Clear Candidate Next Steps
Candidate Agent runs inside the external candidate portal — not in internal Ask AI. Applicants get grounded answers about their application status, upcoming interviews, offer state, and open roles; anonymous visitors get help finding the right job. Strictly read-only, with anonymous mode explicitly walled off from any candidate PII.
Where The Candidate Experience Goes Quiet
Most candidate portals are status pages — log in, see a stage, log out. Candidate Agent turns the portal into something the candidate actually uses to navigate the hiring process — without sending them back to a recruiter for every question.
"Did You Get My Application?"
The candidate applied. Confirmation email landed. Then nothing for two weeks. The status page says "submitted" but doesn't say what comes next or when. The candidate emails the recruiter — multiplied by every applicant, this is most of a coordinator's inbox.
Interview Logistics Live In A Calendar Invite Nobody Re-reads
"What time is my interview? Who am I meeting with? Where do I dial in?" — the answer is in an invite from 10 days ago. The candidate searches their inbox the morning of, finds it, realizes they didn't prep for the right format, and starts at a disadvantage.
Anonymous Visitors Get The Same Generic Career Page
A passive candidate lands on the careers page. "Are there remote backend roles?" — the site shows 47 openings in a static list. Visitor leaves. The roles that matched were three of those 47, but nobody helped them find them.
Career-Site Q&A Always Goes To The Recruiter
"What's the company size? What's the engineering stack? What's the parental leave policy?" — every one of these questions can be answered from grounded company knowledge, but they end up in recruiter inboxes because the careers page doesn't talk back.
Offer Letters Trigger A Wave Of Questions With No Self-Serve Answer
The offer email lands at 6pm. The candidate has questions about equity vesting, start date flexibility, the relocation package, and the signing bonus payment schedule — and the recruiter is offline until tomorrow morning. By the time they get a callback, the candidate has already mentally compared the offer to a competing one and the momentum is gone. Grounded answers about the actual offer terms should be available the moment the email lands.
Stale Status Erodes Trust In The Whole Process
"Submitted" for 18 days. "Under review" for 11 days. The candidate refreshes the portal every morning and sees the same string they saw a week ago. They start telling friends the company "ghosts candidates." The pipeline didn't actually stall — the recruiter was on PTO, then the hiring manager was traveling — but without context, the silence reads as rejection and good candidates take competing offers.
Candidate Agent At A Glance
AI Candidate Assistant
Application status, interview schedule, grounded company answers — in the portal.
Inside Candidate Agent — The Actual Capabilities
Candidate Agent wraps the CandidatePortalAiService — a grounded RAG service running in the external candidate portal, not internal Ask AI. The agent has two surfaces (authenticated and anonymous) and is strictly read-only on both. The candidate's own portal record is the source of truth; the LLM gets context, never write permissions.
Authenticated Mode — Grounded In The Candidate's Own Record
For logged-in candidates, the agent answers from the candidate's actual application record — current status, days in process, the job applied for, upcoming interviews, and offer state. Every answer references what's true for this candidate, not generic pipeline copy.
- Application status — current stage, days in process, plain-English explanation of where the candidate is in the pipeline.
- Interview context — upcoming interview dates, format, duration, and panel composition where available.
- Offer state — proposed start date, response deadline, and days remaining to respond when an offer is live.
- Job context — title, department, location, salary range, and full job description for the role they applied to.
Anonymous Mode — Grounded Job Search For Passive Visitors
For unauthenticated visitors, the agent answers about open roles and the company itself — never about application data. Job search runs full-text and trigram-matched against actual live postings, with location, level, and salary range surfaced inline.
- Grounded job search — full-text + ILIKE + trigram matching against live job postings, never fabricated openings.
- Salary range and location — surfaced when the posting has them; not invented when it doesn't.
- Application data is walled off — the agent literally returns "log in to check application status" rather than leaking any candidate record.
- Company / role / benefits Q&A — grounded against the careers knowledge base, not hallucinated.
Interview Logistics — Reread The Calendar Without Searching For It
"What time is my interview? Who am I meeting with? What's the format?" — the candidate gets a clear, structured answer pulled from their actual interview record, not the 10-day-old calendar invite. Past interviews show outcomes; upcoming interviews show prep context.
- Upcoming interviews — date, time, type, format (video/phone/onsite), and duration where the record has it.
- Past interviews — completed sessions and outcomes for context.
- Knowledge-base + help-content blending — interview prep guides and FAQ entries surface alongside the candidate's specific schedule.
- Sources cited — KB articles and video resources referenced in answers are returned as structured sources, not just inline text.
Outcomes Teams Can Measure
Candidate Agent compresses the recruiter inbox load and gives candidates a portal that actually answers their questions. Measure against your pre-agent baseline.
- "What's my status?" emails deflected — candidate-portal sessions that answered the status question without an inbound to recruiting.
- Anonymous-to-applied conversion — share of anonymous chat sessions that resulted in a job application started.
- Interview reschedule requests — typically driven by confusion about time / format — should drop when the portal can answer the question.
- Career-site Q&A deflection — company / benefits / stack questions answered in the portal vs forwarded to the recruiter.
- Portal engagement depth — sessions per candidate over the lifetime of their application, vs the historical "login at start, never again" pattern.
Intentionally Read-Only · Anonymous Mode Is Walled Off
Candidate Agent's RISKY_TOOLS list is empty — the agent retrieves and explains, it never changes an application record, withdraws an application, or accepts an offer on the candidate's behalf. Anonymous mode is explicitly walled off from any candidate PII: the system prompt tells the model it has no access to personal application data, and the context-building service doesn't load it.
- Zero write tools — RISKY_TOOLS list is empty. No application withdrawals, no offer responses, no profile edits.
- Anonymous mode separation — anonymous visitors get jobs + KB + company info; no candidate record context is ever loaded into the prompt.
- Grounded, not generated — job search runs against live postings; status answers come from the candidate's record; "don't make up information" is in the system prompt.
- External portal, not Ask AI — Candidate Agent runs in the candidate portal, separate from the internal Ask AI surface that employees see.
WHAT TEAMS TRY INSTEAD
The four alternatives — and why none of them serve applicants AND anonymous visitors in one portal
Talent and recruiting teams trying to deflect applicant questions usually try one of these four. None of them combine authenticated status answers with an anonymous-mode job-search surface that's walled off from candidate PII.
Pasting careers-page copy into ChatGPT, Claude, or Copilot
Candidates copy a job description and ask "is this a fit for me?"
- Authenticated applicants get answers from their own application record — generic AI has no idea who's asking
- Job search runs against live postings, not yesterday's careers-page scrape
- Anonymous mode is explicitly walled off from any candidate PII — a generic chatbot has no notion of that boundary
LinkedIn Recruiter AI / Greenhouse AI / Workday Recruiting Assistant
Vendor-trapped ATS AI inside one recruiting platform
- Runs in the tenant's own candidate portal, not a vendor-branded chat that pulls candidates off the careers site
- Composes with Interview Scheduler, Offer Manager, and Background Checks — not stuck inside one ATS silo
- One agent serves authenticated applicants AND anonymous visitors — most vendor AIs do only one
A custom candidate chatbot on the careers page
An engineering team's six-month build, then forever maintenance of two auth scopes
- Shipped already. Engineering spends zero weeks plumbing authenticated vs anonymous scopes or PII-isolation guarantees
- Read-only by design — RISKY_TOOLS is empty; security review on PII-handling is a one-pager
- Inherits new capabilities (richer status surfacing, new offer-letter Q&A) as the platform evolves
The manual fallback — "email the recruiter"
Every candidate question routes to the recruiting-coordinator inbox
- The "did you get my application?" question answers in seconds — not a 24-hour recruiter-coordinator turnaround
- Interview logistics ("where do I dial in? who am I meeting?") answer the morning of, without inbox-searching
- Offer Q&A self-serves at 6pm when the email lands — momentum stays with the offer instead of slipping to a competing one
PLATFORM LEVERAGE
Candidate Agent inherits everything the platform already runs
A standalone careers-page bot has to plumb each of these. Candidate Agent gets them for free because Recruiting, Interview Scheduler, Offer Manager, and Background Checks already do.
Cross-app data plane
Application records, interview slots, offer state, and background-check status all reach the same agent — no separate sync layer from the ATS into the candidate portal.
Two-surface permission model
Authenticated mode loads the candidate's record; anonymous mode is architecturally walled off from any candidate PII — same enforcement as the underlying recruiting model.
Audit trail on every retrieval
Even read calls in the candidate portal log the candidate (or anonymous session), the tool, and the parameters — same retention as the rest of the platform.
Translation in 100+ languages
Multilingual candidates get application status, interview logistics, and offer Q&A in their own language — same translation service that powers internal apps.
Mobile delivery for applicants
Candidates on phones get answers in the same portal they applied through — no separate app to install, no email-search ritual.
External-portal architectural boundary
Candidate Agent runs in the candidate portal, separate from internal Ask AI — employees and candidates use different agents with different scopes and different audit trails.
INDUSTRY FIT
Industries where candidate experience is the recruiting moat
Candidate Agent helps wherever a strong candidate experience is the difference between accepted offers and ghosting.
Technology
Senior engineering candidates ask grounded stack and team questions in the portal instead of "is the recruiter available?" — and they get answers in minutes, not days.
Healthcare
Clinical applicants asking about credentialing timelines, shift differentials, and start dates self-serve in the portal — recruiting coordinators stop being the answer-desk.
Retail & Hospitality (High-Volume Hiring)
Hundreds of seasonal applicants get status updates and interview logistics on demand — coordinators stop drowning in "did you get my application?" emails.
Field Service & Trades
Drivers, techs, and trades applicants get clear next-step answers on mobile; anonymous visitors find the right opening without a recruiter screen.
Higher Education
Faculty, staff, and student-worker applications all share the same portal experience; anonymous visitors discover roles across schools and departments.
Public Sector
Civic and agency hiring stays defensible — anonymous mode handles citizen job-search inquiries inside FedRAMP-eligible deployment options with no PII spillage.
WHY MANGOAPPS WINS
An embedded agent beats a chatbot, a vendor add-on, or a custom build on every axis
The argument finance, security, talent, and recruiting all share — and the one a single-ATS vendor AI structurally cannot answer.
Cheaper than the alternatives
No per-seat ChatGPT license on the careers page, no LinkedIn Recruiter AI tier, no Greenhouse AI add-on, no six-month custom build.
More secure
Read-only. Anonymous mode is architecturally walled off from candidate PII — not a prompt instruction, an architectural boundary. Every call logs to AiApiLog.
Easier to deploy
Already deployed if you have the candidate portal enabled. Turn the agent on and the existing application, interview, and offer records are queryable the same day.
Easier to use
Lives inside the candidate portal — no separate chat widget, no third-party vendor branding pulling candidates off the careers site.
Easier to manage
Status visibility, anonymous-mode scope, and KB content all sit in the same admin console as every other app. One audit log, one access model.
Easier to extend
Shares the agentic tool framework with every other MangoApps agent. New surfaces (offer Q&A, relocation FAQ) ship as tools, not rewrites.
AI is actually better
A vendor ATS AI sees its own pipeline. Only Candidate Agent serves both authenticated applicants AND anonymous visitors from the same conversational surface, with PII isolation that's architectural — not a prompt.
At our head office within HR, marketing, and finance, we use ChatGPT quite a bit, and MangoApps AI Helper is a useful tool for us. As I go around to the stores, I’ve been showing them how the AI Intranet Helper can find things and answer plain language questions about our resources, and they’re quite excited about that.
Alison van Zijl
Head of Human Resources
Full House
Customer Success
Related Customer Stories
Customer Case Studies
Customer Case Studies
Customer Case Studies
Customer Case Studies
Customer Case Studies
Customer Case Studies
Frequently Asked Questions About Candidate Agent
Candidate Agent runs in the external candidate portal (not internal Ask AI). For authenticated candidates, it answers application status, upcoming and past interviews, offer state, and role context grounded in the candidate's own record. For anonymous visitors, it answers job search, company information, and career-site Q&A grounded in the careers knowledge base.
No. RISKY_TOOLS is empty — Candidate Agent is strictly read-only. It explains state, it doesn't change it. Withdrawing an application, accepting or declining an offer, and updating profile data all happen through explicit candidate-portal actions, where consent is captured.
Anonymous mode is explicitly walled off from candidate records. The system prompt tells the model it has no access to personal application data, and the context-building service never loads it. Anonymous visitors get jobs, KB content, and company information — not someone else's application status.
Job search uses three layered strategies — full-text search, ILIKE matching on title and description, and trigram-based fuzzy matching — all against live job postings in the Recruiting app. The agent never invents an opening; if there are no matches, it says so and offers to show all available positions.
"What's my status?" deflection, anonymous-to-applied conversion, interview-reschedule requests, career-site Q&A deflection, and portal engagement depth (sessions per candidate over the application lifetime). Compare against your pre-agent baseline.
Let's Talk
Since 2008, we've been building the workforce platform — earning the trust of 2 million+ users and an NPS of 78.
Why Choose Us?
- AI-Powered Platform: The most unified workforce experience on the planet.
- Top Security: HITRUST, ISO & SOC 2 certified.
- Exceptional UX: Delightful on mobile and desktop.
- Proven Results: 98% customer retention rate.
Trusted by Legendary Companies:
