{"id":1894,"date":"2025-07-15T18:13:00","date_gmt":"2025-07-15T16:13:00","guid":{"rendered":"https:\/\/laurenswaling.com\/?p=1894"},"modified":"2026-01-15T18:14:05","modified_gmt":"2026-01-15T17:14:05","slug":"ai-matching-in-recruitment-why-generic-ai-isnt-enough","status":"publish","type":"post","link":"https:\/\/laurenswaling.com\/?p=1894&lang=en","title":{"rendered":"AI Matching in Recruitment: Why Generic AI Isn&#8217;t Enough"},"content":{"rendered":"\n<p id=\"ember3141\"><em>AI is transforming hiring. But not all AI is created equal. The Workday lawsuit shows what can go wrong when generic AI is used to screen candidates. This blog explores the difference between traditional hiring, generic large language models (LLMs), and specialized AI for talent matching \u2013 and what HR should do next.<\/em><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p id=\"ember3142\">In the age of AI, hiring is no longer about simply reading r\u00e9sum\u00e9s or posting jobs. From chatbots to automated matching systems, AI is now deeply embedded in how companies find and select talent. But as we learned from the recent Workday lawsuit, the way AI is <em>trained<\/em> and <em>used<\/em> matters \u2013 a lot.<\/p>\n\n\n\n<p id=\"ember3143\">In this blog, we\u2019ll explore three levels of matching in recruitment:<\/p>\n\n\n\n<ol>\n<li><strong>Human-based and keyword-driven matching<\/strong><\/li>\n\n\n\n<li><strong>Generic AI-based matching with LLMs like ChatGPT<\/strong><\/li>\n\n\n\n<li><strong>Specialized matching with trained AI models built for the labor market<\/strong><\/li>\n<\/ol>\n\n\n\n<p id=\"ember3145\">And we\u2019ll explain why only one of these is fit for responsible, fair, and future-ready hiring.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"ember3146\">1. Traditional Matching: Biased but Explainable<\/h3>\n\n\n\n<p id=\"ember3147\">Before AI, recruiters relied on manual CV screening and gut feeling. Early digital tools helped filter candidates using simple rules: for example, \u201cshow me all who mention \u2018Excel\u2019.\u201d<\/p>\n\n\n\n<p id=\"ember3148\">\u2705 <strong>Pros<\/strong><\/p>\n\n\n\n<ul>\n<li>Transparent and intuitive<\/li>\n\n\n\n<li>Recruiters can consider personality and context<\/li>\n<\/ul>\n\n\n\n<p id=\"ember3150\">\u274c <strong>Cons<\/strong><\/p>\n\n\n\n<ul>\n<li>Limited scalability<\/li>\n\n\n\n<li>High risk of unconscious bias<\/li>\n\n\n\n<li>Easily overlooks non-standard candidates<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"ember3152\">2. LLM-Based Matching: Impressive but Misleading<\/h3>\n\n\n\n<p id=\"ember3153\">With the rise of generative AI like ChatGPT or Claude, many platforms now offer instant job matching or r\u00e9sum\u00e9 rewriting. Sounds smart, right?<\/p>\n\n\n\n<p id=\"ember3154\">But here\u2019s the catch: <strong>LLMs are not matching engines.<\/strong> They\u2019re trained to predict words \u2013 not to evaluate skills, competencies, or job-market relevance. Ask them to match someone to a role, and you\u2019ll likely get something plausible-sounding, but statistically shallow. Ask AI for skills and you will get skills, but ask again and you will get different skills. How to start calculating\/matching and how to explain te skills gap?<\/p>\n\n\n\n<p id=\"ember3155\">LLMs don\u2019t understand the labor market. They don\u2019t know that &#8220;logistics coordinator&#8221; and &#8220;supply chain analyst&#8221; share 70% of required skills. They don&#8217;t remember which skills typically lead to success in a specific job. And crucially: they have <strong>no ground truth<\/strong>.<\/p>\n\n\n\n<p id=\"ember3156\">That\u2019s not just a technical problem \u2013 it\u2019s a compliance issue.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"ember3157\">3. Specialized AI Matching: Purpose-Built, Explainable, and Compliant<\/h3>\n\n\n\n<p id=\"ember3158\">The third category is where real progress happens. These are <strong>AI systems trained on millions of job descriptions and CVs<\/strong>, designed specifically for the labor market. They don\u2019t generate text; they detect skills, predict missing competencies, and show how someone matches a job \u2013 or how they can close the gap.f<\/p>\n\n\n\n<p id=\"ember3159\">These models do more than keyword matching. They can:<\/p>\n\n\n\n<ul>\n<li>Interpret context (e.g., \u201cled workshops\u201d implies training skills)<\/li>\n\n\n\n<li>Predict hidden skills based on job history<\/li>\n\n\n\n<li>Identify missing but trainable skills<\/li>\n\n\n\n<li>Suggest lateral moves or career steps<\/li>\n\n\n\n<li>Operate in line with GDPR and the upcoming <strong>EU AI Act<\/strong>: like explaining matching scores and mitigating bias.<\/li>\n<\/ul>\n\n\n\n<p id=\"ember3161\">That\u2019s the kind of explainable AI recruiters \u2013 and regulators \u2013 are calling for. <em>(I would love to tell your more about these systems or give a demo. Just leave me a DM.)<\/em><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"ember3162\">The Workday Lawsuit: A Wake-Up Call<\/h3>\n\n\n\n<p id=\"ember3163\">In the United States, <a href=\"https:\/\/hrexecutive.com\/federal-court-to-consider-ai-in-hiring-as-workday-bias-case-advances\/\"><strong>Workday is now facing a major collective action lawsuit<\/strong><\/a>. The claim? That their AI-based screening software <strong>discriminated against applicants over 40<\/strong> and possibly other groups, leading to automatic rejections without explanation.<\/p>\n\n\n\n<p id=\"ember3164\">A federal judge allowed the case to proceed under the Age Discrimination in Employment Act. The scope? Potentially millions of candidates who applfied through Workday since 2020.<\/p>\n\n\n\n<p id=\"ember3165\">Even if Workday didn\u2019t <em>intend<\/em> to discriminate, the structure of its algorithm may have <strong>amplified existing bias<\/strong> in hiring data \u2013 and failed to detect it.<\/p>\n\n\n\n<p id=\"ember3166\">This is exactly what the EU AI Act wants to prevent: black-box systems that can\u2019t explain decisions, reproduce bias, and affect people\u2019s lives at scale.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"ember3167\">What Should HR Do?<\/h3>\n\n\n\n<p id=\"ember3168\">HR professionals are under pressure to modernize \u2013 but also to protect. The solution is not to stop using AI. It\u2019s to <strong>use the right kind of AI.<\/strong><\/p>\n\n\n\n<p id=\"ember3169\">Here\u2019s how to do it responsibly:<\/p>\n\n\n\n<ul>\n<li>Use <strong>specialized AI<\/strong> built for job-market data, not general-purpose LLMs<\/li>\n\n\n\n<li>Audit your matching process for <strong>bias and explainability<\/strong><\/li>\n\n\n\n<li>Combine AI suggestions with <strong>human oversight<\/strong><\/li>\n\n\n\n<li>Make sure candidates understand <strong>how and why<\/strong> decisions are made<\/li>\n\n\n\n<li>Prepare for <strong>EU AI Act compliance<\/strong> now \u2013 don\u2019t wait for 2026<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"ember3171\">In Summary<\/h3>\n\n\n\n<p id=\"ember3172\">AI can match people to jobs better than humans ever could. But only if it&#8217;s <strong>trained on the right data<\/strong>, <strong>designed for the task<\/strong>, and <strong>transparent enough<\/strong> to be trusted.<\/p>\n\n\n\n<p id=\"ember3173\">Generic AI can assist with writing. Specialized AI can <strong>predict success<\/strong> in a role.<\/p>\n\n\n\n<p id=\"ember3174\">If you want to match talent to opportunity \u2013 fairly, fast, and at scale \u2013 don\u2019t just ask an LLM.<\/p>\n\n\n\n<p id=\"ember3175\">Ask an expert.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>AI is transforming hiring. But not all AI is created equal. The Workday lawsuit shows what can go wrong when generic AI is used to screen candidates. This blog explores the difference between traditional hiring, generic large language models (LLMs), and specialized AI for talent matching \u2013 and what HR should do next. In the [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":1895,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"_uf_show_specific_survey":0,"_uf_disable_surveys":false,"footnotes":""},"categories":[1],"tags":[],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/laurenswaling.com\/index.php?rest_route=\/wp\/v2\/posts\/1894"}],"collection":[{"href":"https:\/\/laurenswaling.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/laurenswaling.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/laurenswaling.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/laurenswaling.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1894"}],"version-history":[{"count":1,"href":"https:\/\/laurenswaling.com\/index.php?rest_route=\/wp\/v2\/posts\/1894\/revisions"}],"predecessor-version":[{"id":1896,"href":"https:\/\/laurenswaling.com\/index.php?rest_route=\/wp\/v2\/posts\/1894\/revisions\/1896"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/laurenswaling.com\/index.php?rest_route=\/wp\/v2\/media\/1895"}],"wp:attachment":[{"href":"https:\/\/laurenswaling.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1894"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/laurenswaling.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1894"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/laurenswaling.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1894"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}