Candidate Experience Audit: How to Find and Fix the Weak Spots in Your Hiring Journey

TL;DR

• A candidate experience audit reviews your hiring journey stage by stage, from the candidate’s side. It uses ATS data and candidate feedback to check four things: how long each step takes, how much effort it asks, what candidates are told, and whether everyone hears back.
• Run it in seven steps:
  • Set the scope and baseline.
  • Map the journey.
  • Mystery-shop it.
  • Survey candidates, including rejected ones.
  • Audit your AI and automation.
  • Score and prioritise.
  • Fix and re-measure.
• The 2026 twist: the biggest new risk is automation that candidates aren’t told about, can’t appeal and never hear back from. In a May 2026 Greenhouse survey, 38% of job seekers said they’d abandoned a hiring process because of AI (vendor survey).
• Where automation helps and hurts: it speeds things up, reduces effort and lets you close the loop with every candidate. It hurts when AI is undisclosed, there’s no human option, or rejections come with no explanation.
• Tools by stage:
  • Screening: Sapia.ai, HireVue, Greenhouse Voice AI, Eightfold
  • Applications: Paradox, Phenom
  • Scheduling: GoodTime, ModernLoop
  • Measurement: Starred, Survale

Completion rates slipping, offer acceptance falling, reviews that mention silence: these are symptoms. A candidate experience audit tells you where they come from.

The pressure has changed shape. LinkedIn was handling around 11,000 applications a minute in mid-2025, up 45% year on year (LinkedIn data via eWeek). Employers have responded with automation, and candidates don’t trust it: only 26% trust AI to evaluate them fairly (Gartner, independent analyst, July 2025).

Most audit guides predate all this. This one gives you:

• six criteria to score against
• a seven-step method you can run in a quarter
• a dedicated step for auditing AI
• a shortlist of tools mapped to what audits usually find

A disclosure: Sapia.ai makes one of the tools below. We’ve assessed 10M+ candidates (vendor-reported) and publish an independent audit of our own scoring, so we’ve held ourselves to the same questions.

The six things a candidate experience audit should score

Score every stage, and every tool, against the same six criteria. They’re based on what practitioners such as SHRM, the CandE Benchmark programme and Starred look for, updated for 2026.

1. Speed and responsiveness. Measure time to first response, time in each stage, and time from decision to telling the candidate. The 2025 CandE award winners left 61% fewer candidates waiting two months or more. Making offers within a week raised candidates’ willingness to refer by 108% (CandE/Survale, 2025; the programme is run by a survey vendor).
2. Effort and accessibility. Check four things: application length, mobile usability, forced logins, and whether steps are timed or need a camera or voice. 35% of job seekers would abandon an application that takes too long (Employ via SIA, vendor survey, April 2025).
3. Clarity and transparency. Do candidates know the steps and timeline? Are they told when AI is involved?
   • 37.8% of candidate comments concern expectation-setting, and 74.5% of those are negative (Starred 2026, vendor data).
   • 70% of candidates who had an AI interview weren’t told upfront (Greenhouse, May 2026, vendor survey).
4. Fairness and evidence. For any step that scores or filters candidates, ask four questions:
   • Is the score built from the candidate’s own answers (CV-blind), answers blended with CV data, or a profile match?
   • Is there an independent, published bias audit of this tool, analysed role by role?
   • What validity evidence is published?
   • Can candidates ask for a human?
5. See our take on human-in-the-loop hiring.
6. Closing the loop. Does every candidate get an outcome, and something useful?
   • 61% of job seekers have been ghosted after an interview (Greenhouse, Dec 2024, vendor survey).
   • Rejected candidates give a candidate net promoter score (cNPS) of –7, against +17 overall (Starred 2026).
   • Specific feedback raised willingness to refer by 69% (CandE 2025).
7. Measurement and stack fit. Can you track cNPS by stage and outcome (hired, rejected, withdrawn) and tie it to ATS data? For tools, also check ATS integration, pricing and time to value.

How to run a candidate experience audit in seven steps

Step 1: Set the scope and pull your baseline

Pick one or two role families, such as frontline volume roles and graduates, rather than auditing the whole company. From your ATS, pull:

• applications started vs completed, by device
• drop-off at each step
• time in each stage
• how long candidates wait to hear their outcome
• the share with no outcome after 30 and 60 days
• offer acceptance
• withdrawal reasons

Output: a one-page baseline for each role family.

Step 2: Map the journey

Write out the stages in order:

1. Attract
2. Apply
3. Screen
4. Schedule and interview
5. Offer
6. Reject
7. Onboard

List every touchpoint at each stage, and flag every one that is automated or AI-driven, such as auto-replies, knockout questions, ranking, AI interviews and automated rejections. Our candidate experience playbook is a good starting point.

Output: a stage map with every automated touchpoint marked.

Step 3: Mystery-shop it on a phone

Apply to your own roles on a phone, logged out. Time each step, and note every unclear instruction, dead end and silent gap. Then repeat with a deliberately weak application, so you experience the rejection path most candidates take.

Output: a first-hand list of friction points, with timings.

Step 4: Survey candidates, including the ones you rejected

Send short automated surveys at three points: after they apply, after interview, and on rejection or withdrawal. A cNPS question plus one or two open questions is enough.

Prioritise rejected candidates. Starred reports that 69.49% of rejections happen at the application stage, and it gets a 31% response rate from rejected candidates (vendor data). Our guide to candidate satisfaction surveys covers what to ask.

Output: stage-level feedback that explains your baseline numbers.

Step 5: Audit your AI and automation

Most audit guides miss this step, and in 2026 it carries legal weight. For every automated touchpoint from Step 2, ask the questions below. This isn’t legal advice, so confirm your obligations with counsel.

Are candidates told?

• EU AI Act Article 50 transparency duties apply from 2 August 2026.
• Illinois HB 3773 has required notice since 1 January 2026.
• Colorado SB 26-189 requires notice and a plain-language explanation of adverse decisions from 1 January 2027 (Crowell).

What is the score built from? Ask each vendor to state in writing whether it uses answers only, answers blended with CV data, or a profile match.

Is there a role-by-role independent audit? Get the auditor and the year. A 2026 study of around 4 million pymetrics-screened applications found racial disparities at role level that pooled audits had masked (Stanford Report). Ask every vendor, Sapia included, for role-by-role results.

Can candidates reach a human or contest the outcome?

• 46% of candidates want a human-review option (Greenhouse 2026, vendor survey).
• The UK’s Data (Use and Access) Act has given candidates the right to human intervention and to contest automated decisions since February 2026, and the ICO is reviewing employers’ automated hiring (ICO).
• In the EU, GDPR Article 22 still restricts solely automated decisions with significant effects.

Is it accessible? Timed, camera-only or voice-only steps exclude people, and response-time limits can disadvantage non-native speakers (University of Melbourne, 2025).

Do automated gates create adverse impact? Check pass-through rates by group at each one.

Output: a register of your AI and automation. For each touchpoint, record the vendor, whether it’s disclosed, what the score is built from, the audit evidence, the route to a human, and accessibility.

Step 6: Score and prioritise

Score each stage 1–5 against the six criteria, and plot fixes by impact and effort. The usual top findings are:

• silent rejections
• slow scheduling
• over-long applications
• undisclosed AI screening

Output: a ranked fix list, with owners.

Step 7: Fix, then re-measure

Fix the top three. Re-run the baseline and surveys after one hiring cycle. Re-audit at least quarterly for volume hiring, and after any new tool.

The candidate experience audit table

Use this as your checklist for each stage. Treat the red flags as starting points, not benchmarks.

How automation can improve candidate experience (and where it makes it worse)

Automation isn’t good or bad for candidate experience in itself. What matters is what you automate, and how.

Where automation improves candidate experience

• Speed (criterion 1). Self-scheduling, reminders and status updates remove the waiting candidates hate most. CandE award winners use automation to tell candidates their outcome within days, not weeks.
• Effort (criterion 2). Form-free mobile applications and untimed screening fit around candidates’ lives. Paradox reports 72% average application completion (vendor-reported), and Sapia’s chat interview reached 82.6% at Woolworths (customer-reported).
• Closing the loop (criterion 5). At volume, automation is the only realistic way to give every candidate an outcome, and something useful. Every candidate who completes a Sapia interview gets personalised feedback.
• Measurement (criterion 6). Surveys triggered at each stage reach rejected and withdrawn candidates, not just hires.

Where automation hurts, and how to avoid it

• Undisclosed AI with no human option. 38% of job seekers have abandoned a process because of AI. The top trigger was AI-scored video with no human review (Greenhouse, May 2026, vendor survey). Fix: disclose upfront, explain what’s scored, and offer a person.
• Rigid bots. Reviewers complain about questions that “fall outside the automated path” and voice bots that glitch on background noise (G2, Sep 2026). 47% of candidates say chatbots make recruiting feel impersonal (CareerPlug, 2025, vendor). Fix: an easy route to a human.
• Rejection with nothing back. Rejected candidates call automated processes impersonal, even “lazy” (Starred 2026). Fix: personalised feedback, not a template.
• Inaccessible formats. In March 2025 the ACLU filed a complaint on behalf of a deaf applicant who said she was denied captioning in a HireVue interview (HR Dive). HireVue says its AI assessment wasn’t used, and the complaint is an allegation, not a finding. Fix: untimed options, text alternatives and a clear accommodation route.

The rule of thumb: automate the logistics and the loop, keep people in the decisions, and audit every automated gate.

Candidate experience automation tools, mapped to what your audit finds

This isn’t a ranking of every candidate experience platform. It’s a shortlist that matches tools to the problems audits most often find, grouped by the stage they fix. For each tool that scores candidates, we give its evidence profile:

• what the candidate produces
• what the score is built from
• what validity evidence is published
• what independent audit exists
• what integrity controls it has

Facts as of September 2026. Several vendors have launched, acquired or re-priced products in the last 18 months, so check each vendor’s current site before you buy.

Stage 1: Screening and interviews (where audits find the most damage)

Most rejections happen here, rejected candidates are the least satisfied group, and AI screening is now the most-cited reason candidates abandon a process. Step 5’s questions matter most at this stage.

Sapia.ai

Best for: high-volume, frontline and graduate hiring where your audit shows drop-off, silence or fairness questions at screening, on top of the ATS you already run.

What it does: every applicant completes an untimed, mobile, text-based structured chat interview. Everyone gets the same job-related questions, can stop and come back later, and receives personalised feedback afterwards. Async video is available as a second stage. Sapia also offers:

• automated interview booking
• ATS integrations, including SuccessFactors, SmartRecruiters and Avature
• the Talent Intelligence Assistant (TIA), which finds strong past candidates for new roles based on what they said in their interviews

Evidence profile:

• Candidate produces: written answers to structured interview questions.
• Score built from: the candidate’s answers only – not their CV, school or employers.
• Published validity: published research, less extensive than SHL’s.
• Independent audit: BLDS, late 2022. Its 72 tests found no evidence of practically significant disparate impact. US and Canada models only.
• Integrity controls: AI-answer detection – 98% detection and a 1% false-positive rate (vendor-reported).

Strengths: feedback for every candidate at scale; no camera, no voice and no timer; CV-blind scoring with a named auditor; 82.6% completion and 9.2/10 candidate experience at Woolworths (customer-reported).

Where it falls short:

• It fixes one stage, not the application, scheduling or survey stages.
• Its validity documentation is thinner than SHL’s.
• The audit is from late 2022 and covers US and Canada models only. Ask us for newer, role-by-role results.
• A written format can disadvantage candidates who find writing hard. AI-text detectors can also wrongly flag non-native English writers (Liang et al., 2023), so ask us for that group’s false-positive rate.

Pricing: quote-based.

Verdict: the strongest fix when your audit points at screening. Pair it with the tools below for the other stages.

HireVue

Best for: enterprise and graduate volume hiring that wants video, voice and assessments with the longest science record among AI interviewers.

What it does:

• a two-way voice AI Interviewer, launched June 2026 on Hireguide technology acquired in March
• on-demand and live video interviews
• assessments and games
• scheduling automation

Scale: 1,150+ customers (vendor-reported).

Evidence profile:

• Candidate produces: voice, video, game and test responses.
• Score built from: for video, “only what is said by the candidate” (2024 explainability statement). Not stated for the AI Interviewer.
• Published validity: machine scores compared with expert ratings (vendor-reported).
• Independent audit: DCI Consulting, 2023, pooled across employers. None for the voice product.
• Integrity controls: fraud controls, with no published figures.

Strengths: the deepest science record in this group; a published statement that video scoring uses answers only; one platform from screening to scheduling.

Where it falls short:

• No audit for the voice product.
• One-way video and voice are the formats candidates most often abandon without human review.
• Reviewers report broken candidate links and problems with ATS and calendar integrations.

Pricing: quote-only.

Verdict: the science-heavy enterprise choice for video and voice. Sapia fits better where candidates need an untimed, camera-free format and feedback.

Greenhouse Voice AI

Best for: structured-hiring teams already on Greenhouse.

What it does: Voice AI, built on Ezra, which Greenhouse acquired in May 2026, runs a two-way voice interview in 11 languages. It’s scored against “questions and scoring criteria your team defines”, and recruiters can override the score. Every Greenhouse tier also includes self-scheduling and a candidate survey.

Evidence profile:

• Candidate produces: voice answers.
• Score built from: answers against your rubric.
• Published validity: none published.
• Independent audit: monthly Warden AI audits (per Greenhouse). Greenhouse also holds ISO 42001 certification for AI governance.
• Integrity controls: none described.

Strengths: you own the rubric and can override scores; audited every month; surveys built into the ATS.

Where it falls short:

• Voice-only screening is harder for some candidates, and no untimed text option is described.
• No published validity evidence.
• Reviewers report scheduling emails landing in junk folders.
• It’s only an option if Greenhouse is your ATS.

Pricing: quote-only; Voice AI sold separately.

Verdict: the natural pick for Greenhouse customers. Sapia fits other ATSs, CV-blind text screening and feedback for every candidate.

Eightfold AI Interviewer

Best for: enterprises wanting one AI interview covering screening, coding and case studies.

What it does: “360 Interview” (August 2026) combines screening, role fit, coding, a case study and a language check in one voice or video session of about 60 minutes. Identity is checked through CLEAR or ID.me.

Evidence profile:

• Candidate produces: voice or video answers, code and case work.
• Score built from: skills shown in the conversation, “not on résumé keywords” (vendor).
• Published validity: none published.
• Independent audit: BABL AI, June 2026, pooled. Race and ethnicity were tested on synthetic data (results).
• Integrity controls: identity verification and proctoring.

Strengths: the most recent published audit in this group; broad coverage in a single session; 93% candidate satisfaction (vendor-reported).

Where it falls short:

• A 60-minute session is a heavy ask at the top of the funnel.
• The audit is pooled, and race and ethnicity were tested on synthetic data.
• No published validity evidence.

Pricing: quote-only.

Verdict: strong for professional and technical roles. Sapia fits a shorter, untimed first step.

Also worth knowing:

• SmartRecruiters Winston (SAP-owned): Winston Match scores education, skills and experience, a score that blends in profile data. We found no published audit.
• Humanly and Sense (staffing): conversational screening, with no published audit.

Stage 2: Applications and engagement

Paradox (Workday)

Best for: high-volume frontline hiring, especially for Workday customers (Workday has owned Paradox since October 2025).

What it does: its candidate experience agent offers:

• applications by SMS or WhatsApp, with no login
• chat screening
• self-scheduling
• offers by text
• surveys at each stage
• 100+ languages

Evidence profile:

• Candidate produces: chat answers to qualifying questions.
• Score built from: recruiter-set knockout rules.
• Published validity: none published.
• Independent audit: none published.
• Integrity controls: none published.

Strengths: 72% average application completion (vendor-reported); fast self-scheduling; covers application through to offer.

Where it falls short:

• Reviewers say questions can “fall outside the automated path”.
• Implementation takes time.
• Knockout rules need your own adverse-impact checks.
• Its long-term roadmap for ATSs other than Workday isn’t public.

Pricing: quote-only.

Verdict: the benchmark for chat-based applications at volume.

Phenom

Best for: enterprises wanting one platform from careers site to interview.

What it does: careers site, CRM, chatbot and scheduling, plus AI agents for voice screening, interviews and fraud detection (March 2026).

Evidence profile:

• Candidate produces: a profile or CV, screening answers and assessments.
• Score built from: the Fit Score matches on “Skills, Experience, Job title, Location” – a profile match.
• Published validity: Phenom cites a 2026 audit but names no auditor, and we found no report.
• Independent audit: the same unnamed 2026 audit.
• Integrity controls: the fraud agent uses facial recognition and voice matching, with no published figures.

Strengths: the widest coverage of any tool here; built-in fraud detection.

Where it falls short:

• Reviewers say the voice bot “will skip and glitch because of small noises”.
• Steep learning curve.
• The audit claim can’t be verified.
• Facial recognition raises biometric-consent questions to put to the vendor.

Pricing: quote-only.

Verdict: the broadest suite. Audit its AI touchpoints carefully.

Stage 3: Scheduling and interview logistics

GoodTime

Best for: complex interviews with several panel members.

What it does: its AI agent Cori (May 2026) schedules and reschedules interviews, sends reminders, chases interviewer feedback, and messages candidates by email and SMS. It doesn’t score candidates.

Strengths: fewer no-shows; faster decisions.

Where it falls short: reviewers report available slots being missed; complex set-up.

Pricing: quote-based, priced on candidate volume (GoodTime).

Verdict: the fix for days of back-and-forth over panel interviews.

ModernLoop

Best for: scheduling that runs automatically when a candidate moves stage in the ATS.

What it does: “Zero Click Scheduling”, a candidate portal and Taylor AI phone screens. Scheduling doesn’t score candidates. Taylor AI does, but ModernLoop publishes nothing on its method, validity or audit.

Strengths: removes manual steps; one place for candidates to manage logistics.

Where it falls short: reviewers say rescheduling is difficult; Taylor AI’s evidence is unpublished.

Pricing: not published.

Verdict: a lean scheduling fix. Evaluate its AI screening separately.

Stage 4: Measurement (the audit’s own toolkit)

Starred

Best for: cNPS at every stage, including from rejected candidates.

What it does: automated surveys from careers page to onboarding, AI analysis of comments, and public benchmarks. It integrates with Greenhouse, Workday, SuccessFactors and others, and doesn’t score candidates.

Strengths: a 31% response rate from rejected candidates (vendor-reported); benchmarks to compare against.

Where it falls short: no write-back to the ATS; few public reviews.

Pricing: quote-only.

Verdict: the easiest way to run Step 4 all year round.

Survale

Best for: always-on surveys with CandE benchmarking.

What it does: candidate, onboarding and quality-of-hire surveys triggered from your ATS. It has owned the CandE Benchmark since April 2025, and doesn’t score candidates.

Strengths: access to CandE benchmarks; automated surveys at scale.

Where it falls short: reviewers report dashboard and permission issues; the benchmark is now run by a survey vendor.

Pricing: quote-only.

Verdict: the pick when CandE benchmarking matters.

Where Sapia fits (and where it doesn’t)

Our lane: screening and interviews. We offer an untimed, mobile, text-first structured chat interview with three features:

• it scores only the candidate’s answers, not their CV
• every candidate gets personalised feedback
• it detects AI-generated answers, with published (vendor-reported) figures

It’s backed by an independent, published audit (BLDS, late 2022, US and Canada models). When a candidate, regulator or your legal team asks “why?”, the answer is what the candidate said.

Where we don’t fit, and who to choose instead:

• Application friction → Paradox or Phenom.
• Panel scheduling → GoodTime or ModernLoop.
• cNPS and benchmarking → Starred or Survale.
• A voice interviewer inside your ATS → Greenhouse Voice AI or HireVue.
• Coding and case studies → Eightfold.
• The deepest published validity documentation → SHL.

Sapia isn’t an ATS, careers site or survey tool. It sits on top of the ATS you already run.

In practice (customer-reported): Woolworths reached 82.6% completion and 9.2/10 candidate experience across around 27,000 hires in under 10 weeks (volume hiring). Starbucks reclaimed around 1,900 screening hours a month.

What you should still ask us:

• newer audit results, broken down by role
• our AI-detection false-positive rate for non-native English writers
• validity evidence for roles like yours

If your audit points at screening, book a demo.

How to choose candidate experience automation – a 5-step process

1. Start from the audit, not the vendor list. Buy for your top three findings.
2. Match the tool to the stage. Scheduling, surveys, applications and screening are different purchases.
3. For anything that scores candidates, get the evidence in writing:
   • what the score is built from
   • what validity evidence is published
   • the latest independent audit: auditor, date, product, and whether it’s role by role or pooled
   • what integrity controls it has
4. “None published” is an answer.
5. Check disclosure, human review and accessibility against EU AI Act Article 50, Illinois HB 3773, Colorado SB 26-189, NYC Local Law 144 and the UK’s rules on automated decisions. Confirm with counsel.
6. Pilot on one role family and re-run the audit. Compare these against your baseline:
   • completion
   • time in stage
   • time to outcome
   • cNPS by outcome
   • pass-through by group

Conclusion

A candidate experience audit is a data-backed, stage-by-stage review of hiring from the candidate’s side, and in 2026 it has to include your AI and automation. Automation fixes slowness, effort and silence. It does new damage when it’s undisclosed, offers no route to a person, or leaves candidates with nothing.

By stage:

• Screening: Sapia
• Applications: Paradox or Phenom
• Scheduling: GoodTime or ModernLoop
• Measurement: Starred or Survale

Start with the audit, even if you plan to buy. It tells you what to fix first and gives you the baseline to prove the fix worked.

FAQs

About Author

Barb Hyman

CEO & Founder