How to Evaluate AI Interviewing ATS Integrations: An 8-Question Vendor Checklist

Every AI interviewing vendor says "yes" to "Does it integrate with our ATS?" That "yes" spans a Zapier connection pasting notes into a text field to a certified native integration writing structured scores, triggering dispositions, and logging every decision for audit.

Ask at the layer level or you won't know what you're buying. These eight questions are what TA ops and HRIS teams should bring to any vendor demo. For the full integration architecture behind them, see our AI interviewing ATS integration guide.

TL;DR: Require layer-level answers at your next vendor demo, not marketing summaries. A vendor who walks through all eight concretely is running a production-grade integration. One who keeps returning to "yes, we integrate" is selling you a label, not an architecture.

The 8-Question Checklist: Ask These at Every Vendor Demo

Question 1: Is your [ATS] integration native/certified, API, or Zapier, and what's the maintenance model?

Who updates it when the ATS changes its API?

Three different things get called "integration": native/certified (vendor-built for your ATS), API (your IT team owns it), and Zapier/webhook (no real-time data, limited field mapping, no audit trail).

A Zapier integration is not an enterprise integration. If a vendor describes their Workday or Oracle connection as "via Zapier," that's a significant limitation for EEOC and OFCCP defensibility.

The maintenance model matters equally. Native: vendor updates when the ATS API changes. API: your IT team owns that work.

• What good looks like: The vendor names the architecture type immediately, explains who owns updates, and can point to certified partner status with your ATS.
• What weak looks like: Vague answers about "connecting via API," inability to confirm who updates the integration when the ATS changes, or describing Workday/Oracle connectivity as running "through Zapier."

Question 2: What data do you pull from the ATS at the trigger layer? Can you pull custom job fields, or only standard ones?

At minimum, the AI tool pulls candidate name, contact details, and job title. Better integrations pull custom qualification criteria from the requisition, the fields that define what actually disqualifies or advances a candidate

Job title only = generic interview. Custom qualification fields = the tool probes criteria your hiring managers actually care about. (See how this works in practice in our AI interviewing overview.)

• What good looks like: The vendor confirms it can read custom requisition fields and demonstrate this in your ATS schema.
• What weak looks like: "We pull the standard candidate and job data" with no mention of custom fields.

Question 3: What writes back to the candidate record after the interview? Can you write structured competency scores to custom fields, not just a note?

After the interview, the candidate record should receive: completion status, structured competency scores with question-level detail, a transcript or recording link, and a recommended disposition. The distinction that matters: a score number in a field vs. structured competency scores recruiters can sort, filter, and defend

Unreliable write-back kills adoption. Recruiters can't see AI output where they work, start reconciling two systems by hand, and revert to manual screening.

• What good looks like: Structured competency scores writing to custom ATS fields; transcript link in the activity timeline; disposition recommendation in a designated field.
• What weak looks like: Results delivered as a pasted text note, a single summary score, or visible only inside the vendor's own dashboard.

Question 4: What is the trigger: automatic on status change, or a manual recruiter action? Show me how a candidate enters the interview.

Tcides when and how consistently an AI interview fires: req sync, status change, manual recruiter action, or API call. Automatic triggers produce consistent coverage. Manual triggers produce variance, and every downstream metric becomes unreliable. (For more on how scheduling automation fits into the trigger layer, see our scheduling solution.)

Make the vendor walk you through the exact sequence: what fires the interview in the ATS, and what happens when a recruiter skips that step.

• What good looks like: Automatic firing on ATS status change or req sync, with no recruiter action required for consistent coverage.
• What weak looks like: "The recruiter clicks a button to send the interview invite" or any trigger that depends on individual recruiter behavior.

Question 5: What is your audit log output, and does it satisfy OFCCP and NYC Local Law 144 documentation requirements? Walk me through a sample export.

Five elements make an ATS record defensible: completion status and timestamp, structured competency scores with question-level detail, disposition action and criteria, candidate disclosures, and bias audit classification

Ask to see a sample audit export. Don't let them describe it. What you see should answer whether it holds up in an EEOC or OFCCP review.

• What good looks like: A structured, exportable log with disposition reason, timestamp, candidate disclosures, and bias classification - all traceable to the candidate record in your ATS.
• What weak looks like: "We log everything in our system" with no clear export path into the ATS, or audit output that lives only in the vendor dashboard.

Question 6: How are integration issues supported: a dedicated integration team, or general support? Who do we call when a write-back fails?

Write-backs fail. Fields stop mapping. API changes break connections. What matters: how fast issues get caught, and who owns resolution. A dedicated integration team with proactive monitoring is not the same as a general support queue waiting for your ticket.

• What good looks like: A named integration support path, proactive monitoring for write-back failures, and SLA commitments for integration issues.
• What weak looks like: "You can submit a ticket through our support portal" with no integration-specific escalation process.

Question 7: What is the implementation timeline for [ATS] specifically, from kickoff to live?

Four weeks at 100 hires/week becomes four months at 1,000/week with a complex ATS. Get the specific number for your ATS and volume, plus what it requires from your side: IT hours, ATS admin access, field mapping decisions

• What good looks like: A specific timeline for your ATS with a clear scope of what your team needs to provide.
• What weak looks like: A generic "typically 4-6 weeks" that doesn't account for your ATS or volume.

Question 8: Has your integration been certified or audited by the [ATS] vendor? Can you show the certification?

For Workday: certified partner status. For Oracle: UNIConnect certification. These mean the ATS vendor validated the integration, not just that the vendor claims it works. Ask to see the documentation. A vendor with real certification has it on hand

• What good looks like: Certified partner documentation from the ATS vendor, available on request.
• What weak looks like: "We're integrated with Workday" with no certification to show.

Executive takeaway: A vendor who can answer all eight at the layer level is running a production-grade integration. One who keeps returning to "yes, we integrate" is selling you the marketing label, not the architecture.

ATS-by-ATS Quick Reference

• Workday: Native / certified. All 4 layers. Humanly partners with Workday. Key test: confirm certified partner status, not just "connects."
• SAP SuccessFactors: Certified partner / API. Layers 1-3 strong; Layer 4 varies. Key test: test audit output specifically.
• Oracle (HCM / Taleo): Native / certified. All 4 layers with UNIConnect-certified tools. Humanly maintains a native Oracle integration. Key test: verify UNIConnect certification.
• iCIMS: Mostly API. Layers 1-3; trigger logic uneven. Key test: test trigger reliability.
• Greenhouse: API. Layers 1-3 strong. Key test: less common in high-volume hourly.
• Bullhorn: API (limited native). Layers 1-3 partial; Layer 4 weak. Key test: compliance audit output weakest.

The Bottom Line

A vendor who answers all eight at the layer level is running production-grade integration. One who deflects to "yes, we integrate" is selling the label.

Humanly's native integrations with Workday, SAP and Oracle cover all four layers, so structured interview data flows straight into your ATS and any connected AI-powered people analytics platform. Audit-ready write-back is the default. For a full architecture walkthrough, see our AI interviewing ATS integration guide.

Want to run these eight questions against your own ATS schema? Book a technical demo and we'll answer every one at the layer level.