AI in recruitment is most useful in five places: sorting high volumes of CVs, finding specialised candidates, matching anonymised CVs to specialised roles, preparing interview questions, and defining a role before you hire. Use a paid AI plan that does not train on your data, remove personal details from CVs, and keep a person responsible for every hiring decision.
Most writing about AI in recruitment is about tools: which platform screens CVs fastest, which chatbot books interviews. This piece is about the uses I actually rely on, from the flood of CVs that follows a job advert to the thinking that should happen before the advert is written.
There are five, and I have set them out in the order they appear in a hiring process, with the practical detail you need to try them. The last one happens before the process even starts, and it is the one most businesses skip.
Place a role for a packer, a cashier, a factory worker, or a call centre agent on Indeed, PNet, or LinkedIn, and you will be flooded with hundreds, if not thousands, of CVs in the first couple of days. In my experience Indeed and PNet are the more popular of those for this kind of role. Reading every one of them properly is not realistic for most HR teams.
This is one of the best uses of AI in recruitment. You can sort the pile by the things that matter for the role:
The output is a sorted list that puts the most relevant candidates at the top of the shortlist. A person still reads that shortlist and makes the call. The value is that your recruiter spends their time on the candidates most likely to fit, and gets to them within days of the advert going live.
Be deliberate about the rules you give the tool. Sorting by experience is sensible for some roles and unnecessary for others, and a first-time job seeker with no employment history can still be the right hire for an entry-level role. Write down what you are sorting for and why, so the rules can be checked.
Niche roles have the opposite problem: too few suitable CVs, because many of the people you want are not actively applying.
I built a small automation tool for this. It works with LinkedIn Recruiter, which lets you look at second and third-degree connections and more detailed profiles: whether someone may be open to a move, what industry they work in, and which companies they have worked for.
I then run those results through an enrichment tool, the kind more commonly used in sales to identify decision-makers at companies. The principle is the same. It brings in more detail about a person from third-party sources, for example an article where they were quoted as the head of marketing at their company.
This is particularly useful when you are looking for candidates from a niche industry, or for a role that only a small number of people can do. It turns a cold search into a short list of people worth approaching, with enough context to make the first message relevant to them.
Enrichment tools collect personal information about people who have not applied to you, so check how your chosen tool sources its data and how it stands under the Protection of Personal Information Act (POPIA) before you rely on it.
Once you have CVs for a specialised role, a large language model (the kind of AI behind tools such as ChatGPT, Claude, or Copilot) can compare each CV against the role's requirements and help you build a shortlist. It is good at spotting where someone's experience maps onto what you need, even when they describe it in different words from your job specification.
Two rules apply before any CV goes near a model.
You should never put any CV information into a free AI tool. Free consumer versions can use your conversations to improve their models unless you switch that off. OpenAI, for example, says it may use content from its services for individuals, such as ChatGPT, to train its models, while business products such as ChatGPT Team, Enterprise, and the API are not used for training by default.
So use a paid or company plan from a provider that does not use your data to train its models. If you are still choosing a provider, our guide on how to choose the right AI model for your company covers what to look for.
Before you run the CVs through your matching model, strip out the personal details: name, surname, ID number, email address, and mobile number. Give each CV a serial number instead, such as 010203, and keep the key that links serials to people somewhere separate.
This protects candidates' information, and it also helps reduce bias. The model only sees the person's skills and experience, so their name and identity cannot influence the ranking. Other details on a CV, such as a school or a home suburb, can still hint at who someone is, so it is worth removing anything that is not relevant to the role as well.
Under POPIA, information that has been de-identified to the extent that it cannot be re-identified falls outside the Act. A serial number with a key you keep is not that, so treat the anonymised CVs as personal information and handle them accordingly. The anonymising step reduces what you expose, and your POPIA obligations still apply.
This is a simpler use, and one I find very useful: using AI as a thinking partner when you are preparing to interview a specific candidate.
You often go into an interview with some concerns about a candidate, or with areas you want to focus on. Those are exactly the areas where well-prepared questions make the difference between a useful answer and a vague one.
Describe the situation to the AI. Explain what you want to achieve with the interview, list the areas you have concerns about, and ask it to help you come up with questions that will get you answers to those concerns. You get a set of questions to work from, and just as usefully, you get pushed to be clear about what you are trying to find out before you walk into the room.
If you share CV details as part of this, the same rules apply as above: a paid plan that does not train on your data, and personal details removed.
The last big use case for me happens a step before the recruitment process starts, and it is the one most often skipped.
I have seen it time and time again. A business hires someone into a role, only to find out later that the work they are doing, or the skills they have, does not solve the gap that made the business want to recruit in the first place. The thinking behind the hire was something like "if we hire this person, our road map will move faster, our cadence will be better, we will hit these deliverables, or we will be able to launch new products". Nobody tested whether this role, done by this person, would do that.
This is where I use AI as a thinking partner to build what I call a talent card for each role before the hiring starts. A talent card works through questions such as:
Putting all of that together makes you far more calculated in hiring. The AI's job is to ask you the uncomfortable questions: is this the right hire, is this the right time to hire, is this the problem we are trying to solve? That is something that never used to exist for most businesses, and I do not think a headhunter or placement firm could fill that function, because they would not have an intricate understanding of your business.
The reason this matters more now is that the world of work is changing. Businesses can execute with much leaner teams and compete with organisations that historically needed a far larger headcount. When teams are smaller, each hire carries more weight. You need to be confident that the role you are hiring for, and the person you put in it, will serve the business for a longer period of time.
The alternative is a knee-jerk hire made to fix a topical problem. That person will probably solve the problem, and then sit in your business without a clear purpose. A talent card makes that kind of hire much less likely.
These five uses sit at different levels of risk. Preparing interview questions and building talent cards involve little or no candidate data, so they are good places to start. Sorting and matching CVs involve personal information and influence who gets considered, so they need the data rules above and a person who makes the final decision.
That split, between work AI helps you think through and decisions it influences about people, is the same one we set out in our guide to AI strategy for HR leaders, which also covers what POPIA, the Employment Equity Act, and the EU AI Act ask of recruitment tools. If your recruiters are new to these tools, training built around their own work matters more than the choice of tool.
Pick the next role you plan to open and build a talent card for it before anything else. It takes little time, and it will tell you whether the role should be advertised as it stands.
If you want a wider view of how ready your organisation is for AI, our free AI Readiness Diagnostic takes about five minutes. If you would rather talk through AI in your recruitment process, our page on AI consulting for HR teams explains how we work with HR leaders.
The most useful applications I have found are sorting high volumes of CVs for roles such as retail, factory, and call centre work, finding specialised candidates through tools such as LinkedIn Recruiter, matching anonymised CVs to specialised roles, preparing interview questions, and defining what a role is for before you hire.
Not into a free version. Free consumer AI tools can use your conversations to train their models unless you switch that off. Use a paid or company plan that does not train on your data, and remove names, ID numbers, email addresses, and phone numbers from CVs before you use them.
Removing names and identifying details before matching helps, because the model then compares skills and experience. Other details, such as a school or suburb, can still hint at who someone is, so remove anything that is not relevant to the role and keep a person responsible for the shortlist.
A talent card is a short document you build before recruiting for a role. It sets out the role's intended output, its purpose and place in the business, its KPIs, how it should look over one to five years, and whether it would still make sense if the business grew 35% to 40%. AI is useful as a thinking partner for working through those questions.
In my view, no. AI helps with sorting, sourcing, and preparation, and a person should still make every hiring decision. A headhunter or placement firm also cannot build a talent card for you, because they would not have an intricate understanding of your business.
Yes, within the law. Candidate information is personal information under POPIA, including CVs you have anonymised but can link back to a person. Keep a person responsible for hiring decisions, and confirm specific uses with your legal advisers.
Ready to move from reading about AI to getting your organisation ready for it? Start with the diagnostic.
Take the AI Readiness Diagnostic