How to Verify a People-Practice Claim
Use this when someone (a vendor, a consultant, a LinkedIn post, an exec) hands you a claim about people practices that you're about to act on. 30-minute protocol below. 6/6 of the Phase 0 panel described this verification work as their biggest unsolved pain.
Downloadable version: this guideline ships as a scored Excel assessment — the Peoplense Claim Check (4 questions × 1–5, auto-verdict, protocol sheet). Free, CC BY 4.0. It pairs with the quiet-quitting Decision Brief.
What 2026 corpus says about claim reliability
Two studies in our library show why this matters operationally:
- Harvard Kennedy School (2025-10): field study at a multinational financial services firm. Self-ratings systematically lower for women / women of color. Hiding self-ratings reduced anchoring but managers substituted historical ratings as an alternative anchor. Any claim about "fair performance ratings" has to survive this finding. (Read in library)
- ScienceForWork (2025-10): meta-analysis of 146 studies (Van Dijk et al. 2012). Subjective performance ratings systematically underestimate demographically diverse teams (r = −0.06) while overestimating job-related diversity (r = +0.09). Biases largely disappear under objective measurement. Any "research shows X" claim built on subjective ratings is suspect. (Read in library)
- Training Magazine (2026-04): 61% of people worry business leaders deliberately mislead. Trust = Connection × Character × Competence — all three needed. (Read in library)
These aren't abstract — they're the reason a "best practice" can be dead wrong even when it's everywhere.
The four questions to ask
1. Who is making the claim, and what do they gain?
| Source type | Their incentive | Calibration |
|---|---|---|
| Vendor / SaaS | Sell their category | Stats favor their product |
| Consultancy | More billable work | Stats favor problems they sell solutions to |
| Academic / peer-reviewed | Citations, tenure | Lower direct incentive — but check who funded |
| Industry body | Membership relevance | Aggregated across members; may miss non-members |
| Practitioner blog / LinkedIn | Engagement, brand | Anecdotal generalized to universal |
| News summary | Clicks | Original nuance often lost |
This isn't distrust — it's calibration. A vendor claim isn't invalid; just weight it.
2. Is the original source actually accessible — and have I read it?
Most people-practice claims circulate via summary-of-summary:
Academic study → industry-body report → news article → LinkedIn post → another LinkedIn post
Rule: trace it back. If you can't reach the original, the claim is unverified.
Red flags the chain is broken:
- "Studies show…" with no citation
- A specific number ("70% of employees…") with no source
- "Research from [big institution]" without a specific paper named
- "According to a survey…" with no methodology
- Cited in 3 places, each citing the others (circular)
3. Does the claim's context match mine?
| Mismatch | Example |
|---|---|
| Geographic | "Remote work +13% productivity" — US white-collar 2020-21. Doesn't transfer to KSA manufacturing 2026. |
| Industry | "OKRs drive engagement" — tech-sector data. Banking/healthcare/manufacturing behave differently. |
| Time | "Annual reviews are dying" — true 2018, partially recovered since 2023. |
| Subjective vs objective measurement | The ScienceForWork meta-analysis: subjective ratings show diversity penalty; objective measures don't. Same question, two answers. |
| Cultural context | "Direct feedback improves performance" — low-power-distance findings. Needs adaptation in high-power-distance (Hofstede). |
4. What would change my mind?
For any claim you're about to act on: "If [X] were true, I'd conclude this claim is wrong. What's X?"
Example: "Stay interviews reduce turnover 20%." What would change my mind? A randomized comparison where the no-intervention group had equal-or-lower turnover. Does that evidence exist? Most "stay interviews work" claims are observational case studies. Knowing what would falsify the claim is the test of whether you actually understand it.
The 30-minute protocol
| Time | Step |
|---|---|
| 0–5 min | Source-incentive map (Q1) |
| 5–15 min | Trace to original (Q2). Google the specific claim in quotes. Read abstract/exec summary minimum. |
| 15–20 min | Context check (Q3) |
| 20–25 min | Falsification test (Q4) |
| 25–30 min | Write a 3-sentence verification note: claim, source, applicability. File it. |
If it takes more than 30 min, the claim is either worth your own analysis or worth leaving uncited.
The "I've seen this in 10 places" trap
Repetition ≠ verification. Three places where bad claims live forever:
- Vendor sales decks — same stat across vendors who all bought the same syndicated report
- LinkedIn thought-leader posts — high circulation, near-zero verification
- Conference keynotes — same speakers reuse the same stats year after year
When everyone is so sure, no one is checking.
What to do with unverifiable claims
- Don't cite it. Your credibility, not the source's.
- Cite with calibration. "A vendor study suggests…" makes the uncertainty visible.
- Use for direction, not decisions. A claim can shape a hypothesis without being the basis for policy.
The one response that destroys long-term credibility: citing the claim as fact. Works once; fails when the auditor asks where the 70% came from.
Where this came from
The 6/6 unanimous verification-pain finding from the N=6 Phase 0 panel (2026-05-09). Every respondent described compile-from-multiple-sources workflows that took hours. This guideline tries to make that work explicit and repeatable.
Sources in our library
- Self-ratings and bias in performance reviews — Harvard Kennedy School, 2025-10
- Diversity: Performance Impact or Perception Bias? — ScienceForWork, 2025-10
- The 3 C's of trust: Connection, Character and Competence — Training Magazine, 2026-04