Skip to content

AI for recruiters that suggests, drafts, and organizes, while people decide

Recruitment teams are being asked to adopt AI without handing it the decisions that matter.

This page explains how an assistive AI layer inside a recruitment ATS/CRM is meant to work: it produces shortlists, drafts, parsed records, transcripts, and formatted documents, and a recruiter reviews, edits, or rejects each output before anything consequential happens. The aim is a working pattern you can evaluate, not a promise that AI replaces judgment.

The operating principles behind assistive recruiting AI

Five rules shape how the AI features are designed to behave. Read them as a test you can apply to any recruiting tool, not just this one.

  • A person makes the decision. AI may suggest, draft, and organize, but it does not reject a candidate, advance a hiring stage, or send a people-decision on its own.
  • AI activity is visible. Features are labeled in the product so users can tell when AI is contributing, rather than encountering silent automation.
  • Account data stays with the account. Data powers that account's searches, matches, and drafts, is not sold, and access follows roles and permissions.
  • Fit suggestions are inspectable. Matching works from job requirements and profile data the user can see, so a recruiter can override or ignore a suggestion.
  • Actions are accountable. Audit logs, granular permissions, and admin AI controls answer who did what, when, and with which tools enabled.

What the AI actually does, area by area

The capabilities fall into seven areas. For each, the useful question is not only what the AI produces but where the human checkpoint sits.

Finding people. AI sourcing searches external profiles and surfaces people who fit a brief; AI-powered Boolean search turns a plain-English brief into an editable Boolean string; candidate matching ranks people in the database against an open role with reasons; job-candidate matching runs the other direction; natural language search returns matching records from a plain-English request; and candidate-to-candidate matching finds people similar to one the recruiter already likes. Human checkpoint: a recruiter reviews every suggested profile, and no one is contacted or rejected automatically.

Reading documents. The resume parser turns a CV into a structured candidate record, the custom field parser agent extracts agency-specific fields, and the re-parse comparison reads the updated CV again and lays the existing record beside the fresh extraction, so the recruiter can inspect every proposed change before any field is committed. Human checkpoint: the recruiter chooses which fields update.

Drafting writing. First drafts of emails, reply suggestions, multi-step sequences, job descriptions, candidate summaries, client pitches, and submission packs are generated from the record and left for the recruiter to edit. Human checkpoint: nothing sends without recruiter approval.

Transcribing calls. Recorded calls become searchable text attached to the record, notes and call logs are summarized for confirmation, and a meeting notetaker captures notes against the record. Human checkpoint: recordings and transcripts live on records governed by roles, permissions, and retention.

Formatting documents. The resume formatting agent rebuilds a CV into a client-ready house format, redaction rules strip names, contact details, or current employer before sharing, and branding presets apply logo, fonts, and layout. Human checkpoint: admins configure presets and recruiters preview every output before it leaves.

Connecting assistants. The MCP connector lets an external assistant such as Claude query recruitment data, scoped to the connecting user's permissions, and is available on Business and Enterprise plans. Human checkpoint: access is user-initiated and permission-scoped, so an assistant sees only what that user can see.

Getting set up. Onboarding AI walks a team through setup and answers product questions. Human checkpoint: guidance only, with admins deciding what actually gets configured.

The three-step pattern to look for in any AI-assisted action

The same sequence runs across the product, and it is a reasonable template for evaluating any recruiting AI.

  1. AI does the legwork. It surfaces matches, drafts an email, parses a resume, or transcribes a call, and the output is labeled as AI.
  2. The recruiter reviews. Approve, edit, override, or ignore. Shortlisting, submitting, advancing, and rejecting remain a person's call.
  3. The action is logged. Entries land in the Audit Log with actor and timestamp, so a question about who decided something can be answered with evidence.

A pre-purchase checklist for assistive recruiting AI

Use these questions in a demo or security review. Each maps to a behavior described on this page.

  • Can an admin enable or disable individual AI features and agents, and restrict them by role?
  • Does the AI ever reject, advance, or submit on its own? The answer should be no, with the decision left to a person.
  • When an updated CV is re-parsed, is there a side-by-side compare before fields change?
  • Are AI-generated emails, sequences, and pitches held for approval before sending?
  • Does the Audit Log record actor and timestamp for AI-assisted actions?
  • Do external assistant connections inherit the connecting user's permissions rather than broader access?
  • Are recordings and transcripts governed by the account's roles, permissions, and retention settings?
  • Is candidate and client data excluded from AI model training, and is that documented in writing?

Worked example: updating a candidate record after a new CV arrives

A candidate sends an updated CV. Instead of overwriting the record, the recruiter runs a re-parse. The system re-reads the document and presents a side-by-side comparison of the existing record and the new extraction. The recruiter selects which fields to accept, leaving untouched anything that looks wrong or that the candidate did not actually change. Only the chosen fields update, and the action is recorded in the Audit Log with the actor and timestamp.

This is the practical difference between assistive and automatic: the AI produces a candidate set of changes, and a person decides what enters the record. The same review-before-commit logic applies to a drafted outreach email, a suggested shortlist, or a formatted resume before it reaches a client.

How assistive AI maps to hiring regulations

Regulators are converging on one line: AI may assist hiring, but humans must make the decision. The design choices here are intended to sit on the assistive side of that line.

  • GDPR. Data handling is described as GDPR-compliant with support for data subject rights. Because the AI does not make solely automated decisions about people, the Article 22 prohibition on such decisions is not triggered by the AI layer itself.
  • EU AI Act. Recruitment AI is treated as high-risk and requires human oversight. The suggest-review-decide loop is the oversight mechanism, built into the workflow.
  • NYC Local Law 144. The law regulates tools that make hiring decisions automatically. Here, decisions stay with recruiters by design.
  • EEO. Compliance support is intended to help agencies run equitable processes and keep documentation to prove it.
  • Audit evidence. The Audit Log answers who decided something with an actor and a timestamp.
  • Access security. MFA, single sign-on, IP whitelisting, and custom roles secure the environment the AI operates in.

Your own obligations depend on jurisdiction. The documentation is there to support your compliance program, not to replace legal advice.

What IT, security, and legal usually ask

Access and control. AI features operate inside the account's roles and permissions and see only what the user is allowed to see. Admins choose which AI features and agents are enabled. MFA, SSO, and IP whitelisting protect the account.

Data handling. Account data powers that account's AI outputs and is never sold. Processing is described as GDPR-compliant with data subject rights supported.

Provable accountability. The Audit Log captures actions with actor and timestamp. Confidentiality controls include private emails, hidden domains, and off-limits records, and custom roles restrict who can use which capabilities.

Retention and residency. Retention follows the account configuration and applicable law. Processing details and sub-processors are documented in the Trust Center, and legal questions can be answered in writing before purchase. Full technical details are in the data handling policy.

AI capability and human checkpoint map
AreaWhat AI producesHuman checkpoint
Finding peopleShortlists, Boolean strings, ranked matches, similar candidatesRecruiter reviews every suggested profile; no automatic contact or rejection
Reading documentsStructured candidate records and custom fieldsSide-by-side compare; recruiter chooses which fields update
Drafting writingEmails, replies, sequences, JDs, summaries, pitches, submissionsNothing sends without recruiter approval
Transcribing callsSearchable transcripts, note summaries, meeting notesGoverned by roles, permissions, and retention
Formatting documentsClient-ready branded resumes with redactionAdmins set presets; recruiters preview every output
Connecting assistantsExternal assistant queries scoped to the userUser-initiated, permission-scoped access only
Getting set upGuided setup and product answersAdmins decide what gets configured

Descriptions reflect the product's stated design intent; confirm current behavior and plan availability during a demo.

Frequently asked questions

Can AI reject a candidate?

No. AI can suggest matches and fit, but shortlisting, advancing, rejecting, and submitting are always actions taken by a person. There is no auto-rejection anywhere in the product.

Is my data used to train AI models?

No. Account data is used only to serve that account, powering its searches, matches, and drafts. It is not used to train AI models. For current specifics on model providers and training policies, see the Security & Trust documentation or ask directly; the answer is provided in writing.

Can I turn AI features off?

Yes. Admin Settings gives administrators a switch for each AI feature and agent, lets them tune how agents behave, and lets them gate usage by role, so the account's AI footprint can be aligned with its compliance posture.

How does this help with the EU AI Act and laws like NYC LL144?

Those frameworks target automated employment decision-making and require human oversight for high-risk AI. The AI here is assistive: it suggests, a human decides, and the Audit Log records it, which is the oversight pattern regulators are asking for. Your own obligations depend on jurisdiction; documentation is available to support your program.

What happens to call recordings and transcripts?

Recordings and AI transcripts are stored on the relevant record inside the account, governed by roles, permissions, and retention configuration. They are visible to the people you allow, and no one else.

What is MCP, and is it safe?

MCP is the connector that lets an external AI assistant, such as Claude, work with the recruitment data in your account. A user starts the connection, and it carries that account's permission boundaries, so the assistant can reach only the records the connecting user is already allowed to open. It is available on Business and Enterprise plans.