119
Developing ethical and professional AI shortlisting criteria
Our AI shortlisting feature will reduce bias in your shortlisting process, save you hours and help you engage with talent faster, if you use it correctly. So rather than just turn on the technology, we wanted to give you a guide as to how best to use this to ensure fairness and efficiency.
We recommend that you closely follow the following dos and don'ts:
- Only use AI shortlisting to search for evidence on the applicant’s CV
Our AI shortlisting works by having a set task for the AI to complete. The task is to evaluate the applicant’s CV against the criteria that you set out. It has been designed to strictly follow this instruction. You must think about whether the applicant's CV will likely include the criteria you are adding?
Examples of criteria we don’t recommend you use AI shortlisting to answer
a) Asking the AI to look for “soft skills”
- Applicants should be knowledgeable/skilled with excellent verbal communication skills
The AI cannot confirm soft skills like verbal communication skills from a CV.
b) Adding too vague a statement
- Applicants should have experience working independently.
This statement is too vague. The AI is likely to reward applicants who add bland statements like “I can work independently or as part of a team” with a high mark.
You could try rephrasing this statement. For example, if upon reflection you think you need someone who has worked in a senior role at an SME where they had to both assign and also complete work, then you could likely write a new statement to ask the AI to look for this.
Additionally, you may determine that this "independent working" requirement is best verified after you have the opportunity to talk with and dig deeper into the backgrounds of the applicants you shortlist.
c) Asking AI to look for items that might not appear on a CV
- Applicants should be qualified/trained with a full UK driving license
Not every applicant would add to their CV to confirm they've got a full driving license. Whilst this question might be okay if you are recruiting for someone who drives for a living, it would not be ideal if applicants do not think of driving as a primary skill and therefore might not include it on their CV.
- The CV must include any of the following term(s) golf
Although you might require knowledge of golf, perhaps if you are recruiting for a Golf retailer. Many applicants do not put a detailed list of hobbies on their CV, so this question would be best managed as a vacancy-specific question that is added to the application form for every candidate to answer.
2. Asking the AI to search for something that might include bias
Example 1 - Asking for UK requirements when not necessary
Above, you might remember we asked for the AI to search for applicants with a “full UK driving license.” In this case, specifying “full UK driving license” could unintentionally discourage perfectly suitable candidates with equally valid licenses from applying — particularly those who have recently moved to the UK or are EU nationals.
Instead, use inclusive wording such as “Full driving license valid for use in the UK.” This keeps the focus on what’s genuinely required to do the job, rather than needlessly narrowing the pool of applicants.
Being aware of this type of unintentional bias is crucial when instructing an AI to screen or search candidates. If the AI is trained on your instructions, it will replicate those assumptions — reinforcing existing biases in your process. A small tweak in how you phrase things can make your recruitment far more inclusive and fair.
Example 2 - Asking for applicants to have attended certain educational institutions
Another common example of unintentional bias is asking the AI to prioritise applicants who have attended specific universities — such as “Russell Group universities” or “Oxbridge”. While you may believe this is a shorthand for identifying high-calibre candidates, it often results in unfair exclusion of individuals who may have studied elsewhere due to financial, geographical, or personal reasons but are equally capable and qualified.
This kind of filtering tends to favour those from more privileged backgrounds and overlooks the diverse talent that exists across the wider educational landscape. It can also lead to less innovation and fresh thinking within your organisation, as you risk hiring people with very similar experiences and perspectives.
When setting AI criteria, avoid naming specific institutions unless there is a clear and justifiable business reason for doing so. This helps ensure that your hiring process remains fair, inclusive, and focused on what truly matters for success in the role.
Example 3 - Asking for applicants with certain personality traits
It can be tempting to ask the AI to search for candidates who are described as “dynamic,” “confident,” “energetic,” or “charismatic.” These words might seem harmless—even desirable—at first glance. However, they introduce subjectivity and unconscious bias into your hiring process and can lead to the unfair exclusion of qualified candidates who may not express themselves in these ways but still have the skills, experience, and temperament to excel in the role.
For example, “dynamic” often implies extroversion or a fast-paced, high-energy working style, which may not be relevant or necessary for the job. It could disadvantage neurodivergent candidates, introverts, older candidates, or those from cultures where humility or teamwork is more valued than self-promotion. Similarly, traits like “resilience” or “grit” can be loaded terms—sometimes used to dismiss legitimate workplace challenges rather than support wellbeing or reasonable adjustments.
It is also worth repeating that asking the AI to look for these personality traits might not be a good idea to exercise when all it has to work with is a CV document.
3. Best practice for using AI shortlisting
Example 1 - Be explicit
Our AI shortlisting works best when it is given clear instructions as to what you are looking to see in a suitable applicant. Below are some examples where the user would be advised to be more specific:
- Applicants should be knowledgeable/skilled in sales
The above is only a good statement for AI shortlisting, if you are truly open to applicants from any type of sales background.
- Applicants should be knowledgeable/skilled in customer experience
Once again, this is likely too vague. Be clearer about the type of experience you are looking for.
Below are some alternative detailed statements that would help guide the AI to the most relevant applicants:
- Applicants should be knowledgeable/skilled in selling telecoms services into the public sector
- Applicants should be knowledgeable/skilled in delivering both face to face and telephone based customer service in a hospitality environment
Remember, our AI shortlisting will make logical judgements when reviewing a CV. So an applicant might mention that they have worked in a hotel in a role that involved “covering reception.” The AI will know that this likely involved face to face and telephone based customer service.
Example 2 - Make sure your question cannot be misinterpreted
In our testing, we saw some recruiters who created statements that included lists of job titles or skills, such as
- Applicants should be knowledgeable/skilled in Java, JavaScript, HTML
This is not ideal. As you have not been explicit about whether all 3 or just 1 of these skills is needed.
- Applicants should have experience working as a Product Manager, Head of Product
Once again, do you want the AI to search for people with experience as a Head of Product and a Product Manager?
The trick here is to add in “and” or “or” to your statement so it is clear what you are really looking for in suitable applicants.
Summary
When setting criteria for AI to screen and rate applicants, it’s essential to carefully consider whether each requirement is genuinely relevant to the role and free from bias.
Unlike human shortlisting, AI-led shortlisting leaves a clear, documented trail. Our AI generates a summary of its decision-making process, which candidates are entitled to request under GDPR. This means you must be confident that the criteria you've provided are fair, justifiable, and stand up to scrutiny.
Example summary
