| Scope note: This article discusses UK data protection context, including the ICO’s work on automated decision-making in recruitment and the UK GDPR’s automated decision-making provisions (Articles 22A to 22D, as amended by the Data (Use and Access) Act 2025). It is practitioner commentary, not legal advice. For decisions about your own organisation’s processing, a qualified data protection specialist can advise you. |
When an HR lead shows me the prompt behind an AI-written job advert, it is usually one line long: “Write a job description for an operations manager.” What comes back is polished, generic and quietly risky. A wish list of untested requirements, a line about culture fit that invites bias, and nothing that would make the right person apply.
The weak link in that exchange sits on our side of the screen. Good ChatGPT prompts for HR do the opposite of that one-liner. They hand the tool your evidence, set its boundaries, and keep people decisions with people.
This matters most for smaller employers. The CIPD’s Resourcing and talent planning report 2024, a YouGov survey of 1,016 UK HR professionals in April 2024, found that nearly a third of organisations use some form of AI in recruitment, up from 16% in 2022. Around four-fifths of SMEs use none, and the CIPD links recent shifts, such as AI-written job descriptions, to the spread of free generative AI tools and to software vendors building generative AI into their products. For most businesses of 51 to 200 people, the first serious contact with AI in hiring will happen in a chat window long before anyone buys a platform.
So here is the playbook I give SME clients: ten prompts across the hiring and retention journey, each with a guardrail underneath. They work in ChatGPT and in other mainstream AI assistants. Adapt the prompts freely. Keep the guardrails.
Anything that identifies a real candidate or employee, unless your organisation has approved that tool for personal data.
That covers names, contact details, whole CVs, interview notes about a named person, sickness records and anything touching a protected characteristic. Role descriptions, anonymised themes and your own policies are generally safe starting material.
The ICO’s response to its generative AI consultation series sets out how UK data protection law applies to both the development and the use of these tools, so the familiar rules still apply when the interface feels friendly. My working rules for clients are simple. Use a business account your organisation has approved, never a personal login. Replace names with roles, such as “Candidate A” or “the new starter in finance”. Check whether your plan lets the provider train its models on your conversations. On ChatGPT’s personal plans that setting is on by default and can be switched off in Data Controls, while OpenAI’s business plans exclude your data from training by default. And if a prompt only works with personal data in it, take that question to whoever leads on data protection before you take it to ChatGPT.
The best job advert prompts give the tool your evidence first and ask it to write second.
Weak adverts come from asking AI to invent the role. Strong ones start with what you already know: the outcomes the job must deliver, the skills your best performer genuinely uses, and the realities of the working week. I cover the wider mechanics in my guide to using an AI job description writer, so here are the two prompts I rely on most.
Prompt 1: The evidence-first advert
You are helping a UK employer with [number] staff write a job advert for a [role]. The three outcomes this person must deliver in their first 12 months are: [outcomes]. The skills our strongest performer in this role uses are: [skills]. Write an advert of no more than 350 words in plain British English. Include the salary range [£X to £Y], working pattern and location. Limit essential requirements to five. Do not invent benefits, perks or company details I have not given you. After the advert, list separately any requirement that could exclude good candidates without being essential to the outcomes.
Guardrail: The tool can flag a risky requirement. Deciding whether it is justified stays with you.
Prompt 2: The realistic job preview
Using the advert above and only these facts, write a section of about 120 words called “What this role is really like”. One genuine challenge of the job: [challenge]. The pace of a typical week: [description]. Support in the first month: [support]. Keep the tone honest and warm rather than promotional, and add nothing I have not told you.
Guardrail: Supply the facts yourself. Left to fill gaps, AI tools invent plausible detail, and a preview that oversells creates the early resignations it was meant to prevent.
Very few employers do this. The same CIPD report found just 10% use tools that help candidates self-select, such as realistic job previews, while 52% of those recruiting reported more unsuitable applicants than the year before.
Use it to build the questions and the scoring guide, then keep every judgement about a candidate in human hands.
Structured interviews are one of the fairest tools an SME has, and preparing them is where AI earns its place. The CIPD found that 73% of organisations say line managers follow objective scoring criteria, yet only 28% train all interviewers on legal obligations and objective interview practice. A careful prompt narrows that gap in minutes.
Prompt 3: The structured question set
Create a structured interview for a [role] at a UK SME, based only on these four criteria from the job advert: [criteria]. For each criterion, write one behavioural question and one follow-up probe. Do not ask about any protected characteristic under the Equality Act 2010, such as age, disability, race (which includes nationality), religion or belief, sex, sexual orientation or pregnancy. Also avoid questions about health, family plans or culture fit.
Guardrail: Ask a second person to review the final set. AI can still produce questions that are leading or unfair in context.
Prompt 4: The scoring guide
For each question above, write a scoring guide from 1 to 4 describing what a weak, adequate, strong and exceptional answer would contain. Refer to observable evidence only, so that two interviewers would score the same answer the same way.
Guardrail: Do not paste a candidate’s answers back in and ask the tool to score them. That is the moment a drafting aid starts drifting towards an automated decision.
That guardrail is where the regulator is looking. In its Recruitment Rewired report, drawn from more than 30 employers between March 2025 and January 2026, the ICO’s key finding was that many employers were likely relying on solely automated decisions without meaningful human involvement. The ICO is also signalling that where human involvement exists, it should be applied consistently to every candidate at the same stage.
Yes, and onboarding is often where a prompt playbook pays back fastest, because the cost of a poor start shows up within weeks.
The CIPD found that 41% of organisations that selected candidates said new recruits always, mostly or sometimes resigned within the first 12 weeks. The causes I see are usually ordinary: no plan for week one, a manager too stretched to check in, and a role that differs from the one described at interview.
Prompt 5: The first-week plan
Draft a first-week plan for a new [role] joining a team of [number] at a UK company. They need to meet these people, listed by role: [roles]. Include one small piece of real work they can complete by Friday and two check-ins with their manager. Use only the roles I have listed, and present the plan as a simple day-by-day table.
Guardrail: Personalise the plan with the manager afterwards. Prompt with the role, never the new starter’s name.
Prompt 6: The 30-day conversation
Write five open questions a line manager can ask a new starter after 30 days to understand what is going well, what is confusing, and whether the role matches what was promised at interview.
Guardrail: The conversation is the point. Keep the answers between the manager and the new starter rather than feeding them back into the tool.
Use prompts to prepare better conversations and to find themes in anonymised feedback, never to profile individual employees.
Retention is where I see the most enthusiasm and the most risk. Summarising a set of anonymised survey comments is a sensible use. Asking a chatbot which named employee is likely to leave moves into profiling territory, and people are uneasy about it. In the ICO’s recruitment research, the public was particularly wary of profiling-based automation.
Prompt 7: The stay conversation
I am a line manager preparing a stay conversation with a valued team member in a [role]. Suggest six questions that uncover what keeps them here, what might tempt them away, and the one change that would most improve their work. Keep the tone informal and genuinely curious.
Guardrail: Prepare with the tool, then put it away. The conversation itself is entirely human.
Prompt 8: The feedback themes
Below are anonymised comments from our employee survey, with names and identifying details removed. Group them into no more than six themes, give an approximate count for each and quote one representative comment per theme word for word. Do not paraphrase inside quotation marks, and do not speculate about who wrote any comment.
Guardrail: Anonymise before you paste, and spot-check the counts, because AI tools miscount long lists. In a team of six, one comment about a specific project can identify its author.
Keep your approved prompts in one shared document, give it an owner, and review it every quarter.
A prompt library only works when everyone uses the same version. These two prompts keep it healthy.
Prompt 9: The policy explainer
Rewrite the following section of our [policy] in plain English for employees, keeping every obligation, deadline and exception exactly as written. Do not add, remove or soften anything. List anything you were unsure how to simplify.
Guardrail: Check the output against the original. The policy remains the source of truth.
Prompt 10: The prompt reviewer
Here is a prompt my team uses for [task]. Review it for anything that could introduce bias, invite personal data, or ask the tool to make a decision about a person. Suggest a safer version.
Guardrail: Run this on every new prompt before it enters the library.
Once these prompts are working, the bigger question is where AI sits across your whole hiring process. My step-by-step guide to implementing AI in hiring walks through that for SMEs. Start small, write the guardrails down, and make sure a person owns every decision that affects a person. That is how a 120-person business adopts AI with confidence.
For drafting adverts, interview questions, plans and policy summaries without personal data, it is a widely used and practical option. The risk rises when personal data goes in, or when a tool’s output starts deciding something about a person. A qualified data protection specialist can advise on your specific setup.
I advise against it unless your organisation has approved that specific tool for candidate data. CVs are personal data, and pasting them into an unapproved tool raises questions about where that data goes and how it is used.
I would not use a general chatbot to score, rank or reject candidates. The ICO’s recent recruitment work found many employers were likely making solely automated decisions, which is the territory of the ICO’s updated guidance on automated decision-making, now being finalised following consultation.
A general chatbot is a drafting and thinking aid. Recruitment software is built into a process and often acts on candidate data at scale, which changes the governance questions to ask before you buy.
Quarterly suits most SMEs, plus an extra review whenever your tools change or the ICO updates its guidance on automated decision-making.
Sabiha is a Talent Acquisition Director, speaker and author with 16+ years of international hiring experience across the UK, Dubai, South Africa and Malaysia. She has advised 300+ businesses on hiring and retention and was shortlisted for Best Career Coach UK by the CDI. Her book, How to Use AI to Win Talent and Retain People (Trotman), publishes in Autumn 2026.
Global Talent. Ethical AI. Strategic Hiring. Sustainable Retention.
