---
title: Greenhouse Candidate Surveys: Setup and AI Analysis | Remqa
url: https://remqa.com/blog/greenhouse-candidate-surveys-ai-analysis
updated: 2026-11-19
---

[Home](https://remqa.com/) / [Blog](https://remqa.com/blog) / Greenhouse Candidate Surveys: Setup and AI Analysis

Recruiting operations

# Greenhouse candidate surveys: setup, reporting and careful AI analysis

Configure Greenhouse candidate surveys, understand exclusions and report dates, then use reviewed AI summaries without trying to identify respondents.

Remqa · November 19, 2026 · 5 min read

**Short answer:** Set the Greenhouse candidate survey audience by office, department and interview stage, then choose when invitations follow a hire or rejection. Read the reporting date rules before comparing results, and use any custom AI analysis to summarize process feedback without matching anonymous responses to candidates.

- Selecting both an office and a department requires candidates to match both
- Recent survey recipients, active applications elsewhere and Do not email affect eligibility
- The Submitted between filter changes response results but not the completion total
- Keep AI summaries separate from candidate profiles and review identifying details

By the Remqa engineering team · Last updated November 19, 2026

A recruiting team opens its candidate survey report and sees complaints about unclear interviews. Someone asks for an AI summary by department. Before the export leaves Greenhouse, the team needs to answer two questions: which candidates could receive the survey, and which dates the numbers actually describe. Without that context, a polished summary can turn a narrow set of responses into a claim about the entire hiring process.

Greenhouse provides native survey configuration and reporting. The AI analysis described here is a recommended custom workflow around those features. Keep the work focused on process improvements, with a recruiter reviewing the evidence before a summary reaches hiring managers.

## Choose the audience before changing questions

A Site Admin can open Configure, select Candidate survey and choose Edit survey settings. Audience controls include office, department and stage. The office and department refer to the job opening. If you select both, a candidate must match both selections. Treat this as an intersection, not a combined list of everyone in either group.

Stage selection has another constraint. A stage must appear in every job interview plan within the selected office and department to appear as an available option. If the stage you expect is missing, inspect the selected jobs and their plans rather than assuming the survey configuration is broken.

Greenhouse lets you choose when invitations are sent relative to a candidate being hired or rejected. Record the selected timing with the audience configuration. A change to that timing can change when feedback arrives, so keep it visible when comparing reporting periods.

## Explain missing invitations with eligibility rules

An invitation count is not simply the number of rejected candidates. Greenhouse excludes candidates who received a survey from the organization within the past six months, remain active for another job or have Do not email selected on their profile. The survey overview also excludes applications migrated from another ATS. A person with a migrated profile can still receive a survey for a later application submitted directly through Greenhouse.

| Check | What it means for analysis | Recommended operator action |
| --- | --- | --- |
| Office and department | The selected audience may cover only part of hiring | Attach the audience definition to every summary |
| Selected interview stages | Earlier exits may fall outside the audience | Describe which part of the process is represented |
| Another active application | A rejection on one job may not trigger an invitation | Explain eligibility without overriding it |
| Migrated application | Historical ATS imports do not behave like direct applications | Separate migration history from current feedback |

Do not build a second email sender to compensate for every missing invitation. First distinguish an expected exclusion from a configuration issue. A custom sender can conflict with communication preferences and create a second survey process that nobody intended to operate.

## Keep the question set consistent

Greenhouse supports customizable questions and one survey across departments and offices. Start by deciding which process decisions the answers should inform. Interview preparation, communication clarity and coordination are different topics. A useful question asks about one of them rather than combining several into a single rating.

For your own analysis, keep a dated copy of the question text whenever it changes. A renamed question can alter what respondents think they are rating. Do not silently combine old and new wording into one trend. Ask the reviewer to decide whether the responses remain comparable and annotate the summary when they do not.

## Read the report date rules literally

Open Reports, Essential reports and Candidate Surveys in the Share Progress with Executives section. Greenhouse documents the report for Job Admin and above. Apply office, department and Submitted between filters deliberately. The date filter defaults to the current year.

The important distinction is between response results and the completion total. Greenhouse says it does not collect survey date sent data. The report's total number and percentage completed represent all submissions within the selected offices and departments. Changing the date range does not change that total. The response results do reflect the selected submission period.

> **Do not label the completion total as this month's response rate.** A submission date filter does not turn an all-submission completion total into a matched invitation cohort. Keep these measures separate in exports, dashboards and AI prompts.

If leadership wants a different metric, define its numerator, denominator and observation period before building it. Do not ask a model to reconstruct missing invitation dates from comments or candidate activity. An unavailable measurement should remain unavailable, with a clear explanation.

## Build a restricted analysis packet

Greenhouse documents an Excel export for viewing survey comments and notes. That file creates a new copy outside the ATS. Decide who can access it, where it can be processed and when it should be removed before passing it to another system. These are custom operating controls, not guarantees inherited from Greenhouse.

Prepare a packet with the question wording, selected filters, response period and permitted comments. Ask a reviewer to remove names, contact details and unusually specific descriptions before AI processing. Anonymous survey collection does not make every free-text comment safe to circulate without review.

Do not join comments to candidate profiles, scheduling history or rejection timestamps to guess who wrote them. The objective is to improve the recruiting process. A summary should describe an issue and the supporting feedback, not produce a list of suspected respondents.

## Review themes before creating work

1. Ask the model to group comments by a small, written set of process themes.
2. Require each proposed theme to point back to the permitted input passages.
3. Separate direct feedback from the model's interpretation and suggested remedy.
4. Have a recruiter check missing context, identifying details and overgeneralizations.
5. Assign an approved improvement to an owner without attaching candidate identities.

A comment about interview duplication may justify reviewing the interview plan. It does not establish that every interviewer was unprepared. Keep the scale of the conclusion aligned with the input. Preserve disagreement and uncertainty rather than compressing every comment into a confident department score.

Start with a read-only summary. Add task creation only after the team has checked its usefulness and chosen an approval step. Our guide to [AI recruiting agents](https://remqa.com/ai-recruiting-agents) covers how to separate evidence gathering from actions that change recruiting records.

## Common mistakes to avoid

- Calling survey respondents representative of every applicant without checking eligibility.
- Comparing completion totals as if the submission filter changed their denominator.
- Mixing different question wording without marking the change.
- Sending raw comments to a wider audience than the original analysis requires.
- Writing anonymous feedback back onto an individual candidate profile.

Measure the review effort as well as the draft time. If recruiters must repeatedly correct themes or remove personal details, include that work when judging the workflow. The [recruiting AI measurement guide](https://remqa.com/blog/measure-recruiting-ai-time-saved) explains how to include review and rework rather than counting generated summaries alone.

## How Remqa can help with candidate survey analysis

Remqa builds AI agents for recruiting teams. A candidate feedback project can start with a restricted export, a clear reporting definition and a read-only summary for recruiter review. We can help design the evidence packet, access checks and approved handoff into your operating process. Bring the current report and the decision you want it to support to a [recruiting workflow review](https://remqa.com/recruiting-hours). Start with one useful, reviewable output.

## Questions people ask.

**Q: Who can configure the Greenhouse candidate survey audience?**
Greenhouse documents Site Admin permission for choosing survey recipients. Audience settings are under Configure, Candidate survey, then Edit survey settings.

**Q: Why did a rejected candidate not receive a survey?**
Check the selected office, department and stages first. Greenhouse excludes recent survey recipients, candidates still active on another job and profiles marked Do not email. Its overview also excludes applications migrated from another ATS, while allowing later applications submitted directly through Greenhouse.

**Q: Does the report date filter change the survey completion total?**
No. Greenhouse says the total number and percentage completed represent all submissions within the selected offices and departments. Submitted between changes the response results shown, not that completion total.

**Q: Where can we read candidate survey comments?**
Greenhouse documents exporting the candidate surveys report to Excel to view comments and notes. Treat that export as a separate sensitive dataset with an approved audience and retention policy.

**Q: Can an AI agent identify which candidate submitted a survey?**
Do not design it to do that. Use anonymous process feedback to find themes, and avoid joining comments to candidate profiles, interview dates or application histories. AI analysis here is a custom integration recommendation, not a native Greenhouse feature claim.

Book a free 30-minute call with an engineer: https://calendly.com/meghanaj-connect/30min
