Opinions Are Free Now

In the spring of 2023 a New York lawyer filed a brief that cited half a dozen court decisions — names, docket numbers, quotations lifted from the opinions. The cases did not exist. He had asked a chatbot for supporting authority and it had handed him exactly what supporting authority looks like: confident, formatted, fluent, and entirely hollow. When the judge asked him to produce the opinions, the chatbot obligingly wrote those too. He was sanctioned in front of the whole profession.¹

The easy lesson is don't trust the chatbot. The harder one — the one this essay is about — is that the brief read as evidence right up until somebody checked. Fluency had come unhooked from truth, and nobody in the room could tell, because the test we all use to judge a claim at a glance, does this sound right, had quietly stopped carrying any information.

People analytics has been filing that brief for thirty years. It just never needed a machine to write it.

They say the answer sounds right

Walk into most rooms where a people decision gets made and the currency is the confident take. The keynote with the four habits of high-performing teams. The best-practice deck benchmarked against three admired companies nobody actually studied. The vendor white paper. The senior leader who has seen this before. None of it is sourced in any way you could check; all of it is delivered in the cadence of someone who knows. Pfeffer and Sutton named this twenty years ago and were not gentle about it — most management practice, they found, runs on casual benchmarking, half-remembered ideology, and what worked for somebody once, with hard facts the conspicuously missing ingredient.² The field has a folk epistemology, and the folk epistemology is a confident person said it and it sounded right.

Then the machine arrived and did to opinions what the printing press did to pamphlets. A plausible, well-organized, citation-shaped answer to any people question you can phrase now costs a fraction of a cent and arrives before you've finished your coffee. The supply of things that sound right went vertical. That is genuinely useful — I use it every day — and it is also the end of sounds right as a filter, because a filter only works when the thing it screens for is scarce. It isn't anymore.

When opinions are free

Here is the principal issue, and it's an economic one before it's a methodological one.

When producing a confident answer was expensive — it took a consultant, a deck, a degree of nerve — the confidence itself carried a little signal. Somebody had paid something to say it. Now the production cost is zero, and a zero-cost signal is worth what you paid for it. The fluent take has become like a currency printed without limit: still in everyone's pocket, no longer worth anything. Whatever is going to separate a real claim from a plausible one, it cannot be how authoritative the claim sounds, because that is now free and infinite.

The only thing left that is not free is the part the chatbot faked: a source you can actually walk to. Not a citation-shaped string — a real one, that resolves, that was graded by someone who knew how, that you could go read and disagree with. When opinions cost nothing, the sourced claim is the scarce good. Everything in this essay follows from that one sentence.

The discipline already exists

The reassuring part — and the unglamorous part — is that none of this is new, and we don't have to invent the fix. Medicine had this fight first and largely won it. Evidence-based medicine made the radical-sounding claim that a treatment's pedigree and a senior physician's confidence are not evidence, that you rank what you know by how it was found out, and that you appraise the source before you trust the finding.³ Denise Rousseau spent a career importing that discipline into management against real resistance, and the movement she helped start gave it a working shape: there is more than one source of evidence — the scientific literature, your own organization's data, practitioner judgment, stakeholder values — and the move is not to pick one but to appraise each for how trustworthy it actually is before you let it drive a decision.⁴⁵

There is an even older and more humbling finding underneath all of it. In 1954 Paul Meehl gathered up the studies that pitted expert clinical judgment against a plain statistical rule and reported that the rule usually won — a result that has only hardened in the seventy years since.⁶ The expert who has seen this before is, on average, beaten by a checklist built from data. That is not an attack on expertise; it is a statement about where to put your trust when the two disagree. It is also the whole argument of this essay in miniature: the confident human voice is the thing to be suspicious of, and the traceable, gradeable record is the thing to lean on.

So the move is not exotic. Every claim resolves to a source. Every source carries a grade — how it was found out, how much to trust it, whether it would survive being checked. And the claims that don't resolve go in an honest we don't know yet pile instead of getting rounded up into a confident sentence to fill the silence.

Source superiority is a moat, not a slogan

Say all of that out loud in a marketing meeting and it sounds like a tagline. Evidence-based, source-backed, rigorous. Everybody says it; almost nobody can show it; and a thing you assert about your own rigor is, of course, exactly the kind of unsourced confident take this essay is against. So the only honest version is to build the receipts and leave them out where anyone can pull on them.

That is what the rest of this site is. Every public claim we make is wired to the evidence behind it — a corpus read at chapter fidelity, findings extracted to citation grade, a registry that grades its own sources, methods original enough to file — and the links are checked on every build, so a claim that loses its source breaks the build instead of quietly shipping.⁷ You can walk the graph yourself. Some of it, you'll find, traces to a strong prior; some of it traces to thinner ground than we'd like; and where the evidence is thin, the page says so. That last part is the tell. A source-superiority claim that never admits a weak spot is just confidence wearing a lab coat — the same hollow brief, better dressed.

This is the part that costs something, and the cost is the point. Citing your sources is slower than having a take. It is less impressive in the meeting, because the immediate manager is the construct that actually moves retention in your data, and here's the prior it's measured against will never land as cleanly as people leave bad bosses. It exposes you — a take can't be wrong in public the way a sourced claim can. But a claim that can be checked, and graded, and occasionally caught being wrong, is the only kind that's worth anything in a world where the alternative is free. The depth of that record, and the discipline of keeping every claim tied to it, is not a slogan you can copy by Friday. It's a moat you dig for years. That's exactly why it's defensible.

Cite or it doesn't count

Step back to the lawyer and his beautiful, empty brief. What undid him wasn't that the machine was confident — confidence was the easy part, the free part, the part anyone can now generate by the page. What undid him was that when one person finally walked to the source, there was nothing there.

People analytics is about to get the same brief delivered, at scale, in fluent and reasonable prose, to every leader making a decision about other people's lives. The field's old folk test — does it sound right — was already weak; against an infinite supply of things that sound right, it is finished. The replacement isn't a smarter model or a louder expert. It's the oldest discipline in evidence-based practice, dragged into the room and made non-negotiable: show the source, grade it, and if you can't, say so. When opinions are free, the receipt is the product. Cite it, or it doesn't count.

---

This is the proof essay for a portfolio differentiator: every public claim PeopleAnalyst makes is traced to graded evidence, and the depth of that evidence base is the defensible asset. The apparatus this essay points to is real and walkable — the proof graph (every claim → its source, checked on each build), the Library corpus, and the source-graded Principia registry. Its sibling Show Your Work takes up the adjacent move — making the reasoning legible, not just the sources. No claim in this essay is unsourced: each footnote names a real, checkable work, and where the ground is thin the prose says so rather than rounding up.

Footnotes

1. Mata v. Avianca, Inc., No. 22-cv-1461 (S.D.N.Y. 2023), Judge P. Kevin Castel — counsel were sanctioned for submitting a brief containing judicial decisions fabricated by ChatGPT, complete with invented quotations and, when challenged, fabricated full "opinions." Widely reported as the first prominent case of AI-hallucinated legal citations. ↩

2. Jeffrey Pfeffer & Robert I. Sutton, Hard Facts, Dangerous Half-Truths, and Total Nonsense: Profiting from Evidence-Based Management (Harvard Business School Press, 2006) — the argument that much management practice rests on casual benchmarking, ideology, and unexamined belief rather than appraised evidence. ↩

3. David L. Sackett, William M. C. Rosenberg, J. A. Muir Gray, R. Brian Haynes & W. Scott Richardson, "Evidence based medicine: what it is and what it isn't," BMJ 312 (1996): 71–72 — the founding statement that clinical practice should integrate individual expertise with the best available appraised external evidence, ranked by how it was obtained. ↩

4. Denise M. Rousseau, "Is There Such a Thing as 'Evidence-Based Management'?" Academy of Management Review 31, no. 2 (2006): 256–269 — the Academy presidential address that brought the evidence-based discipline into management. ↩

5. Eric Barends, Denise M. Rousseau & Rob B. Briner, Evidence-Based Management: The Basic Principles (Center for Evidence-Based Management, 2014) — defines the practice as integrating four sources of evidence (scientific, organizational, experiential, stakeholder), each critically appraised for trustworthiness and relevance before use. ↩

6. Paul E. Meehl, Clinical versus Statistical Prediction: A Theoretical Analysis and a Review of the Evidence (University of Minnesota Press, 1954); the finding that mechanical/actuarial prediction generally equals or exceeds holistic expert judgment was confirmed meta-analytically by William M. Grove, David H. Zald, Boyd S. Lebow, Beth E. Snitz & Chad Nelson, "Clinical versus Mechanical Prediction: A Meta-Analysis," Psychological Assessment 12, no. 1 (2000): 19–30. ↩

7. The proof graph and its build-time check are described at /evidence; the discipline behind it — claims traced to graded evidence, weak spots disclosed — is the subject of the source-superiority doctrine the site holds itself to. Construct grading and priors draw on the measurement tradition of Lee J. Cronbach & Paul E. Meehl, "Construct Validity in Psychological Tests," Psychological Bulletin 52, no. 4 (1955): 281–302. ↩