After an Offer Is Withdrawn, How Much Employer Trust Remains? Event-Based Employer Brand Research

Between late August and early September 2026, the "mass rescission of campus recruits" incident at a Chinese auto-lighting company, Xingyu Co., kept escalating: the Changzhou municipal human-resources authority reported that the company had recruited 440 graduates of the class of 2026 and rescinded labor contracts with 107 of them, describing the process as crude and poorly communicated; on September 7 the company announced disciplinary measures and follow-up remedies including subsidies and job fairs. After any "offer rescinded" incident like this, the HR team wants an "employer brand damage" survey. But the common approach—asking only the affected graduates how they feel now—misses the actual mechanism: employer-brand trust is a signal aimed at the entire labor market, and the loss mainly comes not from the anger of those affected but from uninvolved people who see the signal and change their application and recommendation behavior.

The design principle for event-based employer brand research is therefore: define three audiences at once—affected graduates, unaffected candidates, and current employees—and measure them separately; operationalize "fairness" into procedural, distributive, interpersonal, and informational components; and split "trust" into observable behaviors (whether people withdrew applications or still attend job fairs) and word-of-mouth intentions with a named object (not a vague "would you recommend" but "would you encourage a friend to apply to this company"). The survey measures the signal; legal compensation and administrative accountability are questions outside its scope.

Define Three Audiences First, Not Just the Victims

The same incident means different things to three groups, and each needs its own recruitment and sampling design from the start:

The denominators differ completely across the three groups. Affected graduates are parties to the incident, so conclusions apply only to those 107 people's perceptions and subsequent resettlement. Candidates and employees answer how the signal changed behavior—that is the quantifiable part of employer-brand loss. A report with only the victims' emotions and no behavior change among candidates and employees cannot answer management's real question: "Can we still hire next year?"

Separate Time Windows: Then, Now, and Future Behavior

A survey in the immediate aftermath captures only momentary emotion and cannot predict behavior three months later. A competent questionnaire separates at least three time layers: perception at the time of the event (recall items, with recall bias flagged), current trust and behavior (whether people are still engaging with other roles), and future intention (whether they would apply next recruiting season or recommend the company to juniors). To track the recovery curve, run waves at months 1, 3, and 6 after the event with the same people as a panel or at least the same measurement instrument, rather than a single "damage" cross-section.

The time window also dictates tense: keep "after this happened, did you…" and "in the next year, would you…" as two separate groups of items. Do not place a timeless attitude item like "do you still trust this company" in the middle; it will fuse momentary emotion with long-term judgment into one number.

The Four Fairness Components: Turning Vague "Unfairness" into Diagnosable Items

Organizational-justice research—whose meta-analysis distinguished and validated procedural, distributive, interpersonal, and informational fairness (Colquitt, 2001)—provides a ready diagnostic skeleton. Each component maps to item sets that point to different corrective actions:

• Procedural fairness: whether the rules of rescission were explained in advance and whether there was an appeal or consultation channel. Example item: "Before deciding to rescind, did the company give you an opportunity to express your views and negotiate?"
• Distributive fairness: whether compensation and remedies matched the loss. Example item: "Did the compensation and job-search support provided match the loss caused by losing this job?"
• Interpersonal fairness: whether the communication treated people with respect. Example item: "Was the person who communicated with you respectful, without humiliation or dismissal?"
• Informational fairness: whether reasons were explained clearly, promptly, and candidly. Example item: "Did the company promptly and candidly explain the reasons for the rescission and the follow-up arrangements?"

The value of the four components is diagnosis: if low scores concentrate on informational fairness, the fix is transparent communication mechanisms; if on procedural fairness, rules and appeal channels; if on distributive fairness, compensation design. Compressing the four into a single "fairness index" is convenient for reporting but erases the differentiation a remedy plan needs.

Illustrative Worked Example: How a Three-Audience Recommendation Gap Should Be Reported

The numbers below are fully illustrative and are used to show the report structure; they do not represent Xingyu or any real study. Assume that six weeks after the incident a company surveyed three groups: 60 affected graduates, 400 unaffected candidates, and 800 current employees:

Reporting only the first column yields "the company is socially dead"; reporting only the third underestimates risk because employees' 63% rests on internal experience, while aware candidates (about 268 people) show only 36% recommendation willingness—usually more than 20 points below the unaware group. For employer brand, candidate recommendation is the core external signal because next year's résumés come from them. The report should give each group's proportions and denominators separately, then compute the aware-versus-unaware candidate behavior gap, instead of averaging three populations into one "employer brand score."

Translate Trust into Behavior and Name the Object of Word-of-Mouth

Attitude items are easily contaminated by social desirability and momentary emotion, especially in event studies. Make two replacements when designing:

1. Replace "do you still trust this company" with a behavior checklist: "After the incident, which of the following have you done—withdrew or edited an application, cancelled an interview, stopped following the company's recruiting, continued normal engagement, or expressed negative comments to others"—multi-select plus frequency. Behavior is closer to real decisions than declarations.
2. Name the object and situation of "would you recommend": "Would you encourage a same-major friend to apply to this company" and "Would you post a positive review of this experience on job forums." A scale such as NPS is interpretable only when the object is explicit; a generic "recommendation willingness" mixes "satisfied with the product" with "trust in the employer." For NPS's three-group structure and error interpretation, see NPS survey design.

For candidates, recommendation measurement must also separate "aware of the incident but still applying" from "applying because unaware": the latter means the funnel is not damaged—the information simply has not spread; the former is the result of trust repair. Awareness, consideration, and application are distinct funnel levels and cannot be replaced by a single "liking" score; see brand awareness research for the layered method.

Current Employees Are Not Bystanders; They Amplify the Signal

How employees see "how the company treats young people" reaches the next candidate cohort directly through referrals and social networks. When surveying employees, do not ask only "does the company treat me well"; ask whether they would still refer juniors after this incident, whether they still recommend the company externally, and whether they worry similar practices could affect them later. The last item measures "self-referential fairness"—employees are judging not what the company did wrong but "could the same thing happen to me." Only when the two kinds of items are separated can management tell whether what needs repair is the external employer brand or the internal floor of trust.

Boundaries: What the Questionnaire Does Not Answer

Several questions are beyond what event surveys can answer: whether the conduct was illegal, what compensation is owed, and the outcomes of administrative and labor-supervision review belong to labor disputes and regulatory authority and should follow the labor authorities, the courts, and the company's law-based disclosures; a limited sample cannot infer compliance conclusions. Factual verification of individual cases—who promised what and what happened—requires written records rather than recall items. Data collected while an incident is still developing has a short shelf life; reports must state the collection date and the stage of the incident, rather than writing "the emotion of week 6" as "the permanent state of the employer brand."

Running This Study in WeJot

WeJot can run three separate questionnaires by audience: targeted full reach for affected graduates with a low-response bias check; candidates sampled from the application database and stratified by aware/unaware; and an anonymous short survey for employees. The three questionnaires share the four fairness components and the behavior checklist for comparability; rolling waves can be fielded at months 1, 3, and 6. For writing the methodology note, see how to write the survey methodology section.

The companion recruitment promise and employer trust survey template contains modules for the three audiences with representative items and can be generated and adapted to the event. Incidents pass, but candidates remember signals—the job of measurement is to make the signal visible, classified, and repaired rather than averaged away.