If you were cut in the last month, the model that flagged you almost certainly weighed six inputs. A July 2026 ResumeTemplates.com survey of 1,000 U.S. managers at 500-plus-employee companies handed the industry a leaked feature list, and the press missed the point by leading with the 59% headline. The real story is the weights, and those weights tell you exactly which resume bullets to write next.
What the July 2026 survey actually leaked
The survey is a de facto spec sheet for the layoff model: performance scores at 80%, attendance at 57%, salary at 42%, tenure at 32%, sick days at 31%, PTO at 23%, and age at 14%. Every other outlet buried these weights under the softer "59% use AI" lede. Treat them instead as the input vector your next employer will feed a similar tool, and write your resume backwards from there.
The base rates matter. Of the 1,000 managers surveyed:
- 59% use AI to help decide who gets laid off
- 24% use it often or all the time
- 43% have let AI make a layoff call with no human review
- 38% received no training on ethical AI use in HR
- 58% cannot confirm the tool was tested for bias
- 63% believe AI can make fair, unbiased layoff calls
- 91% say they would override an AI recommendation they disagreed with
That last pair is the tell. Managers say they would override, but 43% already do not. The "human in the loop" is a stated preference, not a practice.
From ResumeTemplates.com's July 2026 Pollfish survey of 1,000 U.S. managers at 500-plus-employee firms.
The 31% sick-day signal is the real story
The single most important finding is that 31% of AI-using managers, or 18% of all managers surveyed, feed the model frequent sick days and medical leave, a category of data that is legally protected under the ADA and FMLA. Julia Toothacre of ResumeTemplates.com put it directly: "Sick days, medical leave, and age stand apart from the factors a layoff usually turns on, because discrimination based on age, disability, or protected medical leave is illegal."
Combine that 31% with the 58% of employers who cannot confirm bias testing and you have a live ADA lawsuit waiting to be filed. The Workday class action is already the first-of-its-kind case testing whether AI-based hiring programs can be discriminatory, and California's Fair Employment and Housing Act regulations extending disparate-impact liability to automated-decision systems went into effect October 1, 2025. NYC Local Law 144 still mandates an annual independent bias audit, with fines up to $500 for a first violation and up to $1,500 per subsequent one.
The practical rule for your resume: never disclose the protected data, and never volunteer a gap explanation that names a medical cause. Counter the signal with output metrics that make attendance irrelevant. If you shipped, the model has to weigh the shipping.
The feature vector, translated into resume bullets
Every row below is a bullet-writing prompt. Work them in order of weight. The "% of all managers" column is derived by multiplying the AI-user share by the 59% base rate, except sick days, which the survey reports directly.
| Feature the AI weighs | % of AI-using managers | % of all managers | Resume-bullet counter |
|---|---|---|---|
| Performance / productivity | 80% | ~47% | Lead every bullet with a metric and a denominator |
| Attendance | 57% | ~34% | Name uptime, on-call rotations, deadline hit-rate |
| Salary / cost | 42% | ~25% | Bullet revenue-per-headcount and cost saved |
| Tenure | 32% | ~19% | Date recent skills, not longevity |
| Sick days / medical leave | 31% | 18% | Do not disclose. Counter with output metrics |
| PTO usage | 23% | ~14% | Ship dates that overlap PTO windows (async wins) |
| Age | 14% | ~8% | Strip graduation years, foreground 2024-2026 stack |
Performance (80%): metric, denominator, timeframe
80% of AI-using managers feed the model performance and productivity scores. That is the heaviest weighted feature by a wide margin, and it is also the one your resume can attack directly.
A weak bullet: "Led migration to new billing system."
A bullet the model can score: "Led billing migration for 1.4M active accounts, cut invoice errors from 3.1% to 0.4% in Q2 2026, saved 11 finance-team hours weekly."
Three components in every line:
- A number with a denominator (1.4M of what)
- A before-and-after delta, not just an end state
- A dated window so the tool can weigh recency
This is the exact rewrite work Refolk does for you: paste the posting, get your own resume back with each bullet reshaped around the metrics that specific role actually screens for. It pulls the numbers from your own history rather than inventing them, which matters when the receiving model is looking for consistency across your LinkedIn, GitHub, and application text.
Attendance (57%): the reliability bullet nobody writes
Attendance outweighs salary in the layoff model, 57% to 42%. Job seekers assume they were cut for being expensive; statistically they were more likely cut for showing up irregularly, or for producing at an irregular cadence the model read as absence.
The counter is a reliability bullet, and almost no one writes one. Examples:
- "Held primary on-call for a 24/7 payments service, 6-week rotation, 99.98% response SLA over 14 months."
- "Hit 47 of 48 sprint commitments across 2025, zero missed release trains."
- "Shipped weekly product updates for 68 consecutive weeks (Jan 2025 to Apr 2026)."
Reliability language reads as human dependability to a recruiter and as a strong attendance-proxy signal to the model. It also displaces the need to explain any gap.
Salary (42%): frame yourself as a margin, not a cost
You cannot hide your comp band, but you can attach a return to it. Bullets that pair your cost with your output flip the signal:
- "Owned a $2.1M infra budget, delivered 34% cost reduction (Q3 2025 to Q2 2026) while doubling request volume."
- "Generated $4.8M in expansion revenue against a fully loaded cost of ~$220K."
The model is comparing your salary against a peer set. Give it a ratio in your favor.
Tenure (32%): the trap for eight-year veterans
Tenure at 32% cuts both ways, and this is the non-obvious insight most layoff-survivor advice gets wrong. In prior cycles, "last in, first out" was the rule. The current generation of models is equally likely to flag long-tenured people as stale headcount, especially when the "could AI do this job" filter (34% of AI-using managers) runs on top.
If you have 8-plus years at one employer, do not brag about longevity in the summary. Instead:
- Date every skill and tool to the year you last used it in production
- Split one long role into two or three project-scoped sub-roles with distinct outcomes
- Foreground work from 2024 through 2026, even if it was a smaller share of your time
If you have short stints, invert the same logic: name each stint as a scoped project with a delivered outcome, so the model reads intent rather than churn.
Sick days (31%) and age (14%): the do-not-disclose column
Two rules, no exceptions:
- Do not put graduation years on the resume. Age is a 14% feature and graduation year is its cleanest proxy.
- Do not narrate medical leave, caregiving, or health-related gaps in the resume itself. If a gap needs a label, use "independent projects" or "consulting" with one concrete deliverable.
The legal ground under sick-day and age filtering is shaky enough that the receiving employer's counsel does not want that data on your document. You are doing them a favor by not providing it.
Managers say they would override the AI. 43% already do not. Write for the model, not the manager.
Who actually builds these models
The people who design layoff-scoring systems are a tiny guild, and the gap between builder and user explains why 58% of employers cannot confirm bias testing. In Refolk's index of professional profiles, only 80 U.S. profiles carry a People Analytics, Workforce Analytics, or HR Analytics title, and just 21 hold the pure "People Analytics" title.
The concentration is stark:
- Apple: 7
- Meta, Google, OpenAI, Anthropic, Deloitte: 2 each
- Stripe: 1
- Geography: 5 in NYC, 3 in the SF Bay Area
Roughly 80 people nationwide are building the frameworks that thousands of Fortune 500 managers push buttons on. That is why "the model considered your attendance" often means a vendor default weighting nobody at your employer can explain, let alone defend in a deposition. Write your resume assuming the reader has less insight into the tool than you do.
The "AI could do this job" filter and the 13% cohort
34% of AI-using managers ask the tool to evaluate whether the job could be done by AI, which works out to 20% of all managers. Of the 44% of managers asked to make that evaluation directly, 75% concluded AI could replace the role and 30% went through with it. That is 13% of all managers who have actually replaced a person's function with AI.
If you are in that cohort, the bullet fix is not "I used AI." Every candidate writes that now. The fix is to bullet the supervision of AI at scale:
- "Audited 1,200 GPT-drafted support replies weekly, reduced hallucination rate from 4.2% to 0.6% over Q1 2026."
- "Designed eval harness for internal LLM pipeline, caught 3 regression incidents pre-deploy across 8 months."
- "Trained and deployed a fine-tuned classifier handling 40K daily tickets at 94% precision, replaced a 6-person queue."
The signal you want to send is that you are the person who runs the automation, not the person the automation ran over. Tailoring that framing to each posting is tedious work, and it is exactly what Refolk automates end-to-end: it reads the posting, pulls the AI-supervision evidence from your history, and drafts a cover letter that names the same tool the employer just listed in the JD.
Geography now determines whether you can fight back
Federal enforcement retreated in 2025 while state enforcement advanced, and where you live now determines whether the algorithm that fired you is legally reviewable. The EEOC removed its AI technical-assistance documents in 2025, and Executive Order 14281 (April 23, 2025) directs federal agencies to de-prioritize disparate-impact enforcement. Meanwhile:
- California's FEHA automated-decision regulations took effect October 1, 2025
- NYC Local Law 144 continues to require annual independent bias audits, with penalties up to $500 for a first violation and up to $1,500 per subsequent one
- The Workday class action is the leading federal test of AI hiring discrimination
Per the EEOC's own 2023 baseline, nearly 83% of employers use some form of AI or algorithmic tool in hiring. If you are in California or NYC, a records request or complaint has actual teeth. If you are not, your leverage is on the resume itself, which is why writing to the feature vector matters more than writing to a recruiter's taste.
A layoff survivor resume, section by section
The full rewrite, in order:
- Header. Name, city, one phone, one email. No graduation year. No headshot.
- Summary. Three lines. First line names the function and the last dated production stack. Second line is a single metric bullet. Third line names one tool the JD lists.
- Experience. Each bullet is metric plus denominator plus dated window. Reliability bullets go first in each role.
- Recent skills. Date-tagged: "Terraform (2025-2026), Snowpark (2026), Rust (2024-2025)". This kills the tenure-staleness signal.
- Education. No graduation years. Degree and institution only.
Do this per posting, not once. The 42% cost feature, the 80% performance feature, and the "could AI do this" filter all read the specific words in the JD, so your resume has to match them literally.
FAQ
Should I mention I was laid off on my resume?
No. The resume is not the place. If a role has a hard end date, use it, and let the cover letter or a screening call carry the context. AI layoff decisions at your next employer will weigh your bullets against the JD, not your exit narrative, so spend the space on metrics that pre-empt the 80% performance and 57% attendance signals.
Is it legal for my former employer to use sick-day data in a layoff model?
It sits on shaky legal ground. Julia Toothacre of ResumeTemplates.com noted that sick days, medical leave, and age involve legally protected categories. 31% of AI-using managers still feed sick-day data to the model, and 58% of employers cannot confirm the tool was tested for bias. In California (FEHA, Oct 1 2025) and NYC (Local Law 144), you have concrete recourse; elsewhere, federal disparate-impact enforcement was de-prioritized by EO 14281.
How do I write productivity resume bullets that beat an AI screen?
Three parts in every line: a number with a denominator, a before-and-after delta, and a dated window. "Cut invoice errors from 3.1% to 0.4% across 1.4M accounts in Q2 2026" beats "improved billing accuracy" every time. Performance and productivity scores are the 80% feature, so front-load them and repeat the JD's exact metric vocabulary.
If my job was replaced by AI, what do I write instead?
Do not write "I used AI." Write the supervision: audits, eval harnesses, precision numbers, incidents caught, queues collapsed. 13% of all managers surveyed have replaced a person's function with AI, so the receiving screener already assumes you know the tools. What sets you apart is proof you ran them at scale without breaking things.