AI Resume Intake on WhatsApp: How to Limit and Delete Candidate Data

AI Resume Intake on WhatsApp: How to Limit and Delete Candidate Data

Receive resumes on WhatsApp with AI without keeping data longer than needed: minimal fields, candidate notice, deletion deadlines and access control.

•

Want to apply this in your company?

Book a call with Juan →

To receive resumes through WhatsApp with AI without keeping personal data longer than needed, first define which fields the agent may collect, tell the candidate what the data is for and how long it stays, and schedule automatic deletion of anything you no longer use. The AI can run the conversation and organize a summary, but resumes and personal data need explicit rules for retention, access and disposal. Without those rules, a channel that looked simple turns into a growing pile of personal data spread across phones, spreadsheets and third-party tools.

Key takeaways

  • Collect only what the role requires: contact details, target position, a short summary of experience and availability.
  • Tell the candidate at the start of the conversation what the data is for, who handles it and how long it is kept.
  • Set a disposal deadline for each stage and schedule the deletion, so it does not depend on someone remembering.
  • Separate the AI's organization of information from the human decision, and make clear to the candidate who decides.
  • Test the flow with fictional conversations before you publish the job opening to the public.

Which data the agent should ask for, and what to leave out

The biggest risk in this channel is not the technology. It is the open conversation. When a candidate sees a chat, they tend to send everything at once: documents, full address, a photo, marital status and even health information. If the agent accepts all of it, the company ends up storing data it does not use and should never have had.

The best protection starts with a closed list of fields, written before you give the agent any instructions. For most operational and administrative roles, the list usually includes:

  • Name and a valid contact method for a reply.
  • Position or area of interest.
  • A short summary of experience, in a few lines, without requiring a full copy of the resume.
  • Available shifts and region, when those matter for the role.
  • Salary expectation, only if the role asks for it, and as an optional field.

Keep personal documents such as national ID and tax ID numbers, full home address, photos, marital status, children, religion, origin and health data out of the initial screening. None of these help decide whether someone moves forward at this stage, and asking for them early increases the company's responsibility. If a candidate sends any of this unprompted, the agent should explain that these items are not needed at this stage, and the system should remove the attachment from the workflow without passing it to the recruiter.

The notice to candidates: what to say before the file

The notice must appear before the candidate sends any information. It is not just a formality. It is the moment when the person understands who they are talking to, what the data will be used for and how they can request deletion. A useful notice includes at least the following points:

  • Who is responsible for the job opening and for keeping the information.
  • The specific purpose, which is to assess the application for that role.
  • A statement that the conversation is handled by an AI assistant and that people on the team review it before any decision.
  • Which data will be requested.
  • The retention period and the channel for requesting deletion before that period ends.

An opening message can be short and direct:

You are talking to an AI assistant. Your data will be used only to assess this position and will be deleted within the period stated in this conversation.

The same care with notice and purpose applies to other automated records. Our article on recording meetings with AI and privacy law covers the same principle in another context.

Disposal deadlines: where the data lives and when it leaves

A common mistake is setting a deadline while only thinking about the main database. In practice, the same resume can exist in several places at once: the recruiter's phone conversation, the platform that connects to WhatsApp, activity logs, the tracking spreadsheet, a forwarding email and backup copies. Before choosing any deadline, map these locations and assign a retention rule to each one.

A practical separation, which should be adjusted with legal advice, is to treat three different situations:

  • Candidate not selected: the data leaves the workflow at the end of the defined period, and the candidate can request deletion before then.
  • Candidate in an active selection process: the data stays for the length of that stage, with a record of who accessed it and when.
  • Candidate hired: the data becomes part of the HR routine, with its own legal basis and rules, and is no longer treated as a WhatsApp screening record.

The deadline should be justified by the purpose and documented in the notice. The Brazilian National Data Protection Authority publishes guidance and is a useful reference for understanding how LGPD principles apply to this type of processing. The central point is that disposal happens automatically, with a record of the deletion, and not through a manual action someone has to remember.

For logs, a simple rule applies: record decisions and metadata, such as date, stage, reason for the recommendation and status, but do not store the full content of resumes in observability tools. When a case needs investigating, the team checks the original conversation within the retention period, not an old log file nobody remembers.

A six-step flow for a pilot

If the company wants to start without excessive exposure, a pilot with a single job opening is the safest format. The sequence below works as a reference:

  • First, map the role and write the list of minimum fields before you touch any agent instructions.
  • Second, draft the notice and the reading confirmation, and place them before the first question.
  • Third, configure the agent to collect only the listed fields and to decline unrequested attachments.
  • Fourth, generate a structured summary for the recruiter, without attaching the full conversation transcript.
  • Fifth, schedule deletion by stage and log each deletion without personal content.
  • Sixth, review a sample of conversations with the recruiter and confirm that the agent followed the field list and the notice.

One technical point: the agent can only reply freely within the service window opened by the candidate's own message. When the company needs to contact the candidate again later, the message must follow WhatsApp Business rules and the recipient's consent. This is covered in our piece on the 24 hour WhatsApp API window, which is worth reading before you design any follow-up flow.

Illustrative scenario: a stockroom assistant opening

Illustrative scenario: a building materials store opens a stockroom assistant position and configures the agent to ask for name, phone number, neighborhood of work, a summary of experience and shift availability. A candidate, without being asked, sends a PDF that includes a photo, a tax ID number and a full home address. The agent thanks them, explains that these items are not needed at this stage and says the attachment will be removed from the workflow along with the rest of the conversation at the stated deadline. The recruiter receives only the summary with the listed fields. When the position is filled, the candidates who were not selected have their data deleted on the scheduled date, and the log records only the date and stage of each deletion.

In this example, nothing sensitive reached the store's spreadsheet, and the team did not have to delete anything by hand. This is the kind of result the flow design controls, and it should be verified during the pilot, not assumed.

Who decides, and how a candidate asks for a review

AI screening can organize applications, but removing a person from the process has real consequences. The practical rule is to keep a human in charge of any elimination decision and let the AI handle organization. The NIST AI Risk Management Framework offers a structured way to map these risks before a system goes into production.

A simple task split helps the team stay within the boundary:

  • Can be automated: collecting the listed fields, answering questions about the role, organizing the summary and scheduling deletion.
  • Needs human review: rejecting an application, ranking candidates by score and responding to review requests.
  • Should not be done by the agent: evaluating personal aspects unrelated to the job, such as appearance, assumed age or health data.

Candidates need to know whom to ask for a review of a rejection or a classification. To see how this right connects to the LGPD, read our article on automated decisions and Article 20.

How we apply this method in projects like this

In projects of this kind, we follow the same method we use for other customer and candidate workflows: map the current process and its baseline, limit data and actions, pilot a real flow with logs and a success metric agreed before launch, then hand over the operation and decide whether the case should be expanded, adjusted or closed. In the limitation phase, the field list and the disposal deadline are defined before any configuration. In the pilot phase, the metric is usually operational, such as time to first response and the number of conversations containing data outside the list, not a promise of hiring outcomes. No outcome is guaranteed in this kind of initiative, and the decision to expand depends on what the pilot shows.

Frequently asked questions

Is WhatsApp an appropriate channel for receiving resumes?

Yes. The channel is common and can be used with clear rules. The risk lies less in the app itself and more in what happens around it, such as backups, spreadsheets and third-party tools. Define what comes in, what goes out and where each piece of information is stored.

Do I need consent to use AI in screening?

It depends on the legal basis you adopt and the context of the processing, and this should be validated with a lawyer. Whatever the basis, stating the purpose, the use of AI and the channel for review is a sensible practice and fits the spirit of the LGPD. Do not treat this notice as a substitute for legal analysis.

How long should I keep the resume of someone who was not selected?

The company should set the period based on the purpose and with legal guidance, and it must be documented in the notice. The most important things are that a period exists, that it is applied automatically and that the candidate knows how to request deletion before it ends.

Can I use conversation history to train another model?

Do not assume that this is allowed. Check the terms and settings of your AI provider, avoid sending unnecessary personal data and do not include transcripts in training datasets without a documented legal basis. If the provider processes data outside Brazil, also review the rules for international data transfer.

Sources

WhatsApp recruitingAI data retentionLGPDCandidate privacyHiring automation