AI Tools for HR: How to Pick the Right Tool for Each HR Task and Check What It Gives You

Most lists of AI tools for HR are out of date by the time you read them: products get bought, renamed and closed. So I do not give you a ranking. I give you the map I use in my course: HR task → type of AI tool → how to check the result. Types change much slower than brands.

If you already have a task in mind, go straight to the free AI tool picker.

AI tools for HR are software that drafts, sorts, summarises, answers or acts inside HR work. In practice AI lives in three places:

• Inside the systems you already use. Your HRIS, ATS or LMS has AI features — for example Ask BambooHR or Personio Assistant answer questions about policies.
• A separate service built for one step: checking the wording of a job ad, taking notes in an interview, building a quiz.
• Your own chat — a general assistant: ChatGPT, Claude, Gemini or Copilot.

On top sit agents that connect several systems — for example a custom GPT plus Notion or Google Sheets, or an agent plus Slack plus your ATS.

The real difference between the types is who presses the button. In an assistant, you do; in your ATS, the system does, by your rules; an agent does it on its own.

Embedded AI is like cruise control in a car: already installed, already paid for, never switched on. So the first step is not a demo of a new product. Open the admin panel of your HR system, see which AI features are there, and ask your vendor three questions: which AI features are in my current plan, which are sold separately, and is there a monthly limit on requests?

Then place your team on a scale of AI maturity: 0 — AI is not used; 1 — a personal chat from time to time; 2 — set-up assistants with the company's knowledge base; 3 — AI works inside the working systems; 4 — agents perform actions.

My rule: until you reach level 2, do not buy new systems. Get there on existing tools first.

Take one task and check the conditions from the task router in my course: sensitive data (salaries, ratings, personal or medical data); company context needed every time; the result must go into a process, not into a file on a drive; permissions — different people see different things; an action in the system, not a text; and data in three or more systems.

The route:

• Three or more of the first five — the AI inside your HR system. Context, permissions and the action are already there.
• Fewer — a general assistant. Even one condition is a reason to look at your system's AI first.
• Data in three or more systems outweighs everything: neither embedded AI nor a chat will close it. It needs an orchestration layer — the longest launch and a separate contract, for when your stack has settled.
• A person decides, or the task exists for a conversation — do not automate. AI prepares the material.

And before you choose any separate service, ask: what will I do with this data if the service is switched off tomorrow?

The map for the main HR functions: which type of tool, what to give it, how to check it.

Recruiting. Job descriptions, sourcing search strings and messages to candidates are assistant work. For a job ad give five tasks of the role, five must-haves and three stop factors — things that rule a candidate out whatever their strengths. Services such as Textio keep one standard of wording across all job ads; a prompt does not. Resume screening has two paths: inside the ATS the system scores applications by your criteria, but only those inside it; in an assistant you upload anonymised resumes of any kind, but by hand each time.

Onboarding and employee questions. Count the volume first: 40 questions a week × 6 minutes = 200 hours a year, in a course illustration. Those questions go to a bot built on your knowledge base, in your assistant or in your HRIS, living where people already talk: Slack, Microsoft Teams or Telegram. Reminders by start date go in the HRIS; without one, Zapier or Make can trigger them from a spreadsheet row.

Learning. Any material with content is an input: a policy, an instruction, a webinar transcript, a presentation. The assistant builds the structure, the lesson and the quiz; course builders and video platforms package it.

Performance. Reviews take long not because of writing but because of remembering: 20 employees × 40 minutes = 13 hours per cycle, in a course illustration. AI collects the facts, drafts the review and summarises 360 feedback — for example, the vendors of Effy AI and Culture Amp describe AI summaries of feedback. The rating and the conversation stay with the manager.

Analytics. An assistant can work with an anonymised export — department, role, grade, tenure in months, exit date, exit type. Ask about one hypothesis, not “what is interesting here”.

Employer brand and communications. Give 20 reviews as text and ask for five themes, the number of mentions and one verbatim quote for each. When you adapt a piece for several channels, the fact and the number never change.

One main tool instead of five subscriptions. The decision tree from my course:

• Your work lives in Google Workspace — Gemini.
• Your company is on Microsoft 365 with a licence — Copilot.
• You work with many files and spreadsheets — Claude.
• In all other cases — ChatGPT.

Keep a second one on a free plan to cross-check answers.

An assistant without a knowledge base is a new employee on day one: smart, but does not know your ways. Set up your own — GPTs, Claude Projects, Gems or Copilot Agent Builder — and load four things: your policies, job profiles, email templates and three examples of texts you like. The typical first-run mistake: the instruction describes the assistant, not the task. How to split HR work into tasks — in AI HR Architect: Breaking Down HR Tasks for AI.

I have taught AI in sourcing for about three years, and students sent me every kind of search string from their homework. Sometimes ChatGPT writes a query that makes you think “wow”, sometimes a very primitive one. The quality is unstable, so the check is part of the work.

Four checks work for almost any HR task:

1. Ask for the source. A made-up answer is caught in one move: ask which document the answer comes from.
2. Recount one number by hand. Ask the same question in other words. If you cannot reproduce a figure in a spreadsheet in five minutes, do not show it.
3. Open three links yourself. A Deep Research report comes with sources; check that they say what it claims.
4. Explain it without the system. Can I explain this rejection or decision in my own words, without referring to the system?

Remember that the company answers for what its bot says. A Canadian tribunal ordered Air Canada to refund a customer the fare difference its chatbot had promised against the airline's rules.

Data. Anonymising is like removing the tags from clothes before the dry cleaner. Open the file and try to recognise a person: if you can, it is not anonymised. Never upload names with salaries or medical data. Find the setting that allows training on your chats — on personal plans it can be on by default. One question before each upload: would I be fine if a stranger read this file?

Tasks. Four kinds stay manual on purpose: rare ones, where the set-up takes longer than five years of doing the task; low-value ones — ten minutes a month saved, two weeks spent on the launch; high-risk ones without human control — dismissals, disciplinary decisions, the final rating, salary decisions, rejecting a candidate on a formal criterion; and everything that needs a conversation. If a task exists for contact between people, automation removes not the work, but the point.

AI can prepare material for any of these decisions. The decision stays with a person, because a person answers for the consequences.

AI tools for HR are not a list of brands. They are a route and a check for every task. The full path, from your HRIS and your own assistant to every HR function and agents, is in my AI in HR Course: Artificial Intelligence Training for HR.