What the 2023 amendment actually changed
Rule 702 is one sentence and four subsections. On 1 December 2023 two parts of it changed. The rule now reads:
A witness who is qualified as an expert by knowledge, skill, experience, training, or education may testify in the form of an opinion or otherwise if the proponent demonstrates to the court that it is more likely than not that: (a) the expert's scientific, technical, or other specialized knowledge will help the trier of fact to understand the evidence or to determine a fact in issue; (b) the testimony is based on sufficient facts or data; (c) the testimony is the product of reliable principles and methods; and (d) the expert's opinion reflects a reliable application of the principles and methods to the facts of the case.
The first change is the added preamble, the proponent demonstrates to the court that it is more likely than not that, which did not appear in the old rule. It puts the Rule 104(a) preponderance burden on the face of the text, and on the party offering the opinion.
The second is the rewrite of subsection (d), from the expert has reliably applied the principles and methods to the facts of the case to the expert's opinion reflects a reliable application of the principles and methods to the facts of the case. The question is no longer only whether the expert followed a sound process; it is whether the opinion actually stated is one the method can carry. The amendment targets overstatement.
The Advisory Committee said why: Many courts have held that the critical questions of the sufficiency of an expert's basis, and the application of the expert's methodology, are questions of weight and not admissibility. These rulings are an incorrect application of Rules 702 and 104(a). And the rule, it added, does not permit an expert to make claims unsupported by the expert's basis and methodology. Rule text: Cornell LII.
Who has to prove what, and to whom
Three things follow from the preamble. The proponent carries the burden. It is not enough to observe that the other side has not disproved the method. If a report says traffic fell because of the defendant's conduct and does not say what data established the fall or how the alternatives were ruled out, the gap is the proponent's problem.
The standard is preponderance, and the judge applies it. This is a Rule 104(a) determination: the judge is not asking whether a reasonable juror could accept the opinion, but deciding, more likely than not, whether the four conditions are met.
The old escape hatch is narrower. The reflex answer to a weak-basis argument used to be that it went to weight and belonged on cross-examination. The Committee said in terms that treating basis and application that way misapplies the rule. An argument you could once expect to lose on a motion and win before the jury now has to be answered in the report.
And the rule governs the opinion, not the resume. Qualification is a threshold; reliability is separate, and a well-qualified witness can be excluded on (b), (c), or (d) without any finding that the witness lacks expertise.
Subsections (a) and (c): the elements that rarely decide the motion
Subsection (a) asks whether specialized knowledge will help the trier of fact, and in search matters it is the easiest element. What a jury cannot work out unaided is how a search engine turns a page into a listing: how a crawler — automated software that requests pages and follows the links it finds — discovers a URL; why a page that exists is not necessarily one an engine will return; what a 301 permanent redirect does to the signals attached to an old address; and why a noindex directive removes a page from results while leaving it visible to anyone typing the address.
Subsection (c) asks whether the testimony is the product of reliable principles and methods. The obstacle is that the system under examination is proprietary and changed continuously by its operator: no published ranking algorithm to test, no error rate, no peer-reviewed literature. That does not defeat the inquiry. It relocates it. The method under examination is not the search engine; it is the analyst's protocol. A crawl with stated settings can be reproduced. A Search Console export for a stated property, date range, and dimension can be pulled again. Parsing server logs for verified crawler requests is arithmetic on a fixed record. Each has a defined input, a defined operation, and an output somebody else can obtain by repeating the steps.
Subsection (b): the third-party estimate problem
This is the first place a search opinion is exposed, and it is exposed because of how this industry measures. Most numbers in an SEO report are not measurements. They are models. A rank tracker is software that issues a query to a search engine on a schedule, from a chosen location and device, and records where a URL appeared; it measures a synthetic query, not what a real user saw. A keyword search volume is modeled from panel data and advertising-planner ranges, then bucketed. A commercial traffic estimate is an estimate multiplied by an estimate: an assumed click-through curve applied to an estimated position and volume. A link count describes the size and refresh rate of that vendor's crawl index, not the web.
None of that is disqualifying by itself. What makes it a Rule 702(b) problem is that in most cases the party's own measured data existed. Google Search Console reports the engine's own record of impressions, clicks, average position, and queries, on a rolling sixteen-month window. Analytics records sessions and conversions. Server and CDN logs record every request the server actually answered. Those are records of events. When an opinion says traffic fell forty percent, and the forty percent came from a third-party estimator while first-party data sat in the client's own account, the sufficiency question answers itself badly.
Third-party data is not off limits. For a competitor's site, or a period before the client began measuring, an estimate may be the only data there is. The discipline is to say so: name the source, state what the vendor publishes about how the figure is produced, state the likely direction of error, and never let an estimate carry a conclusion a measurement would have contradicted.
Subsection (d): the causal leap from two aligned timelines
Here is the characteristic failure of this field, and amended subsection (d) was written for it. The method is a comparison of two timelines: the date of the conduct and the date organic traffic fell. The opinion is that the conduct caused the loss. The gap between them is the whole case. Correlation supports a narrow statement — that the decline began within four days of the migration and was concentrated in a defined set of URLs. It does not support the causal statement unless the competing explanations have been tested and, where possible, excluded.
The candidates are rarely exotic. A broad core update — an engine-wide change to the ranking systems, announced and dated by the operator — may have landed in the same window. Demand may be seasonal, and the comparison period a holiday quarter measured against a summer one. A competitor may have launched. Paid media may have been cut, taking branded search demand with it. The analytics tag may have been removed in the deployment now being blamed, making part of the measured fall an artifact.
The response that satisfies (d) is not hedging. Hedged language over an untested inference is still an untested inference. It is elimination work that shows in the exhibits: segment the loss to the URLs actually affected, compare them against untouched sections of the same site, compare organic against the site's other channels over the same period, line the decline up against the published dates of any update in the window, and check a peer set that saw the update but not the conduct.
Where a confound cannot be excluded, the honest opinion is narrower and it survives: the decline in one subtree is attributable to the redirect failures on this record, and the sitewide decline is not. In my experience that is the version that reaches the jury, and the expansive version is the one that draws the motion.
What a compliant opinion states on its face
Rule 702 is satisfied on the page, not at the hearing. Reading a report as an opposing expert, these are what I look for first, and their absence is the outline of a motion:
- Every data source named. Which property, which account, which export, which tool and version, and whether the figure is measured or modeled.
- Every date range stated as a range. Including the comparison period and why it was chosen.
- Every exclusion stated. Filtered bot traffic, removed internal addresses, separated brand queries, omitted subdomains, dropped outlier days.
- Every confound named and addressed. Updates in the window, seasonality, competitor activity, paid-media changes, tracking changes.
- Every query and crawl re-runnable. Stated settings, filters, user agent, and capture date, with files produced so another analyst can repeat the run.
- Every conclusion scoped to what the data reached. Which pages, which queries, which period, and what the opinion does not claim.
Stating exclusions feels like an invitation to cross-examination. It is the reverse. An exclusion disclosed in the report is a methodological choice; an exclusion discovered in deposition is a concealed one.
What happens when part of an opinion fails
Total exclusion is not the common outcome. A motion is usually aimed at particular opinions, and a ruling can leave technical findings standing while striking the causal or quantified ones. An expert who can testify that a migration dropped four thousand URLs from the index, but not to what that cost, has still given testimony only a technical witness can give.
So the structure matters. Entangle the technical and the damages opinions in one narrative and a defect in the second can take the first with it; separate them, and a ruling against one does not automatically reach the other.
The report is also the pleading: an opinion that first appears in a declaration answering the motion invites its own objection, and the scheduling order closes the window for curing a defect before the motion is filed. Whether an opinion in your own case will hold up is an assessment for counsel, often with a second expert reading the report the way the other side will.
Frequently Asked Questions
What did the 2023 amendment to Rule 702 change?
Two things. It added the preamble "the proponent demonstrates to the court that it is more likely than not that," which puts the Rule 104(a) preponderance burden on the face of the rule and places it on the party offering the opinion. And it rewrote subsection (d) from "the expert has reliably applied the principles and methods" to "the expert's opinion reflects a reliable application of the principles and methods." The second change shifts the focus from the expert's process to whether the stated opinion is one the method can actually carry, which makes overstatement an admissibility problem rather than a cross-examination point.Can an SEO expert rely on third-party rank tracking and traffic estimates?
Sometimes, but not as the sole basis for a quantified opinion when first-party data exists. Rank trackers measure synthetic queries from a chosen location and device. Keyword volumes are modeled and bucketed. Commercial traffic estimates apply an assumed click curve to an estimated position for an estimated volume. Where the party's own Search Console, analytics, and server log data were available and producible, using an estimate instead is the weakness Rule 702(b) is aimed at. Third-party data is defensible for a competitor's site or a period before measurement began, provided the source, its method, and the likely error are stated.Who has the burden on a Rule 702 motion?
The proponent of the testimony, by a preponderance. That is now explicit in the rule text as amended in December 2023, and it is a Rule 104(a) question the judge decides rather than a question the jury weighs. Practically, it means the offering party cannot rest on the absence of contrary proof. If the report does not identify the data behind a figure, or does not show how competing explanations were addressed, the unfilled gap counts against the side that offered the opinion, not against the side that pointed it out.Is a drop in rankings after the defendant's conduct enough to show causation?
Not on its own, and this is where amended subsection (d) bites. Two aligned timelines are a correlation. The method supports a statement about when the decline began and which URLs it affected; it does not support a causal conclusion until the alternatives in the same window have been tested. That means core updates, seasonality, competitor activity, paid-media changes, analytics tracking changes, and unrelated site changes. Where a confound cannot be excluded, the opinion that survives is the narrower one that stays inside what the data reached.Does Rule 702 apply differently to a technical expert than to a scientist?
The rule is the same and the gatekeeping obligation is the same. Rule 702(a) expressly reaches "scientific, technical, or other specialized knowledge," and the Supreme Court confirmed in Kumho Tire that a trial court's reliability obligation extends to all three. What differs is which reliability considerations do useful work. Testability and error rate apply readily to a laboratory protocol; for a search analysis they attach to the analyst's documented procedure, not to the search engine's undisclosed ranking system.Can a court admit part of an expert's opinion and exclude the rest?
Yes, and in my experience that is the more common result than wholesale exclusion. A ruling can leave technical findings intact while striking a causal attribution or a quantified loss figure built on top of them. That makes the structure of the report consequential: technical opinions stated on their own data, with the economic consequences handled separately, are far more likely to survive a defect elsewhere than a single undifferentiated narrative in which every conclusion depends on the weakest link in the chain.Should an expert address Rule 702 in the report itself?
The report does not need to recite the rule, but it needs to answer it. Each subsection maps to something that either appears on the page or does not: what specialized knowledge the opinion supplies, what data it rests on and where that data came from, what procedure produced the analysis, and how the stated conclusion is bounded by what the procedure could show. A report drafted that way answers the motion before it is filed. A report that states conclusions and attaches charts leaves the proponent making the showing for the first time under time pressure.Published