The two questions, and why they separate
Every damages claim built on search traffic contains two questions that are routinely argued as one. The first is whether the claimant was damaged at all by this defendant. The second is how much. In most states lost profits must be proven to reasonable certainty, though the formulation varies and some jurisdictions apply a looser rational-basis test; the operative requirement is evidence providing sufficient data from which the trier of fact can properly estimate the loss.
The distinction that matters here is that the fact of damage is generally held to a more demanding standard of proof than the amount. Once injury is established, reasonable estimation of quantum is tolerated. That allocation is brutal in search cases, because it is exactly backwards from where the evidence is strong. The fact of a traffic loss is usually visible in the first hour of looking at the data: sessions fall, impressions fall, the chart bends. What is not visible in that chart is who or what caused it. The half of the case the law holds most strictly is the half the data addresses least.
So the working rule for this whole area is that a search damages opinion is a causation opinion wearing a damages label. If the causal separation is not done, the arithmetic downstream of it does not matter, and no amount of precision in the revenue model repairs it.
Why the plaintiff's history is often the first problem
The new business rule — the doctrine barring lost-profits recovery by an enterprise with no track record — has been abandoned in most states, but it survives in modified form in several. Georgia retains it in its strongest form. Illinois carves out exceptions for products identical to existing ones in known markets and for acquisitions of going concerns. Iowa, Washington, and Pennsylvania apply versions functionally close to the reasonable-certainty standard.
This is not an abstraction in search litigation, where the claimant is frequently a young e-commerce or lead-generation business with eighteen months of trading history and a growth projection shaped like a hockey stick. Two consequences follow. First, the choice of damages method is partly dictated by the jurisdiction's treatment of young enterprises, not only by what the data supports. Second, the plaintiff whose own history is too short to serve as a baseline has to reach for external comparables, which imports a foundation problem in place of a data problem.
Counsel should settle this question early, because it determines which of the three approaches below is even available and therefore what the expert should be asked to do.
The three accepted approaches
Damages practice recognizes three families of method, and search cases use all three.
Before-and-after. Compare performance in a pre-event period against the damage period and treat the difference as the loss, on the assumption that the prior level would have continued but for the conduct. It requires a clean, sufficiently long baseline. It is the method most commonly attempted in search matters and the one most commonly excluded, for the reason set out in the next section.
Yardstick, or benchmark. Substitute a comparator for the claimant's own unusable history — comparable businesses, comparable undamaged locations, or industry averages. Search offers unusually good internal yardsticks: the claimant's other acquisition channels as a control for demand-side shifts, its unaffected page groups as a within-site control for a sitewide change, competitor visibility over the same window as a market control, and query-demand data for the affected keyword set. A design comparing affected against unaffected cohorts across the event date is testable, which is what makes it defensible.
But-for modeling. Construct the world absent the conduct — typically through multivariate regression, market-share analysis, or a projected sales model — and subtract actuals. Regression is the approach most explicitly endorsed in the Reference Manual on Scientific Evidence, and the Second Circuit has approved a residual impact analysis where data limitations foreclosed a full regression. It is also the approach that most often exceeds what a search dataset can support.
Why search is harder than most damages settings
Every one of those methods assumes a counterfactual — what would have happened but for the conduct. In most commercial settings that counterfactual is at least approximately stable. In organic search it is not, and this is the single structural fact that distinguishes the discipline.
- The ranking system changes constantly. Google confirms several broad ranking updates a year, each with a published start date and a rollout measured in weeks, and ships continuous unannounced change besides.
- Seasonality is strong and category-specific, and it is measured in the same series as the alleged harm.
- Competitors act independently. A rival's investment, expansion, or exit moves the claimant's traffic without anyone touching the claimant's site.
- Query demand itself shifts. The number of people searching a phrase is not a constant, and a decline in demand looks identical to a decline in visibility if you only measure sessions.
- The relationship between position and traffic is itself moving. SparkToro's analysis of Similarweb clickstream panel data found 68.01% of United States Google searches ended without a click between January and April 2026. That panel covers desktop and mobile browser sessions and excludes the Google mobile app, so SparkToro's own view is that the real rate is likely higher.
A naive before-and-after on organic sessions charges the defendant with everything Google, the calendar, the market, and the interface did in the same window. That is not a rhetorical criticism. It is the specific defect for which damages models get thrown out.
Where damages models actually get excluded
Three decisions carry most of the weight in this area, and none of them is about search. That is why they work: they state the failure in general terms, and the search version of each is easy to see.
Concord Boat Corp. v. Brunswick Corp., 207 F.3d 1039 (8th Cir. 2000), reversed a verdict because the plaintiffs' econometric model "did not incorporate all aspects of the economic reality" of the market and "failed to account for market events that both sides agreed were not related to any anticompetitive conduct." The search analogue is a model that ignores a core update, a migration, or a competitor's entry inside the damage window. The opinion is available from Justia.
LifeWise Master Funding v. Telebank, 374 F.3d 917 (10th Cir. 2004), affirmed exclusion of an S-curve growth model that "was not in regular usage for predicting future profits, was not peer reviewed, has no uniform usage in any known industry, and is capable of manipulation to achieve virtually any desired result." The witness was LifeWise's own chief executive, not a retained expert, and he was separately held unqualified. Search growth projections are S-curves, which makes the quoted language uncomfortably portable.
Zenith Electronics Corp. v. WH-TV Broadcasting Corp., 395 F.3d 416 (7th Cir. 2005), affirmed exclusion and refused to let internal company projections stand in: "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 "[r]eliable inferences depend on more than say-so, whether the person doing the saying is a corporate manager or a putative expert."
Since the amendment to Federal Rule of Evidence 702 effective 1 December 2023, the failure to separate the defendant's conduct from a concurrent market event is not a weight argument that goes to the jury. It is an admissibility argument the proponent must answer by a preponderance before the opinion is heard.
What every approach has to include
Three requirements sit on top of all three methods, and search experts get them wrong by omission rather than by error.
Avoided costs. Only incremental profit is recoverable. Lost sales that never happened also never incurred cost of goods, fulfillment, payment processing, or the advertising spend attached to them. A traffic delta multiplied by revenue per session is a revenue figure, not a damages figure.
Mitigation. What the claimant did after the loss, and what it reasonably could have done, is part of the record. In search matters this is unusually concrete: whether the defect was remediated, when, and how quickly recovery followed is documented in crawl and index data.
Present value. Future losses are discounted at a rate reflecting the risk-free rate, the company's cost of capital, and company-specific risk.
None of those three is a search question, which is the point. The division of labor that holds up is that the search expert establishes what changed, when, and whether it is attributable to the conduct rather than to an algorithm, a season, or a competitor — then hands a traffic or visibility counterfactual to a financial expert who converts it into money. A search expert who computes the dollar figure invites a Zenith challenge on the financial half and risks the whole opinion. A financial expert who assumes the traffic loss without anyone establishing it has an unsupported input under Rule 702(b).
The same analysis, run by the defense
Everything above is symmetrical, and a reference that only serves claimants is not a reference.
A defendant's most productive line is rarely that the loss did not happen. It is that the loss was not caused by this defendant and that the opposing expert cannot say how much of it was. The questions that do the work are narrow: which confirmed updates overlapped the damage period, and what was done about each; whether pages the defendant never touched moved by a similar proportion; whether the decline began before the conduct; whether the model was run across a date on which nothing happened; and what portion of the decline the analysis leaves unexplained. An expert who cannot answer the last question has not performed the decomposition, and that single answer usually tells you the state of the opinion.
What does not work is the general version — Google changes constantly, therefore causation is unknowable. It proves too much, would defeat every search claim ever brought, and experienced counsel treat it as noise. The specific, dated version is the one that lands.
The three procedures in this section are the operational versions of this problem: separating an algorithm update from the defendant's conduct, measuring what AI answer features actually took, and handling the 2024 Google documentation leak as an evidentiary object rather than as SEO commentary.
Frequently Asked Questions
What has to be proven to recover lost profits in a search case?
Two things, held to different standards. The fact of damage — that this defendant injured the claimant — is generally held to the more demanding standard, usually reasonable certainty. The amount may then be estimated, provided the evidence supplies sufficient data from which the trier of fact can properly calculate it. In search matters this allocation is uncomfortable, because the fact of a traffic decline is visible immediately in analytics while its cause is not. The practical consequence is that most of the expert work, and most of the risk, sits on causation rather than on the arithmetic.Which damages method works best for organic search traffic?
Usually a yardstick design rather than a straight before-and-after. Search supplies internal controls that most damages settings never get: the claimant's other acquisition channels, its own page groups that the conduct never touched, competitor visibility over the same window, and query-demand data for the affected terms. Comparing an affected cohort against an unaffected one across the event date produces a testable estimate with an assessable error, which is what reliability analysis under Rule 702 asks for. A before-and-after model can still work, but only where the baseline is genuinely clean and the confounds in the window are addressed rather than assumed away.How much trading history does a claimant need?
Enough to establish a baseline, and in several states enough to clear the new business rule. That rule — barring lost-profits recovery by enterprises without a track record — has been abandoned in most jurisdictions but survives in modified form in others; Georgia retains it in its strongest form, Illinois carves out exceptions for identical products in known markets and for acquisitions of going concerns, and Iowa, Washington, and Pennsylvania apply versions close to reasonable certainty. Because search claimants are often young businesses with short histories, this question should be settled before the method is chosen, not after.Can a search expert calculate the dollar value of lost traffic?
It is the wrong division of labor and it puts the whole opinion at risk. The defensible split is that the search expert establishes what changed technically, when it changed, whether the change is attributable to the conduct rather than to an algorithm update, seasonality, or a competitor, and what the traffic counterfactual would have been — then stops. A financial expert takes that traffic delta as an input and applies conversion value, incremental margin, avoided costs, mitigation, and discounting. Each expert must be able to state precisely which assumptions came from the other, which is also a disclosure question.What is the most common reason a search damages model is excluded?
Failure to account for events in the damage window that nobody attributes to the defendant. That is the Concord Boat defect: a model that ignores market events both sides agree were unrelated to the challenged conduct. In search the events are a confirmed core update, a site migration, a redesign, a paywall, a competitor's expansion, or the structural decline in clicks per search. Since the December 2023 amendment to Rule 702, that failure is an admissibility question the proponent must answer by a preponderance, not a weight argument saved for cross-examination.How does a defendant attack a search damages claim?
Specifically and with dates. Identify the confirmed ranking updates whose rollout windows overlap the claimed inflection; show that pages the defendant never touched fell by a similar proportion; establish a pre-trend showing the decline began before the conduct; run the opposing expert's own model across a date on which nothing happened; and ask what portion of the decline the analysis leaves unexplained. The general argument that search is unknowable because Google changes constantly does not work — it would defeat every claim in the field, and courts and experienced counsel discount it accordingly.Does the shift toward AI answers make older damages models unreliable?
It makes them incomplete, which for admissibility purposes is much the same thing. Every model built before 2024 implicitly treats the relationship between ranking position and traffic as stationary. Measured zero-click rates say otherwise, and a model with a 2023 baseline running into 2026 spans that change. Charging an industry-wide structural shift to a defendant is the same failure as ignoring a core update, only slower and harder to see in a chart. It also cuts the other way: a defendant now has a real, measurable confound to raise, and an expert who can quantify it is useful to either side.Published