Kumho Tire and Technical Expertise

What the Court decided in 1999

Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999), is the case that governs a search expert, and almost nobody writing in this field mentions it.

Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993), had held that the Federal Rules, not Frye, supply the standard, that the general-acceptance test was absent from and incompatible with the Rules, and that the trial judge has a gatekeeping role under Rule 104(a). It offered four non-exclusive reliability considerations: whether the technique can be and has been tested, whether it has been peer reviewed and published, its known or potential error rate and the standards controlling its operation, and general acceptance. And it directed attention to principles and methodology rather than to the conclusions they generate.

Daubert was a scientific-evidence case, and for six years litigants argued about whether it reached anything else. Kumho Tire answered that. The gatekeeping obligation applies not only to scientific testimony but to technical and other specialized knowledge — all three categories Rule 702 names. The Court declined to draw a workable line between scientific and technical expertise, and made clear the obligation does not lighten when expertise comes from doing the work rather than from a laboratory. The opinion is at Cornell LII.

Why the Daubert factors are a menu, not a checklist

The second half of Kumho Tire gets misread in both directions, and it cuts for and against the technical expert at once.

The Daubert factors are neither mandatory nor exhaustive. A trial court has broad latitude both in deciding how to assess reliability and in the assessment itself, because, as the Court put it, too much depends upon the particular circumstances of the particular case. A judge need not run a technical opinion through four boxes and exclude it when one comes up empty.

That is the half practitioners like to quote. The other half decides most technical challenges: some of Daubert's questions can help to evaluate the reliability even of experience-based testimony, and it will at times be useful to ask them even of a witness whose expertise is based purely on experience. The absence of a peer-reviewed literature does not excuse an expert from the reliability inquiry. It relocates the inquiry to the expert's own procedure.

The flexibility runs to the form of the showing, not to whether a showing is required. A judge may accept a reliability demonstration that looks nothing like a clinical trial, and is not obliged to accept one that consists of a career.

Why a search expert falls on the Kumho side of the line

Everything about search analysis places it in the technical and specialized category, and the reasons are worth stating plainly because an opposing motion will recite them.

The system under examination is proprietary. Its operator does not publish the ranking algorithm or an error rate for it, and changes it continuously — announcing broad core updates, which are engine-wide changes to the ranking systems, several times a year, and making many more unannounced changes. No controlled experiment is available: you cannot hold the index constant or run a treatment against a control on the live engine, and the same query from two devices in two cities can return different results by design.

There is also no licensure, no registration, no mandatory certification, no professional body with disciplinary authority over anyone practicing search optimization, and no accepted textbook whose omission would be malpractice. What exists is the operator's published documentation — dated, versioned, usable as a contemporaneous external referent — and a body of practice.

None of that makes the testimony inadmissible. It makes the reliability inquiry one about the analyst rather than the field, which is precisely the situation Kumho Tire addresses.

A qualification is not a methodology

In the plainest available form: "I have done SEO for thirty years" is a qualification, not a methodology.

Rule 702 opens by asking whether a witness is qualified by knowledge, skill, experience, training, or education. Experience is expressly on that list, and thirty years of it will get a witness past the qualification threshold in most courts. What it does not do is answer subsections (b), (c), and (d), which ask what data the opinion rests on, what principles and methods produced it, and whether the stated opinion reflects a reliable application of them. A tenure answers none of the three.

The failure mode is recognizable at a glance, and I have read it in more than one opposing report. The opinion is stated as a professional judgment, and the support is the author's background plus assertions in the form in my experience, this is what causes that. No stated data set, no stated date range, no stated exclusion, no alternative considered and rejected. Cross-examination on such a report need not attack the conclusion; it only has to establish that no one else could reach or refute it.

The Seventh Circuit put the general version memorably in Zenith Electronics Corp. v. WH-TV Broadcasting Corp., 395 F.3d 416 (7th Cir. 2005): a witness who invokes "my expertise" rather than analytic strategies widely used by specialists is not an expert as Rule 702 defines that term, and reliable inferences depend on more than say-so. That was a damages case, not a search analysis, but the sentence is general and it travels.

What a documented protocol looks like

What survives a Kumho inquiry is a documented, reproducible protocol — in a search matter, six things stated before any conclusion appears:

• Stated sources. Which Search Console property, which analytics view, which log files from which servers and dates, which crawl with what tool and version, which third-party exports. Measured data identified as measured, modeled data as modeled.
• Stated date ranges. The analysis window, the comparison window, the reason for that comparison, and the time zone the data is reported in.
• Stated exclusions. Bot traffic filtered, internal addresses removed, brand queries separated, outlier days dropped — each named, with the rule applied.
• Stated confound controls. Which competing explanations were considered, what was examined to test each, and which could not be excluded.
• Re-runnable queries. Crawl configurations, user agents, filter expressions, and export parameters recorded so another analyst can repeat the run and obtain the same file.
• A preserved record of the raw material. Original exports retained unaltered, derived working files kept separately, so the path from source to chart is traceable.

None of that is exotic. It is ordinary craft, written down. It decides admissibility because it converts an experience-based opinion into a procedure with inputs, operations, and outputs a second analyst can test.

Making the reliability considerations do work where they can

Because the factors are neither mandatory nor exhaustive, the useful exercise is not to force all four but to apply the ones a search protocol can satisfy, and to say why the others do not fit.

Testability

This applies directly and is the strongest ground available. A crawl with stated settings can be re-run. A Search Console export for a stated property, range, and dimension can be pulled again. A log parse is arithmetic on a fixed file. An opposing expert who repeats the run and gets a different number has found something; one who cannot repeat it at all has found something worse.

Error rate and standards controlling operation

There is no error rate for a search engine. There is one for a sampling procedure, a classification rule, or an extrapolation, and where an opinion uses any of those, the rate can be stated. Standards controlling operation are the documented settings: crawl depth, rate limits, rendering mode, user agent, and the step verifying that requests attributed to a crawler actually came from it.

Peer review and general acceptance

There is no peer-reviewed literature on the ranking system. There is dated, published operator documentation and a set of widely used measurement conventions. The defensible version of this factor identifies the contemporaneous guidance the analysis was measured against, with its date, rather than claiming a literature that does not exist.

Qualification and reliability are separate questions

Two distinct challenges get filed under one heading, and conflating them costs an expert more often than either alone.

The first is a qualification challenge: this witness lacks the knowledge, skill, experience, training, or education to give this opinion. Against a practitioner with a long record it usually fails on its face. Where it succeeds is on scope. Fifteen years of technical search work qualifies a witness to opine on crawl behavior, index coverage, and redirect handling. It does not, without more, qualify the same witness to compute lost profits or value a business.

The second is a reliability challenge, and it is indifferent to the resume. A witness can be the most qualified person available and still have produced an opinion resting on insufficient data, an undisclosed method, or an inference the method cannot support. Courts exclude that opinion without any finding that the witness is unqualified, and the amended Rule 702 makes doing so more straightforward than it was. Where a background is unimpeachable and the report states no data sources, no date ranges, and no exclusions, the qualification is doing no work at all.

How an opposing expert gets measured against this

I read opposing reports as often as I write my own, and the Kumho frame is the one I read them in. The questions are the same every time, and worth putting to any expert in this field, including your own.

1. What did you look at? Named accounts, files, exports, and tools with versions. "Industry data" is not an answer.
2. Over what period, compared to what period, and why that one?
3. What did you leave out, and under what rule?
4. What else could explain this, and what did you do to test it? A witness who has considered no alternative has not applied a method; one who names three and shows the work has.
5. Can I reproduce it? Given the settings and the source files I should reach the same number. If I cannot, the opinion rests on its author.
6. Where does your opinion stop? An expert who can state the boundary has one. An expert who cannot has an impression.

A report that answers those six in writing is difficult to exclude and unpleasant to cross-examine, because every answer is on the page and checkable. A report that answers none of them relies on the court accepting experience in place of method — the substitution Kumho Tire tells a trial judge to examine.

Frequently Asked Questions

Published 2026-08-11