AI in Hiring: What's Actually Useful vs Marketing Hype

The AI Hiring Hype Cycle

In 2023, every hiring platform added "AI" to its marketing. In 2026, that wave has partially receded — the tools that worked are deeply embedded, and the tools that didn't have quietly disappeared or been re-branded as "intelligent matching."

What remains is a landscape where AI genuinely does some things very well, some things adequately, and some things it was never fit to do in the first place. For both candidates and employers, understanding the actual capability map matters enormously.

How ATS AI Screening Actually Works

The "AI screening" most ATS systems use is not a large language model reading your CV with human-like comprehension. It's primarily:

1. Keyword and semantic vector matching. Your CV is converted to a numerical embedding (a vector representation of its content). The job description is similarly vectorised. The system ranks applications by cosine similarity — how close your vector is to the job description's vector. This is why keyword presence matters and why unusual phrasing of common skills can hurt your score. 2. Historical pattern matching. Some systems are trained on which types of profiles led to hires in similar roles at similar companies. This creates obvious problems: it tends to replicate historical hiring patterns, including biases baked into who was hired before. The UK's ICO and the EU AI Act now both require disclosure when AI is used in employment decisions and provide candidates rights to explanation. 3. Rule-based filters. Despite the AI branding, most systems still apply hard filters: minimum years of experience, required degree level, specific certifications. These filters run before the AI scoring and are often the real gatekeepers. What this means for candidates:

• •Use precise technology names (not just "distributed systems" but "Kafka, Kubernetes, Flink")
• •Match the structure of your experience description to what the JD asks for
• •Don't assume cleverness scores higher than clarity — it often doesn't

What AI Can and Can't Do in Recruitment

Where AI is genuinely useful:

• •Scheduling coordination (genuinely reduces friction)
• •Reference check synthesis (extracting themes from long written references)
• •Job description optimisation (flagging biased language before posting)
• •Offer benchmarking (analysing comp data at scale)
• •Interview question generation (surface-level, but speeds up prep)

Where AI is oversold:

• •Personality and culture fit prediction (correlation to actual job performance is weak at best)
• •Video interview "sentiment analysis" (measuring facial expressions for culture fit has no credible scientific basis — multiple studies have failed to find predictive validity)
• •Automated rejection decisions without human review (legally and ethically problematic in UK/EU contexts)

Where AI simply doesn't work:

• •Predicting long-term job performance from CV content
• •Replacing the human judgment needed in final selection decisions
• •Understanding context that requires real understanding (career pivots, unusual paths, international credentials)

How Jobs and Careers Uses AI Differently

The fundamental problem with most hiring AI is the black box: neither candidate nor employer can explain why a decision was made. Jobs and Careers' AI approach is built on a different principle: proof of work, not pattern matching.

Instead of using AI to score candidates against invisible criteria, our AI agent JAC does visible, explainable work:

• •It reads the job description and generates specific research about the company
• •It identifies which of your existing profile elements are most relevant to each specific role
• •It drafts application materials that are explicitly grounded in things you can verify are true
• •It flags gaps between your profile and the role, giving you the choice of how to address them

The output is auditable. You can see what JAC found, why it made the connections it made, and decide whether you agree. This is a meaningfully different approach from systems that give you a "match score" with no explanation.

The Candidate's Guide to Writing for AI Systems

Given how ATS AI actually works, here's the practical guide:

Keyword mirroring (smart, not mechanical): Read the job description carefully. Note the 8–10 most specific technical terms used. Ensure those terms appear in your CV where accurate. Don't stuff keywords — use them in the context of real accomplishments. Structured consistency: Use standard section headings (Experience, Education, Skills). Keep formatting clean and simple. Avoid tables, columns, text boxes, and images — they confuse parsers. Quantification for semantic richness: Numbers and specific metrics make your CV vectors more distinct and more similar to the accomplishment-focused language used in high-quality job descriptions. "Improved latency" is vague. "Reduced p99 API latency from 2.3s to 340ms" is specific and semantically rich. Relevance ordering: Put the most relevant experience first, even if it's not most recent. Many ATS systems weight the first entries in your experience section more heavily. ATS testing: Tools like Jobscan and Resume Worded can simulate ATS matching against a specific JD. Use them as a sanity check, not a guarantee — but they catch obvious mismatches.

The bottom line: AI in hiring is real, but its capabilities are far more narrow than the marketing suggests. Understanding how it actually works is a genuine competitive advantage for candidates who do.