Paid Search Advertising

What the account actually contains, and how much of it belongs to Google

Paid search — buying ad placement against a user's query — is billed on clicks, which is why it is called PPC, pay per click. An attorney handed login credentials to a Google Ads account will see campaigns, keywords, spend, clicks, impressions and conversions, and will reasonably assume that is the record. It is a view onto it. Almost none of the underlying data is stored anywhere the advertiser controls.

That distinction decides how a case is worked. The advertiser can export what the interface will render, at the granularity it offers, for as long as the platform keeps it. It cannot suspend deletion, see the auction from the other side, or obtain the platform's determinations about its traffic. What it can do is copy the data out early, in native format, with the query, date range, time zone, currency and extraction date recorded alongside. Two honest extracts of the same period lacking those annotations will not reconcile, and reconciling them will consume a deposition. Retention is uneven rather than uniformly short: Google publishes no retention period for aggregate campaign and keyword statistics, and the documented limits sit on the specific resources carrying the forensic detail.

Change history is the record that answers who did what, and when

In a dispute between an advertiser and the agency or contractor that ran the account, the single most useful artifact is the change history. Google states that it "allows you to revisit the changes you've made to your campaign over the last 2 years," and it records changes to ads, assets, audiences, budgets, bid adjustments, conversion settings, targeting, keywords and campaign status — including changes made through automated rules, the API and Google Ads Editor — attributed by user.

Two limits on that record matter. First, the two-year window is an interface window. The API's change_event resource returns only the last 30 days; Google's own API support channel says so, with no workaround for older changes. The older material is therefore captured by the interface's own download or by screen capture, a weaker artifact that should be collected while a witness can still authenticate the collection. Second, Google states that change history does not track password changes, so account-access questions have to be answered from somewhere else.

What the change history supports is a chronology: a budget cut on a date, conversion tracking reconfigured, a campaign paused, targeting widened, and which login did it. What it does not supply is intent or authority. A change attributed to a user is not evidence the client approved it, and the approval — or its absence — lives in email, in the scope of work, or nowhere. Google's description of the account history is the right exhibit for a witness who claims the record shows more or less than it does.

The search terms report has been rewritten twice, and the gap is structural

The search terms report lists the queries that triggered an advertiser's ads. It is central to competitor-bidding matters, to disputes about wasted budget, and to any question about what an account was actually buying rather than what it thought it was buying.

It is also not a complete list, and has not been since September 2020. Google's current language is that the report "is a list of search terms that a significant number of people have used, and that resulted in your ad being shown," and that "some search terms that don't have enough query activity are omitted from the search terms report" on privacy grounds. In September 2021 Google restored coverage for queries from 1 February 2021 forward, and stated that data collected before 1 September 2020 which did not meet the current thresholds would remain available only until 1 February 2022.

The consequence is a structural gap between 1 September 2020 and 1 February 2021 where sub-threshold terms were never restored, and a deletion of pre-September-2020 sub-threshold terms in February 2022. An expert who charts search-term coverage across those boundaries is charting Google's reporting policy, not the advertiser's or a competitor's behavior. Google's help page on the report states the threshold rule in its own words.

Click-level data expires in ninety days, usually before counsel is retained

The only Google-side resource that exposes individual clicks is click_view, keyed on the GCLID — the identifier Google appends to a click's landing-page URL. Google's Ads API support channel has quoted the documentation's own restriction: queries including ClickView must be limited to a single day and can be requested for dates back to 90 days before the request, with no workaround for older data.

That is the most time-critical item in this discipline. A dispute surfacing in month five has no click-level record from Google at all, only aggregate counts by campaign. Anything more granular then has to come from the advertiser's own web server logs — timestamp, source IP, user agent, referrer, and the GCLID carried on the URL — or from a tag manager, or from a third-party click-quality vendor's capture. Those are the advertiser's or the vendor's evidence, not the platform's, and they are authenticated differently.

The landing-page log is worth naming specifically, because it is the one dataset in paid search the advertiser genuinely controls and because it joins to the platform's click export on the click identifier. Divergence between the clicks the platform reports and the sessions the server logged for the same GCLID set is a real, measurable quantity.

Filtered clicks, credits, and what the invalid-activity record does not separate

Google defines invalid clicks as clicks "that aren't the result of genuine user interest, including intentionally fraudulent traffic and accidental or duplicate clicks," and states that advertisers are not charged for them. Detection runs in two stages: real-time filtering before billing, and post-invoice detection producing what Google calls invalid activity credits. The documented remedy is explicit — "Clicks determined to be invalid will result in adjustments or credits, not a refund." That distinction is load-bearing for a damages theory, and sharper still for an advertiser that has left the platform holding a credit it cannot realize.

The advertiser sees two things. The "Invalid clicks" column covers traffic received over roughly the last 60 days, with no per-click detail and no reason codes. The Invalid Activity Credit Report in Report Editor is more useful: credited clicks, credited interactions, credited amount and adjusted performance metrics, by campaign and by network, for Search and Performance Max campaigns.

Read the second one carefully before it goes in an exhibit. Google states that the credits reflect invalid traffic or ad interactions later determined to be on inventory violating an AdSense program policy, and the report does not separate the two. A credited click is therefore not necessarily a click Google classified as fraudulent, and an expert presenting credited amounts as a fraud measurement has taken on an argument that Google's own documentation defeats.

Conversions in the same column are not all the same kind of number

Where a case turns on performance rather than conduct, the number in dispute is usually conversions. Google states that modeled conversions "use data that doesn't identify individual users to estimate conversions that Google is unable to observe directly," that "in the 'Conversions' column, Google reports both modeled and observed conversions," and — the sentence that matters most under cross-examination — that its modeling determines whether an ad interaction led to the online conversion but "doesn't determine whether or not a conversion happened."

Read precisely, that is a narrower claim than critics usually attribute to Google: the model assigns attribution, not existence. It is also a claim no outside expert can validate, because the model is not published and its inputs are not available for inspection. Google lists cookie limitations, consent requirements in the European Economic Area, iOS tracking restrictions and cross-device journeys as the conditions under which modeling is applied.

The honest position for an expert is that a conversion count from this column is a mixture of counted and estimated events in an undisclosed proportion, produced by the party whose delivery is being measured. It may still be the best available evidence. It is not a measurement, and a damages model that treats it as one has an exposed flank. Google's page on modeled conversions is short, and usually more efficient to put in front of an opposing expert than to argue about.

Microsoft Advertising keeps a different set of books

Cross-platform comparisons are common in agency-performance matters and are where careless work shows. Microsoft Advertising classifies clicks into three tiers rather than two: standard-quality clicks, "most likely from potential customers"; low-quality clicks, "clicks that typically don't result in conversions or value"; and invalid clicks, "clicks that often originate from spiders, robots, questionable sources, or test servers." Microsoft states advertisers are billed only for standard-quality clicks, and that bills are adjusted where clicks are later determined low-quality or invalid.

Two differences follow. Microsoft exposes a "Low-quality clicks" metric and a "Low-quality click rate" percentage — not a column labeled invalid clicks — so a side-by-side table of "invalid clicks" across the two platforms compares different definitions. And Microsoft documents a refund path Google does not: where an account is deactivated holding a refund-eligible balance arising from invalid-click credits, it states it will attempt to refund the primary payment method automatically.

Neither platform publishes its detection logic or documents an appeal route for an advertiser disputing a determination. In both cases the platform is the sole judge of traffic on which it also earns revenue — a structural point serious people argue about on both sides, and one an expert should describe rather than adjudicate.

What a paid search analysis does not settle

Advertiser-side data can show that a click was anomalous. Anomalous is not fraudulent. Every signal available from the account and the server log — a repeated IP, a short dwell time, no conversion, an odd hour, a data-center address range — has innocent explanations: shared corporate network address translation, carrier CGNAT, VPN and private-relay traffic, accidental taps on mobile creative, and shoppers who left because the offer was wrong for them.

Attribution to a person is harder still and is usually not solvable from advertiser data at all. An IP address is not a person. Traffic resolving to a competitor's netblock establishes that packets came from that network — not who was at the keyboard, not whether the machine was compromised, and not whether an employee was doing ordinary competitive research, which is not misconduct. Establishing that a named party directed invalid clicks generally requires payment records, communications and third-party logs available only in discovery.

Two further limits belong in any report. The "Invalid clicks" figure is the platform's count of what its own systems filtered; it carries no denominator for what was missed, so a low number is equally consistent with clean traffic and with undetected sophisticated traffic. And an industry-wide loss estimate says nothing about whether a particular account lost a particular sum — aggregate prevalence is not individual causation. Where a paid search claim fails, the reason is very often that the ninety-day and sixty-day windows closed before anyone looked — a preservation failure rather than a merits finding, and worth saying plainly in either direction.

Frequently Asked Questions

Published 2026-08-12