The short answer
The best AI for HR and recruiting in 2026 is Claude on Anthropic's Team plan, because it writes and reads HR documents well and Team content is not used for training by default. Microsoft 365 Copilot is the runner-up for Microsoft companies, and Gemini Notebook is the best free option for policy questions. Under the EU AI Act, AI used to filter applications or evaluate candidates is high risk, with those rules now applying from 2 December 2027.
| Our pick | Claude (Team plan), by Anthropic |
|---|---|
| Runner-up | Microsoft 365 Copilot |
| Best free option | Gemini Notebook Standard, by Google |
| Keep away from AI | Ranking or rejecting candidates without a compliant system and a human decision |
| EU AI Act, hiring | High risk (Annex III, point 4); rules apply from 2 December 2027 |
| Already banned | AI that infers emotions in the workplace, since 2 February 2025 |
Our pick for HR and recruiting: Claude on the Team plan
Why: most of the time HR saves with AI goes on writing and reading: job adverts, interview guides, policies, handbooks, offer letters and long employee relations files. That is where Claude, made by Anthropic, is strongest. Claude Opus 5.5, released on 22 September 2026, is the main model on paid plans, and since 16 September 2026 any Claude conversation can take on a whole task and hand back an editable document, spreadsheet or deck. Claude Docs export to Word, PDF, Markdown or Google Docs, so a policy draft goes straight into your usual tools.
Who it suits: HR teams of 2 to 150 people who handle personal data every day. The Team plan matters more than the model here: Anthropic does not use Team or Enterprise content for training by default, and Team adds SSO and admin controls. Enterprise adds SCIM, audit logs, a Compliance API and custom data retention, which larger HR functions usually need.
The catch: the personal Free, Pro, Max and Ultra plans can use your chats for training unless you switch off Help Improve our AI models under Settings, then Privacy. Never paste candidate or employee data into a personal account. Every plan also has usage limits, and Anthropic says data for Claude Enterprise and its API is stored in the US, which some European employers will need to clear with their data protection officer. Read our Claude guide for the plan details.
Runner-up: Microsoft 365 Copilot, if your company runs on Microsoft
Why: Microsoft 365 Copilot works inside Outlook, Teams, Word, Excel and PowerPoint using your organisation's own files and mail, under enterprise data protection. For work accounts, Microsoft says prompts, responses and Microsoft Graph data are not used to train foundation models, and sensitivity labels, retention and audit still apply. In Teams it writes meeting recaps, and since October 2026 it can do so without keeping a transcript, which is useful for sensitive conversations. Copilot Cowork, launched in September 2026, takes on multi-step work such as turning a stack of exit interview notes into a themed report.
Who it suits: HR teams whose policies, contracts and people data already live in SharePoint, OneDrive and Outlook, and whose IT team has bought Copilot licences.
The catch: there are two Copilots. The free personal app is a different product, and Microsoft has not documented how its updated app (from 18 August 2026) handles training, so keep HR work out of it. Feature names also change monthly, and many new features start in Microsoft's Frontier early access programme. See our Copilot guide.
Best free option: Gemini Notebook for policy and handbook questions
Why: Gemini Notebook, called NotebookLM until 16 July 2026, answers questions using only the sources you add, with citations back to the passage. Load your employee handbook, leave policy and benefits guide, and it becomes a policy assistant that shows where every answer comes from. The free Standard plan allows 100 notebooks with 50 sources each, and each source can be up to 500,000 words or 200 MB.
Who it suits: HR generalists and small businesses who want fast, checkable answers from their own policy documents, and anyone building a first FAQ for managers.
The catch: it only knows what you give it, so an out-of-date policy produces confident, well-cited, wrong answers. Use it for policy documents, not personal files. On a personal Google account Google says notebook content is not used to directly train its foundational models unless you give feedback, and feedback can be read by trained reviewers. With a work or school account there is no human review and no training. Our Gemini Notebook guide covers the limits.
Bias in AI hiring: what goes wrong and how to guard against it
Language models learn from vast amounts of human writing, and human writing about work carries human bias. In HR that shows up in three places. Job adverts drift towards wording that puts some groups off applying. Summaries of applications can weigh the wrong things, such as a gap in employment or an unfamiliar university. And if you ask a chatbot to rank candidates, it will give you a confident order with no reliable way to explain it.
The EU AI Act takes this seriously. For high-risk systems, the providers who build them must examine their data for possible biases that could lead to discrimination prohibited under EU law (Article 10(2)(f)) and take measures to detect, prevent and mitigate them (Article 10(2)(g)). A general chatbot you use on your own has had none of that checking for your hiring process. Practical rules that hold up:
- Use AI on the documents around hiring, not on judging people. Drafting the advert, the interview guide and the rejection email is low risk. Scoring candidates is not.
- Write the criteria first. Agree a scorecard before you see applications, and have humans score against it.
- Ask AI to audit your wording, not your candidates. A prompt that flags gendered or exclusionary language in an advert is a good use. A prompt that says "which of these five is best" is not.
- Keep a record. Note where AI helped and who made each decision, so you can explain it to a candidate, a works council or a regulator.
The EU AI Act and AI in hiring: what applies and when
This is a summary of official EU sources, not legal advice. Check your own situation with your legal team or an employment lawyer.
- Hiring AI is high risk. Annex III, point 4 of the AI Act lists AI systems intended for recruitment or selection, in particular to place targeted job adverts, to analyse and filter job applications and to evaluate candidates. It also covers AI used for decisions on promotion or termination, to allocate tasks based on behaviour or personal traits, or to monitor and evaluate performance.
- The date moved. The European Commission's AI Act page (updated 19 October 2026) says the rules for Annex III high-risk uses, which include employment, have been extended to 2 December 2027 by the AI Omnibus, which entered into force on 27 July 2026. They were originally due on 2 August 2026.
- Some rules already apply. Prohibited practices and the AI literacy duty have applied since 2 February 2025. Since the AI Omnibus came into force on 27 July 2026, Article 4 asks providers and deployers to take measures to support the development of AI literacy, without a fixed level to reach, and the Commission's Q&A, updated on 20 October 2026, says organisations can keep an internal record of the training they run. One prohibition matters directly to HR: AI systems that infer the emotions of people in the workplace are banned, except for medical or safety reasons (Article 5(1)(f)). Avoid any interview tool that claims to read candidates' emotions.
- Employers have duties as deployers. Under Article 26, an employer using a high-risk system must use it according to the provider's instructions, give human oversight to people with the training and authority to do it, monitor how it works, keep the logs it generates for at least six months, and inform workers' representatives and affected workers before putting it into use at the workplace. People affected by decisions it makes or assists must be told it is being used.
What this means today: if you are buying a dedicated screening or assessment tool, ask the vendor now how it will meet the high-risk requirements by December 2027, and plan your worker consultation. For the background, see what the EU AI Act is.
How to choose an AI tool for HR: three questions
- Will personal data go in? If yes, you need a work plan that excludes your data from training by default (Claude Team or Enterprise, Microsoft 365 Copilot, ChatGPT Business or Enterprise) and a data processing agreement. If no, a free tool is fine.
- Where does your HR material live? Microsoft 365 points to Copilot. Google Workspace points to Gemini and Gemini Notebook. Mixed or document-heavy work points to Claude. Compare them in ChatGPT vs Claude and Claude vs Gemini.
- Is the AI making or shaping a decision about a person? If yes, you are close to or inside the high-risk category. Use a purpose-built system with documented compliance, keep a human in charge and consult your works council, rather than improvising with a chatbot.
Workflow: one hire from brief to first day, with AI in the right places
- Intake. Record the hiring manager's brief in a Teams or Meet call, then ask your assistant to turn the recap into a role profile: purpose, five outcomes for the first year, must-have and nice-to-have skills.
- Advert. Draft the job advert from the role profile, then run the bias check prompt below on it. Remove requirements that are habits rather than needs, such as degree requirements for roles that do not need one.
- Scorecard. Generate a scorecard with four to six criteria and behavioural anchors for each score. Lock it before applications arrive.
- Screening. Humans read applications and score them against the scorecard. AI stays out of this step unless you use a dedicated system that meets the high-risk rules.
- Interviews. Generate structured questions tied to each criterion, so every candidate gets the same interview. Use AI meeting notes only with every participant's knowledge and consent.
- Decision and offer. The panel decides. AI drafts the offer letter from your template and the agreed terms, and drafts kind, specific rejection emails.
- Onboarding. Turn the role profile and your handbook into a 30, 60 and 90 day plan and a first-week checklist.
More role-specific ideas are in AI for HR teams, AI for recruiters and our HR prompt library.
7 HR and recruiting prompts to copy
Replace the [placeholders]. Use these on a work plan whenever the material includes anything about a real person.
Prompts to copy
Write an inclusive job advert
Write a job advert for a [role] at [company], based on this role profile: [paste]. Keep it under 450 words. Lead with what the person will achieve in year one, list no more than six requirements and mark which are essential. Use plain, gender-neutral language and say we welcome applications from people who meet most, not all, of the criteria.
You get: A focused advert that does not scare off good applicants
Audit an advert for biased wording
Review this job advert for wording that could discourage some groups from applying: gendered words, age signals, unnecessary degree or years-of-experience requirements, jargon and anything that could indirectly discriminate. Quote each phrase, explain the risk in one sentence and suggest a neutral replacement. Advert: [paste]
You get: A line-by-line list of risky phrases with fixes
Build an interview scorecard
Create an interview scorecard for a [role]. Use [4 to 6] criteria taken only from this role profile: [paste]. For each criterion give a one-line definition and what a 1, 3 and 5 answer looks like in observable behaviour. Do not include criteria about personality, culture fit or background.
You get: A scorecard with behavioural anchors your panel can agree on
Structured interview questions
Using this scorecard: [paste], write two behavioural questions and one follow-up probe for each criterion. Questions must be the same for every candidate, open-ended and legal to ask in [country]. Flag any question that could touch on a protected characteristic.
You get: A consistent interview guide tied to your criteria
Draft a policy from your notes
Draft a [hybrid working] policy for a company of [size] in [country] from these notes: [paste]. Use headings for purpose, scope, eligibility, how to request, manager responsibilities and review. Plain English, under 900 words. List any point where local employment law may set requirements I should check with a lawyer.
You get: A first draft plus a list of legal points to verify
Write a kind rejection email
Write a rejection email to a candidate who reached the final interview for [role]. Thank them specifically for [one thing they did well], say the decision was close if that is true: [yes or no], and offer [a short feedback call or written feedback]. Warm, honest, under 150 words, no clichés.
You get: A respectful rejection that protects your employer brand
Turn a handbook into an onboarding plan
From this employee handbook: [attach] and role profile: [paste], create a 30, 60 and 90 day onboarding plan for a new [role]. Include who they should meet, what they should read (quote the handbook section names), and one measurable goal per phase. End with a first-week checklist for the manager.
You get: A ready onboarding plan built from your own documents