The short answer
Recruiters should use AI for writing and organising: job ads and scorecards in ChatGPT, structured interview questions and note summaries in Claude, translation in DeepL and candidate updates through Zapier. Keep screening, ranking and decisions with people. Under the EU AI Act, AI used to filter applications or evaluate candidates is high-risk, and the Digital Omnibus moved those rules to 2 December 2027, while the ban on workplace emotion recognition has applied since February 2025.
| Start with | ChatGPT for ads, Claude for interview guides |
|---|---|
| Best free option | ChatGPT Free for ads with no candidate data |
| Keep human | Screening, ranking, interviews, offers |
| EU AI Act | Recruitment AI is high-risk; rules apply from 2 December 2027 |
| Already banned in the EU | Emotion recognition in workplaces, since February 2025 |
How recruiters are using AI in 2026
Recruiting is full of writing that has to be good but is rarely new: job ads, outreach messages, interview guides, status updates and rejection emails. AI is very good at that layer. It is also tempting to let it go one step further and sort, score or rank people, and that is exactly where the law, the bias risk and your reputation start to bite. The useful line for 2026 is simple: let AI write, summarise and organise; keep people deciding about people.
The general assistants are capable enough that you rarely need a specialist tool to start. ChatGPT on GPT-6 writes quick, clean ads and messages. Claude handles long interview notes and nuanced feedback well, and its Team plan (2 to 150 users) does not train on your content by default. DeepL translates ads naturally, and its paid plans never use your texts for training. Our hiring stack shows how ChatGPT, DeepL, Notion and Zapier fit together for a team hiring in two languages.
The 7 recruiting tasks worth handing to AI
1. Write inclusive job ads and a scorecard (ChatGPT)
Describe the role, the team and the real must-haves, and ask for an ad in plain language with must-haves separated from nice-to-haves, plus an interview scorecard. The scorecard is the most valuable output: it forces the hiring manager to agree on criteria before the first CV arrives. Cut the must-have list hard, because long lists put good people off.
2. Translate ads and candidate messages (DeepL)
For roles advertised in more than one language, DeepL keeps the tone natural and its glossaries fix how your job titles and company terms are translated. Use a paid plan for anything with candidate details: DeepL says texts on the free version may be used to train its models, while paid plans delete texts after translation.
3. Build structured interview questions (Claude)
Give Claude the scorecard and ask for behavioural questions for each criterion, with what a strong and a weak answer sound like. Every candidate then gets the same questions, which is fairer and easier to compare. Claude is good at spotting questions that drift into personal territory, if you ask it to check.
4. Personalise outreach from public information (ChatGPT or Claude)
Paste the public facts that make someone a fit, such as a talk they gave or a project they shipped, and ask for a short first message that mentions one specific thing. Read it as the candidate would. Do not paste whole profiles or contact details into a personal account.
5. Summarise interview notes against the scorecard (Claude)
After an interview, paste your own notes and ask for a summary organised by scorecard criterion, with direct evidence for each. This saves time in debriefs. The summary must describe evidence, not score or rank the candidate; the panel does that.
6. Keep every candidate informed (Zapier)
Candidates mostly complain about silence. A simple Zapier workflow can send an acknowledgement when an application lands and a status message when a stage changes in your tracker. Zapier's free plan covers 100 tasks a month with two-step Zaps, which is enough to test the idea. If you let an AI by Zapier step send messages, switch on "Require approval before running" for that action, so a person checks rejections before they go.
7. Research a market or a hard-to-fill role (Perplexity)
Ask Perplexity where people with a rare skill tend to work, which job titles they use, and what is changing in that market, with sources. It is a starting map, not salary data you can quote to a candidate; check pay ranges against your own benchmarks.
What recruiters should keep human
- Screening, shortlisting and ranking. Deciding who moves forward is a decision about a person. Do not ask a chatbot to rank CVs or score candidates.
- Interviews and the final decision. A panel that uses the scorecard decides, and can explain why.
- Rejections after interview. Draft them with AI if you want, but a person who knows the candidate edits and sends them.
- Offers and negotiation. Pay, start dates and personal circumstances need judgement and discretion.
- Anything involving health, disability or adjustments. This is sensitive data under GDPR and should never go into a general AI tool.
A starter AI stack for recruiters
| Job | Tool | Plan to use | Why |
|---|---|---|---|
| Ads, scorecards and messages | ChatGPT | Business (2 to 200 people) | Fast, clean writing; Business data is not used for training by default |
| Interview guides and note summaries | Claude | Team | Careful with nuance; no training on Team content by default |
| Translation | DeepL | A paid plan | Natural tone; paid texts are not used for training |
| Pipeline | Notion | Free or Plus to start | One shared board for every role and candidate |
| Candidate updates | Zapier | Free to test | Sends acknowledgements and status changes, with approval steps |
For the full comparison, read the best AI for HR and recruiting, and for ready prompts see prompts for HR and prompts for job interviews.
A 30-day AI rollout plan for a recruiting team
- Days 1 to 7: draw the line. Write down which tasks AI may help with (writing, translation, summaries of your own notes, scheduling messages) and which it may not (screening, scoring, ranking). Choose business plans, and confirm with your data protection lead how candidate data may be used. Our AI policy guide has a template.
- Days 8 to 14: rewrite your ads. Run your five most-used job ads through the job-ad prompt below, agree scorecards with hiring managers, and translate the ads that need it.
- Days 15 to 21: structure the interviews. Build question sets from each scorecard and test them in real interviews. Start summarising your own notes by criterion before debriefs.
- Days 22 to 30: close the loop. Set up candidate acknowledgement and status messages with an approval step. Measure time to first response and time spent per hire, then decide what to keep.
The risks for recruiters: the EU AI Act and bias
This is a summary of official sources, not legal advice. Under the EU AI Act, Annex III point 4 lists AI systems intended for the recruitment or selection of people, including placing targeted job adverts, analysing and filtering applications and evaluating candidates, as high-risk. The European Commission says the Digital Omnibus on AI, which entered into force on 27 July 2026, moved the Annex III high-risk rules from August 2026 to 2 December 2027. From then, an employer using such a system must follow the provider's instructions, give oversight to people with the competence and authority to overrule it, keep the system's logs for at least six months, inform workers' representatives and affected workers before using it at work, and tell candidates when it is used to make or help make decisions about them (Article 26).
Some rules already apply. Since 2 February 2025 the Act has banned emotion recognition in workplaces, so avoid any interview tool that claims to read a candidate's emotions from video or voice. The Commission lists AI literacy obligations as applying from the same date, so give everyone who uses AI at work basic training on what it does well and where it fails.
Outside the EU, rules differ. In New York City, Local Law 144 has been enforced since 5 July 2023: employers may only use an automated employment decision tool if it had a bias audit within the last year, a summary of the results is public, and candidates are notified 10 business days in advance.
The everyday risk is bias. A model asked to "pick the strongest CV" can reward the wrong signals, such as a familiar university, an unbroken career or a particular writing style. Keep AI on the writing and organising side of the line, use the same scorecard for everyone, and check your ads for wording that puts groups off. Candidate data is personal data under GDPR, and anything revealing health is special category data with stricter rules, so keep it out of general AI tools entirely.
Prompts to copy
Write an inclusive job ad and scorecard
Write a job ad for a [job title] in our [team] at [company type]. Plain language, a two-sentence summary of why the role matters, five responsibilities, no more than five must-haves kept separate from nice-to-haves, and a [salary range] line. Avoid gendered or age-coded words and jargon. Then write an interview scorecard with four criteria, each with what good evidence looks like.
You get: A job ad and a scorecard the hiring manager can approve
Translate an ad naturally
Translate this job ad into [language] for candidates in [country]. Keep the job title [in English or translated], use the [formal or informal] form of address, and keep the must-have list exactly as written. Flag any phrase that would sound odd to a local reader.
You get: A natural translation with flagged phrases
Build structured interview questions
Here is our scorecard for a [job title]. For each criterion, write two behavioural questions and one follow-up, then describe what a strong and a weak answer would include. Check every question and remove any that touches age, family, health, religion, nationality or other personal matters.
You get: A fair, repeatable interview guide
Personalise a first message
Write a first message to a [role] about our [job title] role. Mention this one public fact about their work: [fact]. Under 80 words, one clear reason the role fits them, one easy question to reply to, no flattery and no buzzwords.
You get: A short outreach message that feels personal
Summarise interview notes by criterion
Here are my notes from an interview and our scorecard. Organise the notes under each scorecard criterion, quoting the evidence from my notes. Do not score, rank or recommend. Where my notes have no evidence for a criterion, write 'No evidence noted'.
You get: Evidence grouped by criterion for the debrief
Write candidate status messages
Write three short, warm candidate messages for [company]: application received (with the date by which they will hear back: [date]), moving to interview (with the next step), and not progressing at the application stage. Each under 90 words, honest and specific, no clichés.
You get: Templates for your automated updates
Map a hard-to-fill market
For a [job title] with [rare skill] in [location], what job titles do people with this skill use, which types of employers hire them, and what has changed in this market in the last 12 months? Give a source link for each point and say when a point is based on a single source.
You get: A sourced map of where to look