Two prompts make Future Student Experience feel like your institution. Everything else — searching, grounding, button URLs, page layout — is platform behaviour and should stay that way.

Edit prompts on Edit Prompts. Attach them on Configure Agents. See Editing a Prompt for substitutions and version history.

What to edit, and what to leave

Do not put apply URLs, fees, or “this course exists” claims in prompts. Those must come from the index and from CTA configuration.

Interview prompt (agent_future_student)

This is the concierge’s system prompt. The library copy is a good starting point; services often only change the opening name (for example “You are the My Murdoch Experience concierge”).

It should tell the agent to:

1. Run a short interview (about six questions, one at a time, no form).
2. Require at least an interest or a career goal before building a page.
3. Ask about home, travel, life stage, and constraints only when they would change the recommendation.
4. Ask about an entry score only if the institution prompt says you use one, and only as optional.
5. Answer side questions with search(), then return to the interview.
6. Use search_catalog() only to check that a named field is offered at all — never to browse or compare courses in chat.
7. Restate the profile, ask for consent, then call create_experience() once.
8. After the tool call, say that the page is being built and stop. The link arrives on its own.

Consent wording

Before create_experience(), the prompt requires the agent to say, in its own words:

• We will use what they told us to build a personalised page.
• The institution and Airgentic may keep it for a limited time to check quality.
• We will not sell it, and we will not put it in a sales system unless they later choose to get in touch or apply.

It must not promise immediate deletion, and it must not promise that no person will ever see it.

If they say no, the agent keeps helping in chat and does not start a job.

Substitutions you will see

The interview prompt deliberately does not include Query Building or catalog allowlists. Those belong to Study Pathways. Deep catalog work happens inside the job.

Role field (not the prompt file)

The agent’s Role on Configure Agents is what the Frontline agent reads when deciding whether to delegate.

Keep it narrower than Study Pathways. A working pattern:

• Do route here: “I don’t know what I want to study”; circumstances (work, family, distance, money, a false start) that should shape the answer.
• Do not route here: a choice between named courses; how to apply / fees / visas; finding a researcher; dated events; a message that already contains a name plus phone or email (that is Lead Gen).

Institution prompt (sub_fse_institution)

This is the prompt you should spend time on. It is attached to the concierge as a component and is also injected into the synthesis job.

It is orientation, not evidence. The first lines should say so. The agent may use it to ask better questions. It must not treat a course, campus offering, scholarship, or entry route as proven just because it appears here.

Write in prose, not YAML. Cover:

University example (shortened):

Murdoch uses a selection rank. It is not the same as a raw ATAR. There are entry routes that do not need a rank at all. Someone who says “I don’t have an ATAR” has not disqualified themselves — say so, and let the job find the actual routes.

Vocational example (shortened):

This is a vocational institute. Recruitment is local. Campus they can reach, travel time, and next intake matter more than almost anything else. There is no ranked school-leaver entry. Do not ask for ATAR. Free TAFE is a government scheme — never imply someone is eligible.

When you add a campus or change how entry works, update this file. Then test in Admin Chat with two or three realistic stories (school leaver, mature-age, “I don’t have a rank”).

Functions the interview may call

create_experience accepts the profile fields the concierge gathered (interests, home, mobility, constraints, optional entry score, optional email if they offered one) plus consent: true. Extra fields the model invents are dropped.

What the job writes (not a prompt you edit)

After create_experience runs, a separate synthesis pass searches your index and produces the page. It is instructed to:

• Use only retrieved pages for facts
• Leave disabled sections empty
• Pick CTA ids from your configured list
• Distinguish facts, recommendations, and hypotheses (for example funding eligibility)
• Never send the student to another institution
• Include an honest trade-off, not a brochure

If retrieval is thin, the page will be shorter and more cautious. That is preferable to invented courses. Fix coverage in the crawl and in sub_fse_institution, not by loosening grounding.

Testing prompts

1. Set the agent to Testing and use Admin Chat, or open a page with data-agent-id="future_student".
2. Walk through a full interview until the link arrives.
3. Confirm the page: your title, your buttons, no invented URLs, disabled sections absent.
4. Try a refusal of consent — no job should start.
5. Try a side question about fees — it should stay with this agent.
6. Try “which nursing course is best” — when the agent is Live, that should go to Study Pathways, not into a new interview.

Use the Trace Log on Admin Chat to see the profile passed into create_experience and which pages the job retrieved.

Related guides

• Future Student Experience
• Configuration
• Putting it on your website
• Editing a Prompt
• Editing an Agent