GPT‑5 Cookbook for International Education

A practical guide for international education teams. A series of copy‑paste prompts, concrete examples, and clear guardrails.

---

What’s new with ChatGPT (GPT-5 and GPT-5.2)

How it differs from earlier ChatGPT models (quick view):

• Unified experience (GPT-5 family): GPT-5 blends fast chat and deep reasoning. For tough tasks, pick a GPT-5 Thinking option (on paid tiers) or say think this through.

• Better instruction-following: Sticks to tone, steps, and formats (tables and checklists) with less cleanup.

• Longer context handling: More reliable with multi-file handbooks, scholarship sheets, and FAQs.

• Clearer uncertainty: More likely to say "needs confirmation" instead of guessing.

• Improved localisation: Stronger with country-specific rules, dates, and plain-English rewrites.

What GPT-5.2 adds on top of GPT-5 / 5.1 (for you and your tech colleagues):

• More disciplined answers: GPT-5.2 is less verbose by default and keeps closer to the requested format (for example, 3–6 bullets, short emails, tables) which is useful for templates, checklists, and student-facing copy.

• Stronger instruction adherence: Better at sticking to your exact prompt, including sector-specific tone and sector rules, and at avoiding drift into off-topic content.

• Better structured reasoning: Improved performance on complex, multi-step tasks such as comparing handbooks, mapping TNE clauses, or building seminar plans.

• Smarter tool use: Works more reliably with web search, file uploads, and internal tools; better at grounding answers in documents instead of guessing.

• Configurable depth for developers: In the API, GPT-5.2 supports a reasoning_effort setting (for example none, low, medium, high) so technical teams can choose between faster responses and deeper reasoning for agents and workflows.

Previously selectable modes you may have seen: GPT-3.5, GPT-4, GPT-4o, GPT-4.1/4.1 mini, o1 (preview/mini), o3, o4-mini. These models still exist and some developers are still building tools with them until OpenAI deprecates them. It is possible that OpenAI might bring some back depending on user feedback.

Before you start (ChatGPT setup)

• Open the ChatGPT app (web or mobile) and choose GPT-5 (or GPT-5.2, if available) in the model picker (if available in your plan).

• For tougher or high-stakes tasks, select a GPT-5 Thinking option (if available), or simply add: think this through or explain your plan first.

• For faster replies, add quick answer or ask for a short paragraph or 3–5 bullets.

---

Tool & mode quick picks (from the beginner guide, adapted for IHE)

Use this as a “which feature when” map with ready prompts.

Tool / Mode	Plan*	What it’s good for in IHE	Paste-ready starter prompt
🔎 Search (Web browsing)	All plans	Live facts (deadlines, fees pages), regulator pages, partner news	Search the web and summarise the latest [Host Regulator] updates that affect [Program]. Link sources; if unclear, write ‘needs confirmation.’
🔬 Deep Research	Plus / Pro / Team / Enterprise / Edu	Structured literature reviews, strategy scans	Deep research: Do a literature review on [topic]; peer-reviewed 2015–2025; evidence table + 1-page synthesis; include DOIs.
🖼️ Vision (image input)	All plans	Check posters/booth layouts; read screenshots of policies	Review this fair poster for clarity and accessibility; suggest 3 improvements; draft alt-text.
📂 File uploads	All plans (higher limits on paid)	Summarise handbooks/FAQs; compare versions	Using these files—[Admissions Handbook 2025.pdf], [Scholarships 2025.xlsx]—answer: [question]. Cite file + section.
📊 Data Analysis	Plus / Pro / Team / Enterprise / Edu	Clean/aggregate CSV/Excel; simple charts/tables	From this CSV of applications, produce country-by-country offers, a bar chart, and 3 insights; export a clean CSV.
📝 Canvas (doc workspace)	Plus / Pro / Team / Enterprise / Edu	Drafting one-pagers, emails, briefs collaboratively	Create a 1-page pre-departure brief for [City] with sections: Etiquette, Transport, Scams, Packing, Scripts, Support.
📁 Projects	Plus / Pro / Team / Enterprise / Edu	Keep research notes & sources together over time	Start a project ‘International Students & Curriculum Enrichment’; store sources and running synthesis here.
⚙️ Custom Instructions	All plans	Set your team voice and default guardrails	When I ask about admissions, always cite file + section and prefer newest dated doc.
🧠 Memory (opt-in)	Availability varies by region/plan	Remember non-sensitive preferences (tone, time zone)	Remember my time zone is [X] and my tone is warm, plain English.
⏰ Scheduled tasks	Pro / Team / Enterprise (availability evolving)	Reminders (e.g., check scholarship page monthly)	Every month, check [URL] for scholarship updates and draft a 4-bullet summary.

*Plan notes: Availability and limits change over time and by region. If a feature doesn’t appear in your account, check the ChatGPT Help Center or your Admin.

Privacy tip: Use Memory only for non-sensitive preferences (tone, time zone). Never store student data in Memory—keep that in your CRM.

---

How to use this guide (2 minutes)

1. Pick a scenario below (e.g., “Admissions missing documents”).

2. Copy the prompt and replace the [brackets] with your details.

3. Paste into ChatGPT and review the draft it produces.

4. Tidy and send (or escalate to a colleague if it’s sensitive).

5. Save good outputs to your shared folder so your team can reuse them.

When this guide says “AI,” think “supercharged assistant that drafts first versions quickly.” You stay in control.

---

Key ideas

• AI Assistant: drafts answers, emails, lists, and checklists; you approve.

• Your knowledge base: the PDF/Word/Sheet files you already have (admissions rules, scholarship tables, FAQs). If you attach or paste them, the AI can quote and summarize them.

• Thinking style:

  • Quick (fast, good for routine) vs Thorough (slower, better for complex or risky).

• Answer length:

  • Brief (3–6 bullets or a short paragraph) vs Detailed (full email, plan, or 1-page summary).

• Escalate to a human for visas, mental health/welfare, academic appeals, or funding decisions.

---

Prompt switches (paste-in toggles)

What they are: One-line instructions you paste into your message to steer ChatGPT without touching any settings. Think of them as simple on/off toggles that control depth (Quick/Thorough), length (Brief/Detailed), tone, sourcing, and safety.

GPT-5.2 follows these switches more reliably than earlier models, especially for length and sticking to the requested format.

How to use them:

• Place at the start or end of your prompt (either works).

• They affect the current message only; paste again if you want the behavior to persist.

• You can combine several; if two conflict, the last one wins.

• When referencing files, use exact file names and sections so the assistant can cite correctly.

Add any of these lines to the top or bottom of a prompt:

Use only attachments

Use ONLY the attached files. If you cite a rule, name the file + section.

If unsure

If not in the files, write "needs confirmation" and list what to check.

Quick vs Thorough

Be quick / Be thorough. Plan briefly before answering.

Brief vs Detailed

Keep it brief (3–6 bullets or a short paragraph) / Give a detailed one-pager.

Plan first

Start with a 1-line plan, then answer.

Safety

If risk or crisis terms appear, draft a calm message, include the correct hotline for [Country] and our campus support contacts, and add a one-line internal hand-off note I can paste into [System] (summary + time). Do not provide clinical advice.

Tone

Warm, clear, student-friendly tone.

Localise

Use [Country] requirements and [Time Zone] deadlines.

Why bullets/paragraphs (not word counts)?

• Easier for staff and students—no counting needed.

• Consistent across languages and screen readers.

• Reduces cut-offs and padding; focuses on substance.

• Maps naturally to outputs (checklists vs short replies).

---

60-minute Quick Start (for a small pilot)

Goal: Get value today—no tech setup required.

1. Gather (10 min): Find 3–5 files you trust (e.g., “Admissions Handbook 2025”, “Scholarships 2025”, “Visa FAQ”, “Support Directory”).

2. Choose 1–2 scenarios (5 min): e.g., “Fair Q&A replies” and “Admissions missing docs emails.”

3. Copy the prompts from the scenarios below (15 min).

4. Test with real questions from last week’s inbox (20 min).

5. Decide next steps (10 min): keep, tweak, or expand to another scenario.

Success looks like: fewer back-and-forth emails, faster first drafts, fewer mistakes in requirements.

---

Scenario 1 — Recruitment fair Q&A (and follow-ups)

Use when: prospects ask about entry requirements, fees, scholarships, deadlines.

Prompt:

You are my Recruitment Fair Assistant.
Use ONLY: [Admissions Handbook 2025.pdf], [Scholarships 2025.xlsx], [FAQ.docx].
Question: "[paste the student’s question]" from [country].
Reply briefly (3–6 bullets or a short paragraph), warm and clear. Include:
- The exact requirement(s) for this student’s country (cite file + section)
- The next step with a link or attachment name
- 2 brief follow-up questions to keep the conversation going
If the answer isn’t in the attachments, say "needs confirmation" and list what to check.

What you’ll get: a short, accurate reply plus 2 follow-ups. Quality check: Does the answer name a source (file + section)? If not, mark needs confirmation.

---

Scenario 2 — Admissions “missing documents” email (humane tone)

Use when: an application is incomplete.

Prompt:

Write a warm email to [Applicant Name] about missing documents for the [Program] starting in [Month Year].
Include:
- A short friendly opener
- A bullet list of the exact missing documents (paste the list below)
- How to submit them: portal + reply-with-attachment
- The deadline in [applicant’s local timezone]
- Mention scholarships only if listed in [Scholarships 2025.xlsx] (cite sheet/row)
Keep it to a short email (6–8 sentences). End with an encouraging line.

Tip: Paste the list of missing items directly into the prompt.

---

Scenario 3 — Education Abroad pre-departure brief

Use when: preparing students for a destination.

Prompt:

Create a 1-page pre-departure brief for [City, Country] for [Discipline] students.
Include:
- Everyday etiquette and classroom norms (3 points)
- Transport, housing, and common scams (5 bullet tips)
- A packing checklist (10 items)
- 3 “what to say” scripts for tricky moments
- Links or titles of our campus support services
Add a "Wellbeing & escalation" box: signpost official university support; avoid diagnosis.
Write in plain language. Avoid stereotypes; focus on practical tips.

---

Scenario 4 — Student Support first-line triage

Use when: a student writes about wellbeing, housing, fees, or admin.

Prompt:

Classify this student message into {information | book appointment | emergency}. Then:
- If information: give a clear answer + a next-step link.
- If book appointment: draft a short reply with a booking link.
- If emergency keywords appear (self-harm, violence, immediate danger): write a calm, supportive message, include the correct hotline for [Country] and our campus support contacts, and advise immediate human escalation. Add a one-line hand-off note I can paste into [System] (summary + time)
Message: "[paste student message]"
Tone: empathetic, clear, and brief.
Rule of thumb: Anything safety-related → provide hotline + campus support info; do not give clinical advice; include an internal hand-off note template.

---

Scenario 5 — Transnational Education (TNE) compliance check

Use when: checking a partner program against host-country rules.

Prompt:

Role: Educational Compliance Expert.
Compare our program summary [paste or attach Program Spec] with [paste or attach Host Regulator Rules].
List in a simple table:
- Gaps by clause (what’s missing or mismatched)
- Severity (high/medium/low)
- Evidence we need
- Suggested fix wording (1–2 sentences each)
Cite both sources by clause. Flag uncertainty for human review.

Table headers: Clause | Gap | Severity | Evidence | Suggested Fix | Source(s)

---

Scenario 6 — Global engagement: shortlist partners + draft MoU outline

Use when: exploring new partnerships.

Prompt:

Create a shortlist of 10 potential partners in [Region] for our priorities: [themes, e.g., AI in healthcare; SDG 3; student mobility]. For each institution, provide:
- 2 research overlaps with our strengths
- 1 mobility idea
- Risks/considerations (1 line)
Then write a one-page MoU outline for the top 3 (scope, governance, data sharing, review cycle). Note any policy misalignments explicitly.
Keep it practical.

---

Scenario 7 — International employability coaching

Use when: turning global experiences into job-ready bullets.

Prompt:

Turn these student reflections into achievement-style CV bullets for roles in [Target Country/Industry]. Use action verb + task + impact. Tag each bullet with the USEM or CareerEDGE element it evidences. Then draft 3 interview stories in STAR format.
Reflections: [paste]
Add a note: "Please review and approve before sending".

---

Scenario 8 — International marketing: localized content pack

Use when: creating market-specific outreach.

Prompt:

Produce a content kit for [Market] about [Program]:
- 1 hero email (short and punchy—6–8 sentences, local tone and deadlines)
- 3 social posts in the style of [Platform], with appropriate emoji/hashtag norms
- 5 FAQs with exact scholarship lines and dates (cite sheet/row)
- UTM tags for links
- Alt-text suggestions for images
Segment by audience where helpful.

---

Scenario 9 — Teaching: culturally responsive, personalised seminar generator

Use when: you want a seminar that adapts to different cohorts (personalisation at scale).

Prompt:

Create a 60–90 minute seminar on [Topic] for [Module/Course], [Level], [Modality], class size [n].
Cohort segments: [e.g., international first-year (EAL); commuters; working professionals; neurodiverse; remote across time zones].

Deliver:
- 3 learning outcomes (plain language)
- 5-block session arc with timings (activate → explore → apply → reflect → wrap)
- Per-block adaptations per segment (one line each; asset-based; no stereotypes)
- 2 micro-assessments with model answers
- Academic integrity note (allowed AI use; declaration steps) aligned to [Policy] (cite if attached; otherwise write “needs confirmation” + checks)
- Accessibility checklist (alt-text, captions, reading order)

Output:
- **Table:** Block | Time | Activity | Purpose | Materials | Adaptations (by segment)
- Then 5 practical facilitation tips and 2 culturally relevant example cases (local + global).

Tone: warm, student-friendly, plain language.

---

Scenario 10 — Research: Deep Research (paid) and a free-tier alternative

Use when: conducting an academic literature review or evidence brief.

Activate Deep Research (paid plans)

• In a new chat, open Tools (🔧) and select Deep research. (If you use Projects, you can start Deep research inside a project.)

• Expect a multi-minute run with citations; ChatGPT may ask a short setup question first.

• Availability and monthly limits vary by plan; if you don’t see it, use the Free-tier alternative below.

Prompt — Deep Research ON (example you can paste)

Deep research: Conduct an academic literature review on “learning experiences of international students in the United States and how they enrich curriculum and pedagogy in higher education.”
Scope & rules:
- Peer-reviewed sources only, 2015–2025; prioritise systematic reviews/meta-analyses.
- Include key frameworks where relevant (e.g., internationalisation at home, intercultural competence, global citizenship).
- Deliverables: (1) Evidence map table (Study | Year | Context | Design | N | Findings | Limits); (2) 1-page synthesis (what’s known/unknown; implications for curriculum design); (3) 5 gaps & future research ideas; (4) Full citation list with DOI/URL.
- Do not speculate; if evidence is thin, write “needs confirmation” and list what to check.

Prompt — Free-tier alternative (no Deep Research)

Use web browsing to run a rapid scoping review on “learning experiences of international students in the United States and how they enrich curriculum and pedagogy in higher education.”
Rules to reduce errors:
- Search broadly but **include only peer-reviewed sources** (2015–2025). Prefer systematic reviews/meta-analyses; avoid blogs/forums.
- **No fabricated citations.** If you can’t verify a claim, write “needs confirmation.”
- Limit to **8–12** best sources; provide full citations with DOI/URL.
- Show a short inclusion/exclusion note and any disagreements between studies.
- Output: (1) Evidence table (Study | Year | Context | Design | N | Findings | Limits); (2) 6–8 bullet synthesis; (3) 3 implications for curriculum design.

---

Do / Don’t (cheat sheet)

What this is: A quick safety-and-quality checklist for you and a set of ready-to-paste lines you can drop into your prompt so ChatGPT follows the rules.

How to use it:

• Treat each Do/Don’t as a toggle you can paste verbatim (see “Paste-able switches” below).

• Put them at the start or end of your message. If you add several and they clash, the last one wins.

• Use them alongside the Prompt switches section for length/tone.

• Note: ChatGPT cannot escalate or log in your systems; it can only draft the message and a hand-off note for you to paste.

Paste-able switches

Do

Keep it short and clear.

Keep it brief (3–6 bullets or a short paragraph) in plain language.

Name your sources.

If you cite a rule, name the file + section.

Save reusable outputs.

Format the answer so I can paste it into our template (title + bullets + source).

Don’t

Don’t guess on visas or mental health.

If the question is about visas or welfare/mental health, write "for follow-up by [Team/Office]" and do not give advice; include our contact and hours.

Don’t include sensitive data.

Use only what’s necessary; remove IDs and bank details.

Don’t send without a human check.

Add a one-line Reviewer checklist at the end: key facts, links, and anything needing confirmation.

Mini examples

• Admissions reply (short + sourced):

Keep it brief (a short paragraph). Use only the attachments. If you cite a rule, name the file + section. If not in the files, write "needs confirmation" and list checks.

• Student support (safety first):

If distress or danger terms appear, write a calm message, include the correct hotline for [Country] and our campus support contacts, and advise immediate human escalation. Add a one-line hand-off note I can paste into [System] (summary + time).

---

Governance, privacy, and fairness (simple checklist)

• Data: Share only what’s necessary. Remove IDs and bank details. Follow your DPIA/records policy.

• Bias: Test content with diverse names and backgrounds. Avoid stereotypes.

• Accessibility: Use plain language, headings, and alt-text.

• Escalation: Define clear hand-offs for visas, welfare/mental health, appeals, and funding.

• Audit: Keep a small log of high-stakes AI outputs (date, purpose, who checked).

Footer block (paste under each scenario):

Safety & Governance: Use only attached docs; minimise personal data. If uncertain, write "needs confirmation" and list checks. Avoid stereotypes; add alt-text for images. Matters involving visas, welfare/mental health, appeals, or funding must be handled by human staff; include contact details in drafts.

---

Troubleshooting & quality control

• If the AI sounds unsure: It should say needs confirmation and list what to verify.

• Too long? Say make it a short paragraph or give me 5 bullets.

• Misses the point? Paste the exact sentence from your policy and say align to this.

• Sources conflict? Choose the newest dated document or escalate.

Ask the model: Score and improve your last answer using this rubric.

---

Quick scoring rubric (18 points)

• Instruction following (0–3)

• Factual accuracy (0–3)

• Cultural sensitivity (0–3)

• Safety (0–3)

• Actionability (0–3)

• Source use & citations (0–3) — names file + section; uses “needs confirmation” correctly.

---

Roll-out plan (30/60/90 days)

Days 0–30: Pilot 2 scenarios (e.g., fair Q&A + missing-docs emails). Create a tiny shared library of your best prompts and outputs. Days 31–60: Add student support triage + pre-departure briefs. Run a 2-hour staff workshop (live practice + do/don’t + escalation rules). Days 61–90: Extend to TNE checks + partner scouting + marketing kits. Add simple metrics: response time, satisfaction, and error rate.

---

Tiny prompt snippets you can copy today

Cite or say “needs confirmation”

Use only the attachments. If you quote a rule, name the file + section. If not in the KB, write “needs confirmation” and list what to check.

Role clarity

You are a Recruitment Fair Assistant. Keep answers brief (a short paragraph), warm, factual, and cite the KB.

Wellbeing handoff

If distress or danger keywords appear, include the correct hotline for [Country] and our campus support contacts, encourage contacting our support,  and add a one-line internal hand-off note I can paste into [System] (summary + time).

---

Appendix

Example prompt for a web research agent:

// You are a helpful, warm web research agent. Your job is to deeply and thoroughly research the web and provide long, detailed, comprehensive, well written, and well structured answers grounded in reliable sources. Your answers should be engaging, informative, concrete, and approachable. You MUST adhere perfectly to the guidelines below.

############################################
CORE MISSION
############################################
Answer the user’s question fully and helpfully, with enough evidence that a skeptical reader can trust it.
Never invent facts. If you can’t verify something, say so clearly and explain what you did find.
Default to being detailed and useful rather than short, unless the user explicitly asks for brevity.
Go one step further: after answering the direct question, add high-value adjacent material that supports the user’s underlying goal without drifting off-topic. Don’t just state conclusions—add an explanatory layer. When a claim matters, explain the underlying mechanism/causal chain (what causes it, what it affects, what usually gets misunderstood) in plain language.

############################################
PERSONA
############################################
You are the world’s greatest research assistant.
Engage warmly, enthusiastically, and honestly, while avoiding any ungrounded or sycophantic flattery.
Adopt whatever persona the user asks you to take.
Default tone: natural, conversational, and playful rather than formal or robotic, unless the subject matter requires seriousness.
Match the vibe of the request: for casual conversation lean supportive; for work/task-focused requests lean straightforward and helpful.

############################################
FACTUALITY AND ACCURACY (NON-NEGOTIABLE)
############################################
You MUST browse the web and include citations for all non-creative queries, unless:
The user explicitly tells you not to browse, OR
The request is purely creative and you are absolutely sure web research is unnecessary (example: “write a poem about flowers”).
If you are on the fence about whether browsing would help, you MUST browse.

You MUST browse for:
“Latest/current/today” or time-sensitive topics (news, politics, sports, prices, laws, schedules, product specs, rankings/records, office-holders).
Up-to-date or niche topics where details may have changed recently (weather, exchange rates, economic indicators, standards/regulations, software libraries that could be updated, scientific developments, cultural trends, recent media/entertainment developments).
Travel and trip planning (destinations, venues, logistics, hours, closures, booking constraints, safety changes).
Recommendations of any kind (because what exists, what’s good, what’s open, and what’s safe can change).
Generic/high-level topics (example: “what is an AI agent?” or “openai”) to ensure accuracy and current framing.
Navigational queries (finding a resource, site, official page, doc, definition, source-of-truth reference, etc.).
Any query containing a term you’re unsure about, suspect is a typo, or has ambiguous meaning.

For news queries, prioritize more recent events, and explicitly compare:
The publish date of each source, AND
The date the event happened (if different).

############################################
CITATIONS (REQUIRED)
############################################
When you use web info, you MUST include citations.
Place citations after each paragraph (or after a tight block of closely related sentences) that contains non-obvious web-derived claims.
Do not invent citations. If the user asked you not to browse, do not cite web sources.
Use multiple sources for key claims when possible, prioritizing primary sources and high-quality outlets.

############################################
HOW YOU RESEARCH
############################################
You must conduct deep research in order to provide a comprehensive and off-the-charts informative answer. Provide as much color around your answer as possible, and aim to surprise and delight the user with your effort, attention to detail, and nonobvious insights.

Start with multiple targeted searches. Use parallel searches when helpful. Do not ever rely on a single query.
Deeply and thoroughly research until you have sufficient information to give an accurate, comprehensive answer with strong supporting detail.

Begin broad enough to capture the main answer and the most likely interpretations.
Add targeted follow-up searches to fill gaps, resolve disagreements, or confirm the most important claims.
If the topic is time-sensitive, explicitly check for recent updates.
If the query implies comparisons, options, or recommendations, gather enough coverage to make the tradeoffs clear (not just a single source).
Keep iterating until additional searching is unlikely to materially change the answer or add meaningful missing detail.
If evidence is thin, keep searching rather than guessing.
If a source is a PDF and details depend on figures/tables, use PDF viewing/screenshot rather than guessing.

Only stop when all are true:
You answered the user’s actual question and every subpart.
You found concrete examples and high-value adjacent material.
You found sufficient sources for core claims.

############################################
WRITING GUIDELINES
############################################
Be direct: Start answering immediately.
Be comprehensive: Answer every part of the user’s query. Your answer should be very detailed and long unless the user request is extremely simplistic. If your response is long, include a short summary at the top.
Use simple language: full sentences, short words, concrete verbs, active voice, one main idea per sentence.
Avoid jargon or esoteric language unless the conversation unambiguously indicates the user is an expert.

Use readable formatting:
Use Markdown unless the user specifies otherwise.
Use plain-text section labels and bullets for scannability.
Use tables when the reader’s job is to compare or choose among options (when multiple items share attributes and a grid makes differences pop faster than prose).
Do NOT add potential follow-up questions or clarifying questions at the beginning or end of the response unless the user has explicitly asked for them.

############################################
REQUIRED “VALUE-ADD” BEHAVIOR (DETAIL/RICHNESS)
############################################
Concrete examples: You MUST provide concrete examples whenever helpful (named entities, mechanisms, case examples, specific numbers/dates, “how it works” detail). For queries that ask you to explain a topic, you can also occasionally include an analogy if it helps.
Do not be overly brief by default: even for straightforward questions, your response should include relevant, well-sourced material that makes the answer more useful (context, background, implications, notable details, comparisons, practical takeaways).
In general, provide additional well-researched material whenever it clearly helps the user’s goal.

Before you finalize, do a quick completeness pass:

Did I answer every subpart

Did each major section include explanation + at least one concrete detail/example when possible

Did I include tradeoffs/decision criteria where relevant

############################################
HANDLING AMBIGUITY (WITHOUT ASKING QUESTIONS)
############################################
Never ask clarifying or follow-up questions unless the user explicitly asks you to.
If the query is ambiguous, state your best-guess interpretation plainly, then comprehensively cover the most likely intent. If there are multiple most likely intents, then comprehensively cover each one (in this case you will end up needing to provide a full, long answer for each intent interpretation), rather than asking questions.

############################################
IF YOU CANNOT FULLY COMPLY WITH A REQUEST
############################################
Do not lead with a blunt refusal if you can safely provide something helpful immediately.
First deliver what you can (safe partial answers, verified material, or a closely related helpful alternative), then clearly state any limitations (policy limits, missing/behind-paywall data, unverifiable claims).
If something cannot be verified, say so plainly, explain what you did verify, what remains unknown, and the best next step to resolve it (without asking the user a question).

---