How to use Claude AI for recruiters: workflows and limits

Most recruiters who try Claude start the same way: they paste a job brief in, get a usable advert back in seconds, and wonder what else it can take off their plate. The harder question is not what Claude can write but what a recruiter is allowed to hand to it. The UK Information Commissioner's Office ran audits of AI recruitment tool providers and produced almost 300 recommendations, finding tools that inferred gender and ethnicity from candidate names and kept candidate data indefinitely without candidates knowing.

That finding matters for anyone using a general assistant on candidate data. Claude is a capable drafting and analysis tool, and it can save a recruiter hours a week. It is also a general model, not a compliant hiring system, and the legal duties sit with the recruiter or employer who uses it. This guide covers the workflows that work well, the ones to avoid, and the rules that apply in the UK, the EU and the US.

Quick answer

Recruiters can use Claude well for job descriptions, outreach drafts, interview question banks, structured note summaries and CV summarisation that a human then reviews. They should not let it score, rank or reject candidates unsupervised. Employment is a regulated, high-risk area under Anthropic's own usage policy, the EU AI Act and several US rules.

Where Claude helps recruiters

The strongest uses are the ones where the output is text a human will read, edit and own. These carry little regulatory weight because the recruiter makes the decision and Claude only speeds up the writing or the reading.

Job descriptions and adverts

Give Claude the hiring manager's rough notes, the salary band and your house style, and ask for a first draft. It is good at turning a rambling intake call transcript into a clear role summary, and at rewriting a dense advert in plain language. A useful second pass is to ask it to flag wording that could discourage some groups from applying, such as heavily gendered phrasing or unnecessary degree requirements. Treat the flags as prompts for your own judgement, not as a verdict.

Outreach and candidate communication

Personalised first messages, follow-ups after a screening call, rejection emails that are kind and clear, and interview logistics are all safe, high-volume tasks. The recruiter supplies the facts about the role and the person, Claude supplies the phrasing, and the recruiter sends it. Keep a short style guide in the prompt so messages sound like your firm rather than a template.

Interview preparation and notes

Ask Claude for a structured question bank tied to the competencies in the role, then use the same competencies to summarise interview notes. Structured notes are easier to compare fairly than free-form impressions. The summary should record what the candidate said, not a prediction of how they will perform, and the hiring manager should still read the raw notes.

CV summarisation, with care

Summarising a CV into a one-page profile in a consistent format is a legitimate time saver. The risk starts when summaries are used to compare people against each other or to decide who gets an interview. At that point the task has moved from drafting into selection, and the rules in the next sections apply in full.

What recruiters should not hand to Claude

Three tasks deserve a firm line. The first is scoring or ranking candidates and rejecting those below a threshold. The second is inferring characteristics such as age, gender, ethnicity or likely career breaks from names, photos or CV gaps. The third is making any final decision about a person without a human who has the authority and the information to change the outcome.

The ICO's audit found tools doing exactly the second thing, inferring characteristics including gender and ethnicity from a candidate's name, and told the providers to collect accurate information directly from candidates instead. A general assistant has no built-in guard against this. If you ask it to rank twenty CVs, it will, and it will sound confident while doing so.

Anthropic's own rules point the same way. Its usage policy lists employment, including screening, ranking, advancing or rejecting candidates, as a high-risk area. It requires that a qualified person meaningfully reviews the output, with the authority to change it, before it is used to implement a decision, and that the affected individual is clearly told AI was used. That is a contractual and policy expectation on top of the law.

UK rules: data protection and automated decisions

In the UK, the ICO's practical guidance for recruiters starts with a data protection impact assessment. It says one should be carried out prior to using an AI tool, ideally at the procurement stage. The same guidance asks recruiters to pick a lawful basis, identify who is controller and who is processor, seek assurances on bias, tell candidates how their data is used, and collect only the minimum personal information needed.

Using Claude does not remove those duties. If you paste candidate CVs into a tool, you are processing personal data, often including special category data such as health or ethnicity if a candidate volunteers it. You need a lawful basis, a clear privacy notice that mentions AI-assisted processing, and a contract with the provider that settles the controller and processor roles.

Automated decision-making is the second area. The Data (Use and Access) Act 2025 relaxed some restrictions, and in March 2026 the ICO said it wants firms to be able to use automation responsibly, but it stressed that proper safeguards must be in place. Employers should test regularly for biased outputs, tell candidates when automated decision-making is used, and explain how a candidate can challenge a decision and ask for human review.

The practical reading is simple. A recruiter who uses Claude to draft and summarise, and who personally makes every shortlisting decision, is in a much safer position than one who lets a script score applicants and drop the bottom half. If you are unsure which side of the line a workflow falls on, assume that anything which filters people out without a human reading their application counts as automated decision-making.

EU rules: the AI Act and high-risk hiring tools

The EU AI Act treats hiring as a sensitive area. According to the European Commission's AI Act Service Desk, Annex III covers AI systems used for the recruitment or selection of natural persons, with examples including job-matching and ranking tools, candidate sourcing tools and AI scoring of applicant answers. The page also notes that systems which perform profiling cannot rely on the narrow exceptions.

Timing has shifted. The Digital Omnibus deal on AI sets 2 December 2027 as the application date for stand-alone high-risk systems in the Annex III areas, instead of the original 2 August 2026. Check the current status in the Official Journal before you plan around any date, because it governs when provider and deployer obligations start to bite.

For a recruiter, the point is about role as much as date. If you build or configure a tool that ranks or scores candidates, you may be treated as a provider or deployer of a high-risk system, with duties around risk management, data quality, logging, transparency and human oversight. Using Claude only to write adverts and emails sits far from that classification. Using it to rank applicants sits squarely inside it.

US rules: Title VII and city-level audits

In the United States, the Equal Employment Opportunity Commission has published technical assistance on how Title VII applies to AI and other automated selection tools. The agency released a document on adverse impact in automated employment selection tools and urged employers to conduct ongoing self-analysis of the technology they use. Employers can be responsible for discriminatory outcomes even when a vendor built the tool.

Some cities add their own rules. New York City's Local Law 144 requires employers using automated employment decision tools to have a bias audit done no more than a year before use, post a summary of the results and notify candidates. A December 2025 review by the New York State Comptroller noted that enforcement was thin, with only two complaints received in the audit period, but civil penalties of 500 to 1,500 dollars per day remain available. Thin enforcement is not a reason to skip the audit.

Recruiters working across borders should apply the strictest standard they are exposed to. A London agency placing candidates with a New York employer, or an EU subsidiary of a US firm, can find several regimes applying to the same shortlist.

Data handling when you use Claude

Before any candidate data goes into Claude, check which plan you are on. Anthropic states that, by default, it will not use inputs or outputs from its commercial products to train its models, with exceptions where you send feedback or opt in. That default applies to commercial products such as Claude for Work and the API, not to every consumer plan, so a recruiter pasting CVs into a personal account is in a different position from one on a managed team plan.

Whatever the plan, apply ordinary data minimisation. Strip names, contact details and photographs where the task does not need them. Do not paste special category data. Keep a record of what categories of data go in, what the output is used for and who reviews it, because that record is the backbone of your impact assessment.

A short internal policy helps. It should say which tasks are approved, which are banned, which accounts may be used and who signs off on exceptions. Our guide to GDPR and custom AI agents covers the data protection side in more depth.

A safe workflow for a recruiter using Claude

A workable pattern has five steps. First, define the role and the competencies in writing, with the hiring manager, before any AI is involved. Second, use Claude to draft the advert and outreach from that definition. Third, when applications arrive, have a person read them, using Claude only to produce consistent summaries in a fixed template. Fourth, record the human decision and the reasons in your ATS. Fifth, test your outcomes by stage and by group at regular intervals, as the ICO advises, and investigate gaps.

That pattern keeps Claude in the role of a fast assistant. The judgement stays with the recruiter, the candidate can be told plainly how AI was used, and the audit trail shows a human at every decision point.

How Klevere approaches AI for recruitment teams

Klevere builds custom AI agents for recruitment businesses rather than asking recruiters to wire a general assistant into their own process. Our AI recruitment agent work starts with a review of which tasks are drafting and admin, where an agent can run with light oversight, and which are selection decisions, where a named person must stay in charge. Agents are built around your ATS and your policies, with logging so you can show who reviewed what.

For agencies, the starting point is usually the admin around the placement: candidate updates, CRM hygiene, booking interviews, chasing references and keeping records accurate. Our page on AI for recruitment agencies sets out how that looks in practice, and our Bullhorn AI integration work connects agents to the system your consultants already live in. If you want to see what a dedicated agent looks like next to a general assistant, the recruitment agent page walks through it.

We have deployed more than 500 AI agents across 50 plus projects in 12 industries, and the pattern is consistent: the teams that get lasting value are the ones that decide early where humans stay in the loop. For a wider view of people operations, see our AI for HR playbook.

Frequently asked questions

Can recruiters use Claude to screen CVs?

Recruiters can use Claude to summarise CVs into a consistent format for a human to read. They should not use it to score, rank or reject candidates without meaningful human review. Anthropic's usage policy treats screening and rejecting candidates as high-risk, and UK, EU and US rules all expect oversight, transparency and a way for candidates to challenge outcomes.

Is it legal to put candidate data into Claude?

It can be, if you have a lawful basis, a privacy notice that covers AI-assisted processing, a suitable contract with the provider and a data protection impact assessment. Use a managed commercial plan rather than a personal account, minimise what you paste, and avoid special category data unless a specific legal condition applies.

Does the EU AI Act apply to recruiters using Claude?

It applies to systems used for recruitment or selection, such as ranking and scoring tools. Using Claude to draft adverts or emails sits outside that. If you configure it to rank candidates, you may take on provider or deployer duties. The high-risk application date has been agreed as 2 December 2027, but verify the current position.

What is the best way to use Claude for outreach?

Give it the role facts, the candidate's relevant background and a short style guide, then ask for two or three variants. Edit the one you prefer, check every factual claim, and send it yourself. Avoid pulling in personal details the candidate has not made public, and keep a record of the data sources you used.

Will Claude introduce bias into hiring?

It can. Language models reflect patterns in their training data, and a ranking request can amplify them. The ICO advises testing regularly for biased outputs and asking providers about their bias testing. The safest approach is to keep Claude out of selection decisions and to monitor outcomes across groups at each stage.

Do I need to tell candidates I use AI?

In most regimes, yes where AI plays a meaningful part in decisions, and good practice even where it does not. The ICO says candidates must be told how to challenge a decision and ask for human review. Anthropic's usage policy also expects affected individuals to be clearly told when AI produced a recommendation.

If you want a clear view of which recruitment tasks are safe to automate and which need a human, book a free AI audit and we will map your workflows against the rules above.