Defamation in Search Results

The question is not whether it is false, it is what was there

In a defamation matter the search expert is not retained to decide whether the statement is false. That is for the finder of fact. The expert's questions are narrower: what appeared, where, for which queries, on which dates, to whom, and for how long. That is a reconstruction problem, and it is harder than it sounds, because what is being reconstructed was never stored anywhere as an authoritative record.

A defamatory web page has an existence you can document - a URL, a host, a publication date, a revision history in the content management system that produced it. Its prominence does not. Prominence is an event that happened to one user, in one place, on one device, at one moment, and nobody archived it. So when counsel says the page "was the first result for his name," the useful response is a series of questions: first for whom, from where, on what date, signed in or out, on a phone or a desktop, and what proof of any of that survives?

The reconstruction is possible, and it is often the strongest part of the plaintiff's technical case. But it is built from dated, documented, repeatable observations, not from a memory and an undated screenshot.

Why a screenshot does not prove what Google showed

A SERP (search engine results page, the page of results a query returns) is assembled for the person requesting it. Google says so in its own documentation, and the list of reasons it gives is the most important technical fact for anyone offering a search result as an exhibit.

Google identifies time first: "Time may be one of the biggest reasons why results vary between people," because new content is published continuously and ranking changes roll out gradually across data centers. It identifies location - "If you search pizza in a particular city, you get results relevant to that location" - along with language, device type, personalization, and context from recent searches. Personalization "changes the order of the search results," and even with it off the system still uses location, language, device type, and current searches. Google's explanation of why results differ between people is the citation.

The consequence is blunt. A screenshot proves what one browser, in one location, on one device, at one moment, in one signed-in state, was shown. It does not prove what "Google shows." A capture protocol that survives cross-examination records the query exactly as entered; UTC and local timestamps; the country and language parameters used; a signed-out state and clean browser profile; device, viewport, and user agent; and a full-page capture rather than a crop. Repeat across dates and locations and the snapshot becomes a series, which is what reconstruction requires.

The queries that matter are the ones your own data may hide

Reputation matters run on name-based queries: the person's name; the name plus a city, a profession, or an employer; the name plus "reviews," "lawsuit," "fraud," or "scam." For a business it is the brand name with the same modifiers. Those queries are low volume relative to commercial head terms, high intent, because whoever runs them is looking for the plaintiff specifically, and stable in intent over time, which makes before-and-after comparison more tractable than in a commercial traffic case.

The low-volume property creates a trap. Google Search Console (Google's free reporting console for a site whose ownership has been verified, reporting the queries, impressions, clicks, and average positions Google recorded for it) suppresses rare queries: "some queries are omitted from the report to protect user privacy." A plaintiff who owns a website may open Search Console expecting the name queries at issue and find nothing - not because they did not occur, but because they were anonymized.

Search Console also reports only on a property you control. In most defamation matters the page sits on someone else's domain, so its impressions, clicks, and positions are held by a non-party - reachable through third-party discovery, not through tooling.

Prominence is a property of a layout, not a number

"Position three" meant something fixed when a results page was ten organic results (results returned from the index without payment, as distinct from advertisements). It does not now, and an opinion reporting a position without describing the page it sat on describes a layout that may not have existed.

A knowledge panel - the boxed summary about a person or business that appears alongside results - is a distinct problem from a defamatory web page. Google states that panels "are automatically generated, and information that appears in a knowledge panel comes from various sources across the web," and are "updated automatically as information changes on the web." A claimed entity can suggest edits, but changes go for review. The remedy is a suggestion, not an edit, and the source is aggregation rather than one document, so the ordinary path of suing the publisher does not map onto it.

An AI Overview - the generated summary that sometimes sits at the top of the page - is shown, Google says, only where its systems determine it is "additive to classic Search, and as such, often don't trigger." Its presence on the day of your capture is not evidence of its presence on the day of publication. Google warns that "AI responses may include mistakes," and folds traffic from these surfaces into overall Search Console figures.

Search Console does group performance by search appearance. Note the asymmetry: a plaintiff can show what its own site lost by appearance type, and almost never what the defamatory result gained.

Reach can be bounded, and it usually cannot be calculated

The tempting move is to take the result's position, apply a published click-through rate curve, multiply by an estimated search volume, and produce a count of people who saw it. I would attack that opinion, and I expect it to be attacked when I offer the analysis.

The curves in circulation are vendor aggregates, built across industries, query types, and page layouts that have nothing to do with the queries in your matter. Their methodologies are disclosed in summary, they are not audited, and they move substantially year to year. Using one as the sole basis for a reach opinion runs straight into Rule 702(b), which asks whether the testimony rests on sufficient facts or data. A generic curve is not data about this case.

Click behavior is not even stable over time. Pew Research Center analyzed the browsing of 900 U.S. adults during March 2025, covering 68,879 Google searches, 12,593 of which produced an AI summary. Users who saw a summary clicked a traditional result in 8% of visits; those who did not, 15%.

The defensible alternative, where it exists, is the party's own Search Console click-through rate for the specific queries at the specific positions, because that is the actual audience and it is the search engine's own record. Where the result sits on a third party's site, that data is not yours, and reach can be bounded - by impression counts where obtainable, and by the presence or absence of the result across dated captures - but not computed. State the range. Do not manufacture the number.

First publication, modification, and copies

The single publication rule applies to the internet. The New York Court of Appeals so held in Firth v. State of New York, decided 2 July 2002, reasoning that communications on public websites "resemble those contained in traditional mass media, only on a far grander scale." The limitations period ran from the date the report "was first made available on the Internet," not from each later access, and "the mere addition of unrelated information to a Web site cannot be equated with the repetition of defamatory matter." The opinion is in Cornell's New York Court of Appeals collection.

That doctrine converts into technical questions, which are the questions an expert answers.

• When was the page first published? Answerable from the content management system record, the lastmod value in an XML sitemap, the first Internet Archive capture, and - where the party controls the property - Search Console's URL Inspection discovery data.
• Was the content modified, and was the change related to the defamatory matter? Answerable by diffing archived captures against revision history.
• Does moving the content to a new URL republish it? A migration, a 301 redirect, or a syndication deal creates a new address serving the same words. The rule text points toward a distinction between the same content at a new address and the content repeated to a new audience. I located no controlling decision on the point and would not assert one.

Copies multiply all of this. Each has its own first-publication date, its own indexation history, and its own removal path. Google's spam policies define scraping as "taking content from other sites via automated means and republishing without original value or attribution." Locating copies is exact-phrase work against distinctive strings, and it should happen early, because mirrors outlive originals.

Where these opinions fail

Five failure modes account for most of what I see, and every one is avoidable at the collection stage.

1. The undated capture. A screenshot with no timestamp, location, signed-in state, or device information is an assertion about the witness's memory dressed up as an exhibit.
2. The single observation offered as a pattern. One capture, characterized as "the result ranked first," conflates an observation with a state of the world. A series across dates and locations makes the point and survives.
3. Third-party rank data as the sole basis. Commercial rank trackers estimate position from their own infrastructure, in their own locations, on their own schedules. As corroboration they are useful. As the sole basis for an opinion about what a plaintiff's audience saw, they run into Rule 702(b).
4. Reach computed from a curve. The most common overstatement here, and now an admissibility problem rather than a weight problem, because amended Rule 702(d) asks whether the opinion reflects a reliable application of the method to the facts.
5. Silence about what was not checked. If mirrors were not searched, if one country parameter was used, if the archive was not consulted, say so; an opinion that discloses its own boundaries is harder to impeach.

Where results were captured properly, at intervals, with the environment recorded, the series is one of the few genuinely quantitative artifacts in reputation litigation: what existed, when it appeared, when it moved, and when it stopped.

Frequently Asked Questions

Published 2026-08-11