35 practical workflows

AI Use Case Library

Start with a real professional task. Each use case shows where AI can assist, what information is appropriate, what must stay out and what a person needs to check.

Risk is about the whole workflow

A green task can become amber or red if you add identifiable information, use an unapproved tool or allow the output to influence a consequential decision. Local policy always applies.

Search use cases

35 use cases

Frontline practice

Explain a technical drug concept

Turn an approved technical explanation into plain English for a defined audience.

Green

Suitable AI use: AI can suggest clearer wording, structure and examples while retaining the meaning of the source.
Information you can use: Public guidance, non-confidential source text, audience, purpose and preferred reading level.
Do not enter: Client details, unpublished clinical material or unverified health claims.

Example prompt

Using only the approved source text below, explain [CONCEPT] in calm, non-judgemental UK English for [AUDIENCE]. Explain technical terms immediately, preserve every safety warning and list any point that needs a specialist check. SOURCE: [PASTE PUBLIC OR APPROVED TEXT]

Risk level

Green — organisational policy still applies.

Human review

Check that the explanation is accurate, respectful and appropriate for the audience.

Fact-check

Compare health and drug-effect claims with the supplied authoritative source.

Frontline practice

Prepare a keywork session

Create a flexible session structure around an agreed topic without asking AI to assess a person.

Green

Suitable AI use: AI can organise timing, open questions, reflection and optional educational activities.
Information you can use: A generic topic, session length, broad learning aim and approved resources.
Do not enter: Case notes, risk information, names, distinctive histories or instructions about an individual’s care.

Example prompt

Create a [MINUTES]-minute keywork-session framework on [TOPIC] for a UK drug and alcohol service. Include a collaborative opening, open questions, one optional educational activity, reflection and a closing summary. Do not diagnose, assess risk or prescribe a response. Use fictional examples only.

Risk level

Green — organisational policy still applies.

Human review

Adapt the structure to the person, relationship, role and current situation; do not follow it mechanically.

Fact-check

Verify any health, treatment or legal information before using it.

Frontline practice

Generate reflective questions

Develop non-leading questions that support reflection on a practice topic.

Green

Suitable AI use: AI can provide varied question stems and identify potentially judgemental wording.
Information you can use: The topic, purpose, professional audience and fictional context.
Do not enter: A real person’s disclosures, safeguarding information or an instruction to determine motivation or truthfulness.

Example prompt

Write 12 open, non-leading reflective questions for [AUDIENCE] exploring [TOPIC]. Use UK English, avoid assumptions and moralising language, and include questions about strengths, context, uncertainty and next steps. Explain why any question may need sensitive facilitation.

Risk level

Green — organisational policy still applies.

Human review

Select only questions that fit the relationship, timing and professional purpose.

Fact-check

Check any embedded practice assumptions against current guidance.

Frontline practice

Create a psychoeducation outline

Draft an accessible educational outline from approved information.

Amber

Suitable AI use: AI can organise source-led material into short sections, questions and recap points.
Information you can use: Verified public sources, target audience, learning aim and accessibility requirements.
Do not enter: Personalised medical advice, dosing instructions, client histories or unsupported treatment claims.

Example prompt

Using only the verified material supplied, create a short psychoeducation outline about [TOPIC] for [AUDIENCE]. Include what it means, why it may happen, practical discussion points, uncertainty and where to seek appropriate help. Do not add new clinical claims. SOURCE: [PASTE VERIFIED MATERIAL]

Risk level

Amber — organisational policy still applies.

Human review

A suitably experienced practitioner must check safety, relevance and tone before use.

Fact-check

Trace each factual statement to the supplied source and check that it remains current.

Frontline practice

Plan a groupwork session

Structure a safe, participative group session around a defined educational aim.

Green

Suitable AI use: AI can suggest timing, fictional activities, accessible instructions and debrief questions.
Information you can use: Topic, audience, duration, delivery setting, learning aim and facilitation boundaries.
Do not enter: Real group disclosures, identifiable incidents or an assumption that participants will disclose personal experience.

Example prompt

Design a [MINUTES]-minute groupwork session on [TOPIC] for [AUDIENCE]. Include psychological-safety guidance, optional participation, a fictional activity, small-group questions, a factual recap and debrief. Avoid forced disclosure and flag content requiring specialist review.

Risk level

Green — organisational policy still applies.

Human review

Confirm accessibility, group safety, referral routes and facilitator competence.

Fact-check

Verify drug, health, legal and emergency content against current UK sources.

Frontline practice

Turn guidance into a practice checklist

Convert approved guidance into a usable reminder without replacing the full document.

Amber

Suitable AI use: AI can extract actions, responsibilities and escalation points from supplied text.
Information you can use: The current approved guidance, audience, setting and purpose of the checklist.
Do not enter: Unapproved internal guidance, client information or instructions to invent missing steps.

Example prompt

Turn the supplied approved guidance into a one-page practice checklist for [AUDIENCE]. Preserve exact thresholds, cautions and escalation instructions. Add source-section references and mark every unclear point [CHECK LOCALLY]. Do not fill gaps. GUIDANCE: [PASTE TEXT]

Risk level

Amber — organisational policy still applies.

Human review

The guidance owner must confirm completeness and approve the checklist before use.

Fact-check

Check every item against the original document, including exceptions and footnotes.

Frontline practice

Prepare questions for supervision

Structure reflective supervision around a professional theme rather than analyse a client.

Green

Suitable AI use: AI can generate questions about assumptions, boundaries, evidence and learning.
Information you can use: A general practice theme, role, supervision purpose and non-identifiable context.
Do not enter: Case details, names, safeguarding disclosures, staff allegations or a request for a risk judgement.

Example prompt

Create a reflective supervision question set for a [ROLE] exploring [GENERAL PRACTICE THEME]. Cover assumptions, emotional impact, professional boundaries, evidence, uncertainty, equality, strengths and learning. Do not analyse a real person or suggest a clinical or safeguarding decision.

Risk level

Green — organisational policy still applies.

Human review

The supervisee and supervisor decide what is relevant and follow established escalation processes.

Fact-check

Confirm any references to policy or professional standards locally.

Service management

Compare two policies

Identify similarities, differences and gaps between approved non-confidential policy documents.

Amber

Suitable AI use: AI can build a traceable comparison table and surface questions for policy owners.
Information you can use: Current approved documents, their dates, scope and the comparison questions.
Do not enter: Drafts under legal privilege, staff information, confidential incidents or instructions to decide compliance.

Example prompt

Compare the two supplied policies. Create a table covering scope, responsibilities, prohibited uses, approvals, information governance, safeguarding, review and incident response. Quote section numbers, label ambiguity and do not decide legal compliance. POLICIES: [ATTACH APPROVED DOCUMENTS]

Risk level

Amber — organisational policy still applies.

Human review

Policy, legal and information-governance owners must interpret differences and approve changes.

Fact-check

Check every comparison against the original wording and current external requirements.

Service management

Develop an SOP structure

Create a first-draft standard operating procedure structure for a defined process.

Amber

Suitable AI use: AI can organise purpose, roles, steps, exceptions, records and review points.
Information you can use: The approved process, roles, local requirements and relevant policy extracts.
Do not enter: Real client data, passwords, detailed security controls or invented legal and clinical requirements.

Example prompt

Create an SOP framework for [PROCESS] using only the approved information supplied. Include purpose, scope, roles, step-by-step process, exceptions, escalation, records, quality checks, training and review. Mark missing decisions [TO CONFIRM]. Do not invent thresholds or approvals.

Risk level

Amber — organisational policy still applies.

Human review

Operational, clinical, safeguarding and governance owners must approve relevant sections.

Fact-check

Verify all policy, regulatory and pathway references before issue.

Service management

Create a staff briefing

Turn an approved change or guidance update into a short, actionable briefing.

Green

Suitable AI use: AI can structure what changed, why, who is affected and what staff need to do.
Information you can use: Approved source material, audience, effective date, actions, owner and contact route.
Do not enter: Unannounced personnel information, speculation, protected documents or invented assurances.

Example prompt

Using only the approved material, draft a one-page staff briefing on [CHANGE]. Include what is changing, what is not changing, who is affected, required actions, effective date, support and questions to escalate. Keep facts and interpretation separate. SOURCE: [PASTE MATERIAL]

Risk level

Green — organisational policy still applies.

Human review

The accountable lead must confirm the message, timing, actions and contact details.

Fact-check

Verify dates, responsibilities and quoted requirements against the approved source.

Service management

Analyse anonymised consultation themes

Organise properly anonymised feedback into themes without claiming statistical significance.

Amber

Suitable AI use: AI can cluster comments, identify divergent views and create a transparent coding draft.
Information you can use: De-identified comments approved for processing, purpose, coding questions and known limitations.
Do not enter: Names, quotations that identify someone, small-cell data, allegations or raw sensitive disclosures.

Example prompt

Analyse the anonymised consultation text using a transparent thematic approach. Propose themes, supporting de-identified examples, minority views, contradictions and limitations. Do not infer prevalence, motive or representativeness. TEXT: [PASTE APPROVED ANONYMISED MATERIAL]

Risk level

Amber — organisational policy still applies.

Human review

Review the coding against the original material and involve lived-experience contributors appropriately.

Fact-check

Confirm counts and quotations manually; describe the method and limits accurately.

Service management

Structure an implementation plan

Turn an approved change into a staged plan with ownership, dependencies and review.

Green

Suitable AI use: AI can organise tasks, milestones, risks, communications and evaluation questions.
Information you can use: Agreed objective, constraints, responsible roles, dependencies, timescale and approved decisions.
Do not enter: Passwords, sensitive security details, assumed resources or invented approval and completion dates.

Example prompt

Create an implementation-plan structure for [AGREED CHANGE]. Include outcomes, scope, workstreams, dependencies, responsible roles, milestones, risks, communications, training, service-user involvement, measures and review. Mark unconfirmed information clearly and do not assign decisions to unnamed people.

Risk level

Green — organisational policy still applies.

Human review

Owners must agree feasibility, resources, dates, risks and governance before adoption.

Fact-check

Check every dependency, requirement and milestone against approved plans.

Service management

Generate audit questions

Develop draft assurance questions from an approved standard or policy.

Amber

Suitable AI use: AI can translate requirements into evidence-seeking questions and identify possible evidence types.
Information you can use: The authoritative standard, audit scope, audience and review period.
Do not enter: Identifiable records, unsupported compliance claims or instructions to produce a pass/fail judgement.

Example prompt

Using the supplied standard, draft audit questions for [SCOPE]. For each question state the source requirement, evidence to examine and follow-up question. Distinguish design, implementation and outcome. Do not decide compliance. SOURCE: [PASTE APPROVED STANDARD]

Risk level

Amber — organisational policy still applies.

Human review

The audit lead must validate scope, sampling, evidence and rating method.

Fact-check

Map every question back to the exact source clause.

Service management

Improve public-facing communication

Make approved service information clearer and more accessible.

Green

Suitable AI use: AI can improve headings, sentence length, reading level and call-to-action clarity.
Information you can use: Current approved copy, audience, required facts, accessibility needs and service contact route.
Do not enter: Unpublished client information, unverified promises, changed eligibility or invented contact details.

Example prompt

Rewrite the approved service information for [AUDIENCE] in plain UK English. Preserve eligibility, exclusions, urgent guidance and contact details exactly. Improve structure and explain jargon. List every factual point that needs final confirmation. COPY: [PASTE APPROVED TEXT]

Risk level

Green — organisational policy still applies.

Human review

Service and communications leads must confirm accuracy, accessibility and tone.

Fact-check

Test all links, contacts, opening information and access criteria.

Commissioning & public health

Structure a needs assessment

Build a transparent question and evidence framework for a local needs assessment.

Amber

Suitable AI use: AI can organise domains, datasets, engagement questions and evidence gaps.
Information you can use: Geography, population, decision context, public datasets and approved non-identifiable local information.
Do not enter: Small-cell data, unpublished intelligence, identifiable histories or invented local findings.

Example prompt

Create a needs-assessment framework for [PLACE / POPULATION / DECISION]. Cover prevalence, harm, treatment access, outcomes, inequality, unmet need and lived experience. Separate national evidence, local data and gaps. Do not invent local conclusions or imply causation.

Risk level

Amber — organisational policy still applies.

Human review

Analysts, commissioners, providers and lived-experience contributors must test the framework.

Fact-check

Confirm dataset definitions, coverage, dates and comparability with original publishers.

Commissioning & public health

Interrogate JSNA information

Generate questions that test what a Joint Strategic Needs Assessment can and cannot support.

Green

Suitable AI use: AI can identify definitions, time periods, denominators, comparison groups and missing voices.
Information you can use: Published JSNA extracts, public datasets, decision context and the questions being considered.
Do not enter: Restricted data, small numbers or assumptions presented as established local need.

Example prompt

Review the supplied public JSNA material and create critical questions about definitions, data age, coverage, trends, inequalities, treatment penetration, comparators and gaps. Separate what the material states from what it cannot establish. MATERIAL: [PASTE PUBLIC TEXT]

Risk level

Green — organisational policy still applies.

Human review

A public-health analyst should confirm interpretation and identify relevant local context.

Fact-check

Return to the original datasets and technical notes for every numerical claim.

Commissioning & public health

Compare national guidance

Compare current authoritative documents on a defined commissioning question.

Amber

Suitable AI use: AI can trace agreement, difference, audience, scope and publication date across supplied sources.
Information you can use: Current public guidance documents, comparison question and intended decision audience.
Do not enter: AI-generated citations, unverified summaries or a request to choose policy on the commissioner’s behalf.

Example prompt

Compare the supplied authoritative guidance on [QUESTION]. Use a table for scope, population, recommendation, evidence basis, uncertainty and implementation. Cite page or section numbers and identify conflicts without resolving them. SOURCES: [ATTACH DOCUMENTS]

Risk level

Amber — organisational policy still applies.

Human review

Commissioning and subject leads must interpret applicability and resolve conflicts.

Fact-check

Open every source and confirm that each attributed recommendation is accurate and current.

Commissioning & public health

Create stakeholder consultation questions

Develop inclusive questions for a planned consultation about a service or strategy.

Green

Suitable AI use: AI can suggest neutral questions, formats and prompts for different stakeholder groups.
Information you can use: Consultation purpose, decision scope, participant groups, accessibility needs and known constraints.
Do not enter: Named views, confidential disputes, assumed community positions or leading questions designed to validate a decision.

Example prompt

Draft a neutral consultation framework for [DECISION / SERVICE]. Include core questions, tailored prompts for different groups, accessible formats, equality considerations, how disagreement will be recorded and what is outside scope. Do not assume a preferred outcome.

Risk level

Green — organisational policy still applies.

Human review

Stakeholders and lived-experience partners should review wording, access and power dynamics.

Fact-check

Confirm the decision scope, constraints and consultation process before launch.

Commissioning & public health

Structure a service specification

Create a first-draft specification framework from approved evidence and local decisions.

Amber

Suitable AI use: AI can organise scope, pathways, quality, outcomes, workforce, safeguarding and monitoring.
Information you can use: Verified needs evidence, agreed scope, standards, pathways, intended outcomes and contract constraints.
Do not enter: Confidential bids, invented targets, identifiable data or unsupported legal and clinical requirements.

Example prompt

Create a service-specification structure for [SERVICE]. Include purpose, population, inequalities, scope, access, pathways, workforce, safeguarding, quality, outcomes, reporting, lived-experience involvement and review. Mark missing local decisions [TO CONFIRM].

Risk level

Amber — organisational policy still applies.

Human review

Commissioning, clinical, operational, finance, legal and lived-experience reviewers must approve relevant sections.

Fact-check

Verify standards, measures, legal references and service assumptions.

Commissioning & public health

Develop evaluation questions

Create a balanced evaluation framework for an agreed intervention or service change.

Green

Suitable AI use: AI can organise process, outcome, experience, equality, cost and unintended-effect questions.
Information you can use: Intervention logic, intended outcomes, stakeholders, available measures and constraints.
Do not enter: Invented baseline data, causal conclusions, identifiable responses or measures selected only because they are easy.

Example prompt

Create evaluation questions for [INTERVENTION]. Cover reach, implementation, outcomes, experience, inequalities, costs, unintended effects and sustainability. For each question suggest evidence needed and limitations. Do not infer effectiveness or invent measures.

Risk level

Green — organisational policy still applies.

Human review

Evaluation leads and affected communities should agree questions, methods and interpretation.

Fact-check

Confirm definitions, validated measures and baseline sources.

Commissioning & public health

Summarise drug-trend intelligence

Turn verified intelligence into a cautious briefing that separates signal from established trend.

Amber

Suitable AI use: AI can organise sources, dates, geography, confidence and operational questions.
Information you can use: Approved, non-identifiable intelligence; public reports; dates; geography and source limitations.
Do not enter: Operationally sensitive intelligence, identifiable reports, rumours or claims with no source trail.

Example prompt

Using only the supplied verified material, draft a drug-trend briefing covering signal, sources, geography, timeframe, confidence, uncertainty, possible implications and questions for local verification. Do not turn a report into a confirmed trend. MATERIAL: [PASTE APPROVED CONTENT]

Risk level

Amber — organisational policy still applies.

Human review

An accountable intelligence or public-health lead must confirm interpretation and circulation.

Fact-check

Check every date, substance, geography and numerical claim in the originating source.

Training & workforce development

Create a fictional case study

Develop a balanced fictional scenario tied to a clear learning outcome.

Green

Suitable AI use: AI can vary context, ambiguity, strengths and discussion points without using a real case.
Information you can use: Learning outcome, audience, setting, difficulty, duration and content boundaries.
Do not enter: Details taken from a real person, rare combinations that enable recognition or stereotypes.

Example prompt

Create a clearly fictional UK drug and alcohol practice scenario for [AUDIENCE] exploring [OUTCOME]. Include strengths, social context, uncertainty and more than one defensible response. Avoid stereotypes. Add five questions and facilitator notes.

Risk level

Green — organisational policy still applies.

Human review

Check realism, bias, psychological safety and fit with participants’ responsibilities.

Fact-check

Verify all drug, treatment, law and safeguarding details.

Training & workforce development

Develop a training exercise

Create a timed participative activity around a defined learning outcome.

Green

Suitable AI use: AI can suggest instructions, group tasks, debrief questions and facilitator prompts.
Information you can use: Audience, outcome, time, group size, delivery method and verified content.
Do not enter: Real disclosures, forced personal reflection or unsupported facts presented as teaching points.

Example prompt

Design a [MINUTES]-minute exercise for [AUDIENCE] on [TOPIC]. Include objective, preparation, instructions, fictional material, discussion, debrief and an answer guide. Use a trauma-aware approach and flag facts requiring verification.

Risk level

Green — organisational policy still applies.

Human review

A trainer must check feasibility, inclusion, accuracy and emotional safety.

Fact-check

Verify teaching points against current authoritative sources.

Training & workforce development

Generate a knowledge-check quiz

Draft a formative quiz from supplied approved learning material.

Green

Suitable AI use: AI can create varied question formats and feedback linked to the source.
Information you can use: Approved source, audience, learning outcomes, number and type of questions.
Do not enter: Unsupported facts, trick questions, ambiguous clinical scenarios or a claim that the quiz proves competence.

Example prompt

Using only the supplied material, create a [NUMBER]-question formative quiz for [AUDIENCE]. Include answers, explanations and source-section references. Avoid trick wording and mark ambiguity rather than inventing certainty. SOURCE: [PASTE TEXT]

Risk level

Green — organisational policy still applies.

Human review

Check alignment, accessibility, unambiguous answers and the limits of what the quiz measures.

Fact-check

Confirm every answer in the approved source.

Training & workforce development

Draft facilitator notes

Create concise notes that help a trainer deliver approved content consistently.

Green

Suitable AI use: AI can organise timings, key messages, questions, likely misconceptions and transitions.
Information you can use: Approved session content, slides, audience, timing and escalation arrangements.
Do not enter: Invented answers, unsupported disclosure advice or assumptions about participant experience.

Example prompt

Create facilitator notes for the supplied session. For each section include timing, purpose, key message, question, likely misconception, evidence check and transition. Add guidance for optional participation and handling questions outside scope. MATERIAL: [PASTE APPROVED CONTENT]

Risk level

Green — organisational policy still applies.

Human review

The trainer must adapt delivery and confirm referral, safeguarding and support routes.

Fact-check

Trace teaching claims to current sources and approved materials.

Training & workforce development

Draft learning outcomes

Turn a training purpose into observable, realistic learning outcomes.

Green

Suitable AI use: AI can suggest clear verbs and check whether outcomes match time and audience.
Information you can use: Purpose, audience, starting level, duration and intended practice change.
Do not enter: Claims that attendance creates clinical competence or outcomes beyond the training scope.

Example prompt

Draft 4–6 realistic learning outcomes for [SESSION] aimed at [AUDIENCE] over [DURATION]. Use observable verbs, avoid claiming competence beyond the session and explain how each outcome could be checked formatively.

Risk level

Green — organisational policy still applies.

Human review

The course owner must confirm scope, level and assessment approach.

Fact-check

Check alignment with any cited capability or professional framework.

Training & workforce development

Create a reflective learning activity

Develop an activity that helps practitioners examine judgement and uncertainty.

Green

Suitable AI use: AI can create fictional prompts, alternative perspectives and a structured debrief.
Information you can use: Topic, audience, outcome, time and facilitation safeguards.
Do not enter: Required disclosure of personal trauma, a real case or diagnostic interpretation of participants.

Example prompt

Create a reflective activity for [AUDIENCE] on [TOPIC]. Use a fictional stimulus and invite participants to examine assumptions, uncertainty, power, evidence and alternative perspectives. Include opt-out wording and facilitator debrief notes.

Risk level

Green — organisational policy still applies.

Human review

Check that participation is safe, voluntary and appropriate to the group.

Fact-check

Verify any practice or policy statements included in the stimulus.

Training & workforce development

Build a CPD learning plan

Structure a role-relevant continuing professional development plan.

Green

Suitable AI use: AI can organise objectives, activities, evidence of learning and review questions.
Information you can use: Role, development goal, available approved resources, timescale and supervision arrangements.
Do not enter: Performance judgements, sensitive staff records or claims that suggested activities meet formal requirements.

Example prompt

Create a draft CPD plan for a [ROLE] developing [CAPABILITY]. Include objectives, learning activities, practice application, reflection, evidence and review points. Distinguish suggestions from formal professional requirements.

Risk level

Green — organisational policy still applies.

Human review

The worker and manager or supervisor must agree relevance, support and formal requirements.

Fact-check

Confirm any accreditation or professional-body requirements directly.

Research & service development

Structure a literature search

Turn a broad topic into searchable concepts and a transparent search plan.

Green

Suitable AI use: AI can develop synonyms, inclusion questions and a draft search string for testing.
Information you can use: Research question, population, intervention or exposure, outcomes, setting and date range.
Do not enter: A claim that AI has searched databases it cannot access or invented citations.

Example prompt

Help structure a literature search on [QUESTION]. Break it into concepts, UK and international terminology, synonyms, inclusion and exclusion criteria, databases to consider and draft Boolean search strings. State what requires librarian or researcher review.

Risk level

Green — organisational policy still applies.

Human review

A researcher or information specialist must test and document the final strategy.

Fact-check

Run searches in the named databases and retain the actual search record.

Research & service development

Summarise a research paper

Produce a structured summary of a supplied paper without overstating its findings.

Amber

Suitable AI use: AI can extract study design, sample, results, limitations and relevance from the document.
Information you can use: The full paper, research question, intended audience and required detail.
Do not enter: A citation alone, confidential manuscripts or permission-restricted material.

Example prompt

Summarise the supplied paper under: question, design, sample, setting, intervention or exposure, outcomes, main findings, uncertainty, limitations, conflicts and UK practice relevance. Use only the paper and provide page references. PAPER: [ATTACH APPROVED DOCUMENT]

Risk level

Amber — organisational policy still applies.

Human review

Read the methods, results and limitations directly; do not rely on the summary alone.

Fact-check

Check every number and interpretation in the original paper.

Research & service development

Compare research papers

Compare studies without flattening differences in design, population or quality.

Amber

Suitable AI use: AI can create an evidence table and identify where apparent disagreement may reflect different methods.
Information you can use: Full papers, comparison question and criteria agreed in advance.
Do not enter: AI-generated abstracts, missing full text or a request to calculate an unsupported pooled effect.

Example prompt

Compare the supplied papers on [QUESTION]. Tabulate design, sample, setting, measures, findings, uncertainty, limitations and applicability. Explain where studies cannot be compared directly. Do not rank quality without an agreed appraisal method.

Risk level

Amber — organisational policy still applies.

Human review

A competent reviewer must appraise methods and decide relevance.

Fact-check

Verify each extracted field and quotation in the full paper.

Research & service development

Develop research questions

Refine a service-development topic into answerable research or evaluation questions.

Green

Suitable AI use: AI can test scope, definitions, populations, outcomes and feasibility.
Information you can use: Topic, decision need, existing evidence, setting, constraints and stakeholder priorities.
Do not enter: Assumed causation, a predetermined answer or invented evidence gaps.

Example prompt

Develop candidate research questions for [TOPIC / DECISION]. For each, define population, concept or intervention, comparator where relevant, outcomes, setting, feasibility, ethical issues and decision value. Identify assumptions and evidence needed before selection.

Risk level

Green — organisational policy still applies.

Human review

Researchers, practitioners and affected communities should select and refine the final question.

Fact-check

Run a scoping search to confirm the claimed gap and existing measures.

Research & service development

Create an evidence table

Extract consistent fields from a defined set of supplied studies.

Amber

Suitable AI use: AI can draft a structured table and flag missing information.
Information you can use: Full papers, extraction fields, review question and agreed definitions.
Do not enter: Invented fields, inferred data not reported by authors or a request to conceal uncertainty.

Example prompt

Extract the agreed fields from each supplied study into an evidence table: citation, design, setting, sample, intervention or exposure, comparator, outcomes, results, uncertainty, limitations and applicability. Use NR for not reported and add page references.

Risk level

Amber — organisational policy still applies.

Human review

Use duplicate or sampled checking appropriate to the importance of the review.

Fact-check

Confirm all extracted data and calculations in the original papers.

Research & service development

Draft survey questions

Create a first-draft survey aligned with a defined evaluation question.

Amber

Suitable AI use: AI can suggest neutral wording, response options, routing and accessibility checks.
Information you can use: Purpose, population, constructs, mode, length, safeguarding and analysis plan.
Do not enter: Unnecessary personal data, intrusive questions without purpose or claims that unvalidated items form a validated scale.

Example prompt

Draft a short survey for [PURPOSE / POPULATION]. Map each question to an evaluation aim, use neutral UK English, include appropriate response options and prefer not to say where needed. Flag sensitive items, accessibility needs and any construct requiring a validated measure.

Risk level

Amber — organisational policy still applies.

Human review

Pilot with intended users and obtain ethical, governance or data-protection review where required.

Fact-check

Confirm permissions and scoring for any validated measure used.

Research & service development

Create an interview schedule

Develop open questions and prompts for an approved qualitative interview.

Amber

Suitable AI use: AI can organise a logical, non-leading schedule and suggest optional probes.
Information you can use: Research aim, participant group, duration, consent, distress and referral arrangements.
Do not enter: Participant data, coercive wording, assumptions about lived experience or an unapproved research purpose.

Example prompt

Create a semi-structured interview schedule for [AIM / PARTICIPANTS]. Include a plain introduction, open questions, optional probes, strengths, inequalities, unintended effects, closing and distress protocol prompts. Avoid leading language and forced disclosure.

Risk level

Amber — organisational policy still applies.

Human review

Researchers and lived-experience advisers must review ethics, language, burden and support.

Fact-check

Confirm the final schedule matches approved protocol, consent and governance documents.