You handed over five names, texted your old boss "they might call," and waited for the offer. In 2026, nobody called. A survey link went out, a voice bot dialed, an IP fingerprint got logged, and one of your five references never opened the email. That last part is what killed the offer, not anything anyone said.
The reference check is now software, and it grades your references before it reads them
Late-stage reference checks in 2026 are structured surveys or conversational voice bots with fraud-detection layers, and the scoring engine weighs response rate and pattern consistency before a recruiter ever sees a comment. Glider AI, HiPeople, SkillSurvey (now part of iCIMS), Crosschq, Searchlight, X0PA, GCheck, RefHub, and SoWhat all ship some version of this stack. The candidate-facing experience feels the same as five years ago: you paste in names and emails. The back end does not.
Two architectures dominate:
- Survey platforms (SkillSurvey, HiPeople) send a structured questionnaire, score competencies on a fixed scale, and layer in fraud detection. High volume, shallow depth.
- Conversation platforms (Glider AI voice, plus the newer entrants) call the reference or run a chat interface, capture verbatim responses, and analyze them for consistency. Used more often on senior hires.
Superunit's 2026 vendor roundup calls out that survey platforms hit their ceiling on senior hires because completion rates drop and answers go shallow. Voice fills that gap. Either way, the reference is talking to a machine.
Across 1,400+ recruiting teams including McDonald's, Chili's, and Memorial Sloan Kettering.
The 80% rule: a silent reference hurts more than a mediocre one
SkillSurvey's own validity research says a Reference Response Rate below 80%, combined with an average rating at or below the 35th percentile, is statistically predictive of first-year turnover for cause. That threshold is baked into the scoring engine. If you list five references and one ghosts, you are at 80% and borderline. If two ghost, you are at 60% and auto-flagged before a human reads a word.
This inverts the coaching most candidates were taught. The old rule was "pick people who will say great things." The new rule is "pick people who will actually click the link, on time, from a stable email address."
| Scenario | Submitted | Completed | Response rate | System reads it as |
|---|---|---|---|---|
| Textbook | 5 | 5 | 100% | Positive signal |
| One ghost | 5 | 4 | 80% | Borderline |
| Two ghosts | 5 | 3 | 60% | Turnover risk flag |
| Bare minimum | 3 | 2 | 66% | Turnover risk flag |
| Over-listed hedge | 6 | 5 | 83% | Positive signal, one absorbed |
The takeaway: over-list. When a recruiter asks for three, offer five. When they ask for five, offer six. Each extra name is a cheap insurance policy against a single unreachable inbox.
What the fraud layer is actually looking for
The fraud-detection layer inside HiPeople and Glider is not looking for lies; it is looking for patterns that suggest the candidate wrote the reference themselves or coached the language too closely. Get on the wrong side of these signals and the score tanks regardless of content.
Documented fraud signals across HiPeople and Glider:
- Duplicate contact details across your references (same recovery email, same phone area code cluster).
- Shared IP addresses between the candidate's application and the reference's survey submission.
- VPN usage on the reference side, especially if it masks an unexpected country.
- Phone number metadata: HiPeople verifies the age, provider, and country of the reference's phone. A brand-new burner or a Google Voice number reads badly.
- Unique fingerprints per submitter: HiPeople builds a device and behavior fingerprint for every reference, then flags similarities between the candidate's own fingerprint and any reference's.
- Scripted language: verbatim reuse of phrases across multiple references, or phrasing that mirrors the candidate's cover letter.
- Relationship mismatches: if the candidate said "direct manager, 2022 to 2024" and the reference selects something different in the survey, that gap gets logged.
The uncomfortable implication: the classic favor of sending your reference a bullet list of talking points is now a fraud signal, not a favor. HiPeople and Glider both explicitly analyze for scripted patterns. Coaching the substance is out. Coaching the logistics is in.
The employment-verification trap hidden inside the reference survey
HiPeople embeds employment-verification questions inside the reference survey itself, meaning the reference is silently verifying your resume line by line. A representative prompt from HiPeople's own product literature: "Did Melody work at Apple from Oct 2022 to Aug 2026 as a Software Engineer L2?"
If your resume says "Senior Software Engineer" and your former manager clicks "Software Engineer" because that is what your title was on the org chart, that is now a resume-mismatch flag, not just an inaccurate answer. Same for month-off date ranges, promoted-mid-tenure titles, and team names that got renamed after a reorg.
This is where resume hygiene meets reference hygiene. If your resume overstates a title or fudges a date range, the reference survey will catch it, and the catch will look like fraud rather than a rounding error. Fixing this starts with a resume that actually matches what your references will confirm under a fingerprinted survey. Rewriting your history yourself for every posting is grinding work, and it is the exact work Refolk takes off you: it drafts the resume from your real history, tailors it to the posting, and keeps the dates and titles consistent enough to survive the verification step.
The prep sheet: what to send your references before the link arrives
Send a short, factual, opinion-free note to every reference before the survey invite lands, because voice-AI and short-expiry survey links leave no time for a "hey, what should I say?" text after the fact. The point is logistics and accuracy, not talking points.
A one-message template that respects the fraud layer:
- Heads up on timing: "Between [date] and [date] you'll get an email or a call from [vendor name if known]. It expires in 48 to 72 hours."
- Confirm the channel: "Reply here with the best email and phone for that." Correct contact info is the single biggest driver of response rate.
- Factual overlap only: "For consistency, here is what my resume says about our time working together: [company], [team], [my title], [dates], reporting to [name]." No adjectives. No accomplishments. Just the facts they will be asked to confirm.
- Format warning: "It might be a 10-minute survey or a short voice call with a bot. Both are normal."
- No talking points. Do not send bullet points about what to emphasize. Do not attach the job description. Do not paraphrase your cover letter.
The Greenhouse 2026 Candidate AI Interview Report found 38% of US candidates had withdrawn from a hiring process specifically because it involved an AI-led interview (n=2,950 across US/UK/IE/DE/AU). That same distrust hits your references when a bot calls them cold. Warn them the bot is coming, or they will hang up on it.
The old playbook was coaching what your reference would say. The new one is coaching whether they will pick up the phone.
Who actually reads the score, and why healthcare candidates get hit hardest
The people running these tools are recruiters and talent-acquisition teams, and in Refolk's index of professional profiles, roughly 92,743 US professionals currently hold "Recruiter" or "Talent Acquisition" in their title. The single largest employer concentration in that slice is RCM Healthcare Services, which lines up with the vendor pattern: healthcare is SkillSurvey and HiPeople's biggest volume vertical.
If you are a nurse, clinician, or allied-health candidate, you are disproportionately exposed to these platforms. The upside is that healthcare TA teams process high volume and lean on the score more, which means the mechanical prep (response rate, contact accuracy, resume consistency) pays off more than it does in a bespoke executive search.
The comparison table: what each major vendor grades you on
The scoring criteria are not uniform across vendors, and knowing which platform your target employer uses changes what you optimize for. Ask the recruiter. Most will tell you.
| Vendor | Format | Fraud signals | Notable feature |
|---|---|---|---|
| SkillSurvey (iCIMS) | Structured survey | Response rate scoring, percentile flags | Patented Pre-Hire 360; ~93% of finalized-candidate ratings cluster at the high end |
| HiPeople | Survey + verification | IP, phone metadata, VPN, unique fingerprints | Employment verification embedded in the reference questions |
| Glider AI | Voice or chat | Scripted-language analysis, IP cross-reference, verbatim capture | Conversational voice bot for senior hires |
| Crosschq | Survey | Response-quality scoring | Positioned for executive references |
| Searchlight | Post-hire summarization | Less fraud-focused | AI summarizes reference themes |
The 93% clustering number matters: because candidates pick their own references, almost every rating comes in high. A middling score is not "meh" in this system; it is a red flag by contrast. Optimizing for a great score is not the goal. Not tripping the flag is.
What to do the week before you submit references
The compressed timeline (voice AI dials within hours, survey links expire in 48 to 72 hours) means the prep window is now the week before you name the references, not after. HiPeople customers report time-to-fill down 31% and 62% more hires in a year, which is a nice way of saying the process moves faster than your uncle checks his spam folder.
A tight pre-submission checklist:
- Confirm current email and mobile for every reference in the last 7 days.
- Send the factual-overlap note (dates, title, team) so the survey verification questions align with your resume.
- Tell every reference the invite is coming, the vendor name if you know it, and that a voice bot is possible.
- Over-list by at least one. Six names when they ask for five. Four when they ask for three.
- Cross-check your resume dates and titles against what your references would actually confirm. If any line is a stretch, either fix the line or drop the reference who would flag it.
FAQ
How do I know if an employer uses SkillSurvey, HiPeople, or Glider?
Ask the recruiter directly at the offer stage: "What platform do you use for reference checks?" They will tell you, because they need you to submit contact info anyway. If they will not say, watch the sender domain on the eventual invite email (skillsurvey.com, hipeople.io, glider.ai). Knowing the vendor tells you whether to expect a survey link, a voice call, or both, and lets you warn your references about the exact format.
Is it okay to tell my reference what job I am applying for?
Yes on the company and role. No on the job description, the cover letter, or bullet points of what to emphasize. The fraud layers in HiPeople and Glider look for scripted-language patterns, so any language reused across multiple references (or that mirrors your own application) hurts you. Give context; do not give a script.
What happens if a reference cannot answer the voice-AI call?
Most voice-AI reference systems fall back to a survey link or a rescheduled call, but the clock is short (often 48 to 72 hours). If a reference misses the window, they count as non-responsive, and your Reference Response Rate drops. Below 80%, SkillSurvey's own validity model flags you as a turnover risk. This is why over-listing and confirming availability in advance matters more than picking the single most flattering reference.
Should I use my current manager as a reference?
Only if they already know you are looking. Reference checks are triggered at the offer stage, and voice-AI systems dial faster than you can control the narrative. If your current manager is a surprise recipient, the flag on the survey (about your current employment) may look worse than a strong past manager filling the slot. Default to past managers unless you have explicitly briefed your current one and confirmed they will complete the survey on time.