or job, say so by name. "Move Daniel Osei to offer on the Staff Designer role" leaves no ambiguity. "Move that candidate to the next stage" makes HQ work out who and which stage from the conversation — usually right, but "name it explicitly" is the safer habit. HQ never invents records, so a clear name is the cleanest path.
When to skip HQ and use the UI
HQ is not always the fast path, and a good prompt includes knowing when not to write one.
- A single drag is quicker than a sentence. Moving one candidate between two columns on the board beats describing it.
- Bulk visual work belongs in the table. Multi-selecting rows and editing in the candidates table is faster than narrating each change.
- Precise form work belongs in the form. Building a job's application questions is easier in the editor, where you see the whole thing.
HQ earns its keep on searches, questions that span many records, and multi-step changes. For one quick action, the click usually wins.
When it gets it wrong
If HQ misreads a request, it's nearly always the prompt — too vague, or a pronoun where a name belonged. Rephrase with concrete detail and try again. A long, wandering chat also makes HQ less sharp, because it has more to reconcile each turn; for a fresh task, start a fresh chat. And HQ's searches obey the same rules as the rest of Chosen — if one comes back empty, why a search returns no results is the place to look.