Where AI Gets It Wrong

AI is useful until it starts pretending to understand people

AI is strongest when it is doing structured, repeatable work: summarising information, highlighting keywords, comparing answers against a defined framework, or helping candidates prepare more effectively. It is much weaker when the decision depends on context, judgement, nuance or potential that does not show up neatly in data.

That is where AI gets it wrong in hiring. It can mistake similarity for quality, confidence for competence, and tidy CVs for strong performance. It can also amplify gaps in the data it is trained on, which means it may favour certain career paths, writing styles or patterns of progression over others.

For recruiters and employers, the practical challenge is not to reject AI, but to use it as decision support rather than decision replacement. For careers advisers, the challenge is to help candidates present evidence that AI can recognise without letting the process become unfairly narrow.

Common ways AI gets hiring decisions wrong

AI can fail in several predictable ways. If you know the failure points, you can design around them.

• It overweights keywords. A candidate may match the wording of a job description without having the depth of experience needed.
• It undervalues non-linear careers. Career breaks, portfolio work, self-employment, volunteering or lateral moves can look weaker than they are.
• It confuses style with substance. A polished CV or fluent interview answer can be scored more highly than quieter but stronger evidence.
• It misses transferable skills. Skills gained in different sectors or roles may not be recognised if the labels do not match.
• It struggles with context. A short tenure may reflect redundancy, caring responsibilities or restructuring, not poor performance.
• It can reward sameness. If the system is trained on historic hires, it may reproduce historic preferences rather than future need.

These are not abstract risks. They show up in day-to-day hiring decisions when a recruiter sees a candidate who looks slightly unusual on paper but is clearly capable in conversation, or when a candidate with a strong work ethic and relevant aptitude is filtered out before anyone speaks to them.

A practical rule: separate screening, assessment and judgement

One of the simplest ways to reduce AI error is to stop using one signal to do every job. Build the process in layers.

1. Screen for minimum requirements. Use AI to organise information, not to decide fit on its own.
2. Assess capability with evidence. Use role-based tests, work samples or structured questions to check what the candidate can actually do.
3. Apply human judgement to context. Review the full picture before making a shortlist, interview or offer decision.

This layered approach is especially useful when the role is complex, the talent pool is mixed, or the candidate profile is likely to include non-traditional backgrounds. It also helps advisers explain to candidates why one part of the process matters, and how to prepare for it.

A decision framework recruiters can use before trusting the output

Before you act on an AI recommendation, ask four questions.

1. What is the system actually measuring?

Is it measuring skill, relevance, writing style, keyword match, or something else? If you cannot explain the measure in plain English, do not treat it as a final answer.

2. What evidence is missing?

Does the CV analysis show a gap because the candidate lacks experience, or because the evidence is presented differently? Has the system ignored a portfolio, a project, a qualification equivalence or a career break explanation?

3. What context could change the interpretation?

Ask whether the candidate has moved sectors, taken time out, worked part-time, studied while working, or built experience in a less conventional way. Context often changes the meaning of the same facts.

4. What would I want a human reviewer to check?

If the AI flags a candidate as weak or strong, what would you want a recruiter, hiring manager or adviser to verify before acting on it?

Useful test: if the AI output would change your decision, but you cannot explain why in a way a candidate would understand, it is not ready to be used as a standalone decision.

How to assess candidates fairly when AI is part of the process

Fair assessment is not about removing judgement. It is about making judgement more consistent, more evidence-based and less vulnerable to first impressions.

Use structured criteria, not vague impressions

Define the role in terms of observable behaviours and outputs. For example, instead of asking whether someone seems “commercial”, define what commercial behaviour looks like in the role: managing priorities, understanding customer needs, using data, or making trade-offs.

Compare like with like

When reviewing candidates, compare evidence against the same criteria. A strong candidate with a non-standard CV should not be penalised for not looking like the last hire. Use the same questions, the same scoring scale and the same evidence thresholds.

Look for proof, not polish

Polished language can be persuasive, but it is not the same as performance. Ask for examples, outcomes, metrics, work samples or role-based test results where appropriate. CareerMapper’s role-based tests can help you see how candidates apply knowledge in a job-relevant setting rather than relying only on presentation.

Use interviews to test assumptions

AI may highlight a candidate as a strong match, but the interview should test whether that match holds up under questioning. CareerMapper interview preparation can help candidates practise clearer, more relevant answers, while one-to-one interview reports can help advisers and recruiters identify recurring gaps in evidence, structure or confidence.

Examples of where human judgement changes the outcome

Example 1: The candidate with a career break

An AI screening tool may rank a candidate lower because of a two-year gap. A human reviewer may discover that the gap involved caring responsibilities, freelance work and a qualification completed part-time. The real question is not whether the gap exists, but whether the candidate can do the job now.

Decision question: what evidence shows current capability, not just continuous employment?

Example 2: The candidate from a different sector

A candidate moving from hospitality into customer success may not match sector keywords. But if the role depends on handling pressure, reading customer needs and resolving issues quickly, the transferable evidence may be strong. AI can miss that if it only looks for sector labels.

Decision question: which skills are genuinely sector-specific, and which are transferable?

Example 3: The quiet but capable interviewee

An AI-supported interview summary may favour the candidate who speaks fluently and confidently. A human interviewer may notice that another candidate gives shorter answers but provides precise examples, strong judgement and clear ownership of outcomes. Confidence is not the same as competence.

Decision question: are we rewarding communication style, or the quality of the evidence?

How advisers can help candidates avoid being misread by AI

Careers advisers play an important role in helping candidates present evidence in ways that both people and systems can understand. That does not mean forcing everyone into the same template. It means making their strengths visible.

• Use CV analysis to identify weak signals. Help candidates spot where their experience is buried, vague or under-evidenced.
• Translate experience into outcomes. Encourage candidates to show what changed because of their work, not just what tasks they performed.
• Prepare for structured interviews. Candidates should practise concise examples, clear context and measurable results.
• Use work style assessment carefully. It can help candidates understand how they tend to work, but it should support reflection rather than box them in.
• Build evidence for transferable skills. Volunteering, projects, part-time work and study can all be relevant if framed properly.

CareerMapper’s interview preparation tools and one-to-one interview reports can help advisers spot where a candidate is being underestimated, overexplaining, or failing to connect their experience to the role.

How employers can use CareerMapper without outsourcing judgement

CareerMapper is most useful when it improves clarity, not when it replaces decision-making. Used well, it can help employers and advisers create a more complete picture of the candidate.

• CV analysis can highlight relevant experience, gaps, and areas that need human review.
• Role-based tests can check job-relevant capability more directly than CV wording alone.
• Work style assessment can support conversations about how someone prefers to work, communicate and prioritise.
• Employer candidate overview can bring together evidence from different stages so reviewers are not relying on one signal.
• One-to-one interview reports can help advisers and recruiters see patterns in how candidates respond under pressure.

The point is not to let the platform decide who is best. The point is to make the evidence easier to compare, discuss and challenge.

A simple shortlist check before you make a decision

Before moving a candidate forward, run this quick check:

1. Do we have evidence of capability, or only evidence of similarity?
2. Have we considered context that could explain apparent weaknesses?
3. Are we comparing candidates against the role, or against each other?
4. Would we be comfortable explaining this decision to the candidate?
5. What would a human reviewer add that the AI cannot?

If the answer to the last question is “nothing”, the process is probably too automated. In most real hiring situations, human judgement should add interpretation, not just approval.

What good practice looks like in a mixed AI and human process

A sensible hiring process does not ask AI to be wise. It asks AI to be efficient, consistent and transparent enough to support better human decisions.

Good practice usually includes:

• clear role criteria written before screening begins;
• structured review of CVs and applications;
• human oversight of borderline or unusual cases;
• job-relevant tests where appropriate;
• interviews that probe evidence, not just confidence;
• documentation of why decisions were made.

That approach is better for recruiters, better for employers and better for candidates. It also helps careers advisers give more targeted support, because they can see where a candidate is likely to be misunderstood and where they need stronger evidence.

Final thought: use AI to sharpen judgement, not replace it

AI gets it wrong when it is treated as a verdict rather than a tool. Human judgement gets it wrong too, especially when it is rushed, inconsistent or based on first impressions. The answer is not to choose one over the other. It is to combine them properly.

Use AI to organise evidence. Use structured assessment to test capability. Use human judgement to interpret context. That is where fairer hiring starts.

Frequently asked questions

Can AI be trusted to shortlist candidates?

AI can help with early screening, but it should not be trusted to make final shortlist decisions on its own. It may miss context, transferable skills or non-standard career paths, so human review is still important.

What is the biggest risk when using AI in hiring?

The biggest risk is treating AI output as objective truth. If the system is based on incomplete or historic patterns, it can reproduce bias or overvalue proxies such as keywords, writing style or linear career history.

How can recruiters check whether AI has got a candidate wrong?

Ask what the system measured, what evidence it missed, and whether there is context that changes the interpretation. Then compare the candidate against the role criteria, not just against other applicants.

How can advisers help candidates perform better in AI-supported hiring?

Advisers can help candidates make their evidence clearer through CV analysis, interview preparation and practice with role-based examples. The aim is to make strengths visible without forcing everyone into the same mould.

Where does CareerMapper fit into this?

CareerMapper supports decision-making and candidate development through CV analysis, interview preparation, one-to-one interview reports, role-based tests, work style assessment and employer candidate overviews. It helps organise evidence, but it should sit alongside human judgement.

Should employers remove AI if it sometimes gets things wrong?

Not necessarily. A better approach is to define where AI is useful, where human review is required, and what evidence must be checked before a decision is made. Good process matters more than perfect automation.