Build a search brief an AI agent can use

A useful search brief tells an AI agent what you are deciding, what rules a role out, what you prefer and what it must verify. Ask for original job sources and unresolved questions. You want an inspectable shortlist, not a ranking you must trust.

Start with the decision

Write one sentence that describes the change you are considering:

I want to find out whether moving from analytics consulting into an in-house data role would give me more ownership without reducing my compensation.

This gives the search a purpose. “Find data jobs” does not.

Separate constraints, preferences and unknowns

Constraints make a role impossible or impractical. Examples include work authorisation, a hard location boundary, minimum workload or a required working arrangement.

Preferences help rank roles that remain possible. Examples include sector, company size, management responsibility and the kind of problems you want to solve.

Unknowns require research. An English advert, for example, does not establish that German or French is unnecessary. A “remote” label does not establish that you may work from any country.

Do not turn a preference into a hard filter just to shorten the list. You may hide a useful option before you understand it.

Copyable search brief

DECISION
I want to decide whether / find out if:

CURRENT POSITION
Relevant experience and strengths:
What I want to change:
What I want to keep:

HARD CONSTRAINTS
- Work location and travel boundary:
- Remote or on-site requirement:
- Workload:
- Work authorisation:
- Required compensation, if any:
- Other conditions that rule a role out:

PREFERENCES
- Work I want to do more of:
- Sectors or problems:
- Team or company environment:
- Tools or methods:
- Development goals:

EVIDENCE THE AGENT MUST RETURN
- Employer and exact role title
- Original employer posting URL
- Publication date when supplied by the source
- Date the source was checked
- Physical location and separate remote eligibility
- Explicit language requirements
- Reason for the match, tied to this brief
- Missing or conflicting information

OUTPUT
Return up to [number] verified roles in a comparison table.
Keep interesting employers without a verified current role in a separate list.
Do not infer missing facts. Do not give a fit score.

Filled example: analytics consultant

The person and requirements below are fictional.

DECISION
I want to decide whether an in-house data engineering role in the Zurich
area would give me ownership of production data systems while keeping a
four-day working week possible.

CURRENT POSITION
I have four years in analytics consulting. I build SQL transformations,
maintain Airflow workflows and run client workshops. I have supported cloud
migrations but have not owned production infrastructure or managed staff.

HARD CONSTRAINTS
- Workplace reachable from Winterthur in 60 minutes, or remote arrangement
  explicitly compatible with residence in Switzerland
- 80% workload must be stated or confirmed before a final application
- No more than two planned travel days per month
- Existing Swiss work authorisation

PREFERENCES
- Ownership of a production data pipeline
- Product, energy or public-service organisation
- Small data platform team with an experienced engineering lead
- SQL, Python, Airflow or comparable orchestration tools

EVIDENCE THE AGENT MUST RETURN
- Employer, role and original employer posting
- Date checked and any employer-supplied closing date
- Office location; remote eligibility as a separate field
- Workload and travel evidence
- Explicit language requirements
- Which supplied experience supports the match
- What remains unknown

OUTPUT
Return no more than eight verified roles. Use "unknown" instead of guessing.
Put employers with no verified vacancy in a separate research list.

The search can now consider adjacent titles without ignoring the conditions that matter.

Give the agent operating instructions

Add this prompt after your brief:

Use the brief below to research current roles. Treat hard constraints as
filters only when the source provides enough evidence. If a constraint is
unknown, keep the role separate for manual review rather than marking it as
a match.

For every role, open the employer-owned posting when possible. Return the
source URL and the date you checked it. Keep physical location, remote
eligibility, workload and language requirements in separate fields.

Explain each match using facts from the brief and the posting. Do not infer
that an English advert means no local language is required. Do not infer that
a Swiss employer permits work from anywhere. Do not invent compensation,
dates or responsibilities. Do not use a numerical fit score.

End with:
1. conflicting or missing facts;
2. questions the candidate should verify;
3. interesting employers found without a verified current vacancy.

[PASTE BRIEF]

Inspect the result before acting

For each shortlisted role, check:

1. Does the employer-owned page still exist?
2. Does the title and location match the agent’s summary?
3. Is “remote” defined, or only used as a label?
4. Are language and workload stated, inferred or unknown?
5. Does the match refer to experience you actually supplied?
6. Did the agent quietly discard or relax a hard constraint?

Keep the source beside the summary. If a fact is missing, leave it missing.

Improve the brief from real results

After the first search, change the brief only where the results exposed a problem:

• Too many impossible roles: clarify a genuine constraint.
• Too few useful roles: loosen a preference or add adjacent titles.
• Repeated unknowns: add a required evidence field.
• Superficial matches: describe the work you want, not only a title or tool.
• Good employers but no vacancy: keep a separate watch or research list.

A longer brief helps only when it changes the search or the evidence returned.

Common mistakes

• Asking for “the best jobs” without defining the decision.
• Mixing hard constraints with preferences.
• Filtering out every unknown as though it were a confirmed mismatch.
• Trusting summaries without opening original sources.
• Combining physical location and remote eligibility in one field.
• Treating an English advert as evidence about the working language.
• Asking for a fit score instead of reasons and gaps.
• Letting the agent invent candidate experience to improve a match.

Search-brief check

• ✓The decision and desired change are stated in one sentence
• ✓Hard constraints are genuinely disqualifying
• ✓Preferences help rank rather than silently filter
• ✓Unknowns are visible and assigned a next check
• ✓The output requires original sources and observation dates
• ✓Location, remote eligibility, workload and language are separate
• ✓Possible employers are kept separate from verified vacancies
• ✓Every match can be traced to the brief and current source

Next, use the company research worksheet to investigate the strongest employers, or the AI application workflow when you are ready to tailor factual material.