AI in recruitment — what actually works in 2026

We've spent three years building AI features into Recruitly. Some of them changed how agencies work. Some of them we built, shipped, and quietly pulled because nobody used them. Here's what we've learned.

AI sourcing is the big one

This is where AI makes the biggest difference in recruitment right now. Not chatbots. Not note summaries. Sourcing.

Old-school sourcing means typing keywords into a search bar and scrolling through hundreds of profiles. AI sourcing means describing what you need in plain English and getting a ranked list of candidates who actually fit. "Senior backend engineer, fintech background, open to contract, based in London." The system looks at career trajectory, skills in context, company background, seniority signals. It returns 20 candidates ranked by genuine fit, not alphabetically.

We run this against 800M+ profiles plus whatever's in your own database. Most recruiters tell us it cuts their sourcing time from two hours to about ten minutes per role. That's not an exaggeration. We've watched it happen on screen shares with agencies during onboarding.

CV parsing got quietly good

CV parsing has been around for fifteen years. The old versions were terrible. They'd mangle formatting, miss half the skills, and put the candidate's address in the job title field. Nobody trusted them.

The new generation is different. Modern parsers handle messy Word docs, PDFs with columns, even scanned images. They pull out skills, job history, education, and contact details with about 95% accuracy. That saves roughly five minutes per candidate. For an agency processing 50 CVs a day, that's over four hours saved.

The part that's actually new is CV rewriting. You upload a candidate's CV and the AI rewrites it in your agency's branded template. Removes personal details, reformats the experience section, highlights relevant skills for the specific role. What used to be a 20-minute Word exercise now takes about 30 seconds.

Candidate matching saves the morning scroll

A new job lands. You get 200 applications by lunch. The old way: open each one, skim the CV, make a gut call, move on. Two hours gone before you've shortlisted anyone.

AI matching ranks those 200 applications overnight. By 9am, you've got a shortlist of the top 15 with match scores and explanations of why each person was ranked where they are. Not a black box "87% match" with no context. Actual reasoning: "5 years Python, 2 fintech companies, based in target location, open to contract per LinkedIn signals."

We built our matching to show its working. Recruiters need to trust the ranking, and trust comes from transparency.

Outreach drafting is useful, with a caveat

AI can draft outreach emails, InMails, and WhatsApp messages. It's decent at this. Give it the job spec and the candidate's background and it'll write a personalised message in seconds.

The caveat: don't send AI-written messages without reading them first. Candidates can tell. The messages are a starting point. Add your own voice, fix anything that sounds robotic, then send. The time saving is still significant because drafting from scratch takes much longer than editing a draft.

Where AI outreach really shines is campaign sequences. Writing 200 personalised first-touch emails manually is a full day's work. Having AI draft them and then reviewing in batches takes about an hour.

What doesn't work yet (and we've tried)

Fully automated recruiting. We've had customers ask for a system that takes a job brief and produces a placed candidate with zero human involvement. It doesn't exist. Recruitment is a relationship business. Candidates want to talk to a person, especially for important career moves. Clients want a consultant who understands their industry. AI can make every step faster, but it can't replace the human part.

AI interviewing. We looked at building AI video interview screening. Candidates hate it. The bias concerns are real. We decided not to ship it. Agencies are better off using AI to prepare for interviews (candidate briefing notes, suggested questions based on CV gaps) rather than conducting them.

Chatbots as the primary interface. We tried making a chatbot the main way to interact with the CRM. Turns out recruiters are faster clicking buttons than typing sentences. Chat is great for complex queries ("show me all candidates in Manchester who interviewed for fintech roles last quarter") but terrible for the 50 things you do every day. We kept chat as one option alongside the normal UI.

How to tell if a CRM's AI is real

Three things to check in a demo.

Does it show its reasoning? If the AI recommends a candidate, can you see why? A match score without an explanation is useless. You need to know if the AI is matching on skills, experience, location, or something else entirely.

Does it learn from your data? AI that only works off a generic model is mediocre. Good AI improves as you use it, learning from your placements, your rejections, your sector. Ask the vendor what data the AI trains on and how it improves over time.

Is it in the workflow or on the side? AI features that live in a separate tab get ignored. The AI should be embedded in the screens you already use. When you open a job, matching candidates should already be there. When you view a candidate, suggested roles should already be there. If you have to go looking for the AI, it's not integrated properly.

We wrote a broader guide on choosing a recruitment CRM that covers AI alongside everything else. If you want to see how we've built AI into Recruitly specifically, the AI page has the details.