Candidate Data and AI: What Recruiters Should Never Paste

Candidate data is not ordinary business data. Understanding that difference is the starting point for using AI safely in a recruitment setting, and skipping it is how agencies create problems they did not see coming.

This article covers what makes candidate material different, which categories should never enter an AI tool regardless of which tool it is, which reformatting and summarising uses carry genuinely low risk, and what to say to a candidate who asks. At the end there is a practical never-enter list a consultant can pin to their screen.

This is general guidance for data handling practices, not professional or legal advice. A qualified adviser familiar with your agency's situation is the right person to assess what applies to you.

Why Candidate Data Is Different

When someone sends you their CV, they are handing over personal information for a specific purpose: to be considered for a specific role, or a defined type of work. They made a decision about who would see it and what it would be used for. That decision did not include an AI model operated by a company they have never heard of, running on servers whose location they do not know.

Most candidates have no idea that when a consultant copies their CV into a general-purpose AI assistant, a third party is now processing their personal details. They did not agree to that. They probably would not object to it if it were explained clearly and carefully, but nobody explained it. That gap between what they expected and what happened is the core problem with careless AI use in recruitment.

Ordinary business data, by contrast, is usually generated by or about the business itself: financial reports, internal memos, client notes that the business created. Candidate data was created by an individual about themselves, handed over under an implied or explicit understanding of how it would be used, and often contains information that is sensitive in ways that go beyond what a company file would contain.

Whatever privacy rules apply where you operate, which vary by jurisdiction and are worth checking, the underlying principle is consistent: data should be used in a way that matches the reason it was collected. Candidate data was collected for placement. Using it to train, test, or query an external AI tool is a different use.

This is also a professional credibility issue. Candidates and client companies trust recruitment agencies to handle information carefully. Losing that trust is a business problem even before it becomes anything else.

Categories That Should Never Enter an AI Tool

Some information in a candidate file is sensitive in a way that makes any AI processing inappropriate, regardless of the tool, the prompt, or the intended output.

Never paste:

• National identification numbers: passport numbers, national insurance numbers, tax file numbers, social security numbers, or any government-issued identifier
• Date of birth
• Home address, including partial addresses
• Personal contact details: personal email, personal phone number
• Salary history or current pay, if it appears on the CV or in notes
• Bank or payment details of any kind
• Health information, disabilities, or medical history
• Visa or immigration status
• Information about criminal records or background checks
• Family status, caring responsibilities, or pregnancy-related notes
• Religious beliefs or political views, even if volunteered by the candidate
• Photographs, even if attached to a CV
• Notes from reference calls that contain personal detail beyond professional performance
• Any information a candidate shared informally, in a call or message, that they would not expect to appear in a processed document

The risk here is not primarily that the AI will misuse this data today. The risk is that you cannot verify where it goes, whether it is retained, whether it contributes to model training, or who might access logs. For sensitive categories, the appropriate response to that uncertainty is not to enter them in the first place.

Uses That Are Genuinely Low Risk

Not all AI use in recruitment is problematic. Several common tasks carry low risk when handled correctly.

CV reformatting is the clearest example. If a consultant takes a candidate CV, removes all identifying information (name, contact details, address, ID numbers), and pastes only the professional content into an AI tool to reformat it to an agency template, the risk is low. What remains is a work history and skills summary. It is the kind of information that would appear in an anonymised shortlist anyway.

Anonymised shortlist summaries follow the same logic. A consultant who writes a shortlist note from memory, or who pastes only the professional content after stripping identifiers, is working with information that is low-sensitivity by design.

Job description drafting does not touch candidate data at all and is simply a writing task.

Interview question generation based on a role brief, not a candidate file, carries no candidate data risk.

Internal process documents, template emails, and candidate communication drafts that do not contain personal details are all fine.

The pattern in the low-risk column is consistent: professional content, stripped of identifiers, used for a purpose the candidate would recognise as part of the recruitment process. If you can describe what you are doing in one sentence and it sounds reasonable, it probably is.

What an Agency AI Policy Looks Like in Practice

If you want to set a clear standard across your consultants, the place to start is a written policy that names what is allowed, what is not, and what the default behaviour should be when consultants are uncertain. A policy designed specifically for recruitment agencies is available at getplainstart.com/ai-policy-for-recruitment-agencies.

A policy only works if it is simple enough to follow under pressure.

What Happens on a Busy Friday

Here is a worked example. The names are fictional.

It is 4:45 on a Friday. A consultant at a mid-sized agency, call her Rachel, has a client asking for a shortlist by close of business. She has five CVs to summarise and format. She opens her AI tool and is about to paste the first CV directly.

Her agency has a rule: strip before you paste. Name, contact details, date of birth, and any ID numbers come out before anything enters the tool. Rachel has done this before and has a thirty-second process: she copies the CV into a plain text document, deletes the header block, and pastes the body.

The AI reformats the professional content. Rachel reads it, adjusts two bullet points, and adds the candidate back in her own system under the right identifier. The shortlist goes out. The client is happy. The candidate's personal details never left the agency's own systems.

This is not a perfect process. Rachel's agency would be the first to say their policy needs updating as tools change. But the rule survived contact with a real deadline because it was specific and short. Strip before you paste. That sentence fits on a sticky note.

The Never-Enter List: Pin This Up

Print this and keep it near your screen.

Never enter into any AI tool:

• Full name combined with any other personal detail
• Home or personal address
• Date of birth
• Passport, national ID, or government identifier
• Personal phone number or personal email
• Current or previous salary
• Bank or payment details
• Health conditions or disability information
• Visa or immigration status
• Criminal record information
• Photographs
• Religious or political information
• Informal notes from calls that contain personal detail

Safe to enter after stripping identifiers:

• Employment history and job titles
• Skills and qualifications
• Professional summary (written without identifying detail)
• Role requirements from a job brief
• Interview question frameworks

What to Tell a Candidate Who Asks

Some candidates will ask, especially those in data-sensitive industries or those who have read a news story about AI and privacy. A clear answer builds more trust than a vague one.

A reasonable answer sounds like this: "We use AI tools to help with formatting and note-taking. Before anything goes into an external tool, we remove your personal details so only your professional history and skills are processed. Your contact information, date of birth, and any sensitive information stays within our own systems."

If that is not true of how your agency currently operates, it is worth making it true before a candidate asks.

The question is becoming more common. Candidates who work in technology, healthcare, finance, or legal services are especially likely to ask, and they will notice an uncertain answer.

The Practical Point

AI tools can genuinely reduce the time it takes to format a CV, write a shortlist note, or draft a role brief. That is a real benefit for busy consultants. The question is not whether to use them but which data goes in and which stays out.

Candidate data was shared for a specific purpose by a real person who made a decision to trust your agency. Handling it carefully is not a compliance exercise. It is what the working relationship between an agency and a candidate is built on.