Paid Social Advertising

Three records exist, and only one is usually in the client's hands

For any paid social campaign there are three distinct records, and conflating them is the most common defect in exhibits here.

The first is the advertiser's own account record — the Ads Manager or Campaign Manager export. For that advertiser it is close to complete: creative in every variant, targeting as configured, budget, delivery, spend, reported conversions, change history. It is available to the advertiser without any process at all, and obtainable from an agency by a Rule 34 request or under a contract audit right. The second is the platform's internal record: bid-level data, delivery decisions and content moderation actions the advertiser never sees. That is obtainable only by subpoena, and platforms resist. The third is the public archive, the Ad Library, which is far narrower than litigators expect.

The practical consequence runs one way in almost every matter. Counsel asking an expert witness in Facebook ads to establish what a client ran, spent, targeted and reported is asking a question the account export answers well. Asking the same about an opposing party's campaigns is asking something the public record generally does not answer at all, and the answer has to come from that party's production. Getting the export in hand — with its date of extraction recorded, because an export is a snapshot — is the first step in either posture.

What the Ad Library archives, and for whom

The Ad Library's API is the cleanest evidence of what it actually covers. The ad_type parameter takes the values ALL, POLITICAL_AND_ISSUE_ADS, FINANCIAL_PRODUCTS_AND_SERVICES_ADS, EMPLOYMENT_ADS and HOUSING_ADS. Search runs by keyword across text, images, audio and call-to-action buttons in more than 200 territories.

The critical fact is in the field documentation. The bylines, delivery_by_region and the estimated audience size fields are marked available only for political and issue ads. The spend and reach data people assume is public is, for ordinary commercial advertising, not public. Retention follows the same division. Meta's launch announcement describes the library as bringing in all active ads any Page is running, while ads about social issues, elections and politics are archived for seven years. For a commercial advertiser the Ad Library is a live transparency window, not an archive — when the ad stops running, the public record may simply be gone.

Even the seven-year archive is a rolling window that is now pruning. Beginning 24 May 2025, ads from the archive's first year started reaching the seven-year limit and being removed. Political and issue ads from 2018 and 2019 are already gone or going, and the horizon closes further every month. The operating rule in either category is the same: capture early, and treat what the library returns today as evidence of today. Meta's own API reference is the right citation for which fields exist for which ad types.

Transparency libraries elsewhere are narrower still. TikTok and LinkedIn operate libraries built to meet European obligations, and ad transparency generally exists because of political advertising rules and European regulation rather than any United States commercial disclosure duty. A library scoped to the European Economic Area is not evidence about a United States campaign.

The breakdowns expire on a tiered schedule, and the tiers keep tightening

Meta's Ads Insights API — the route by which campaign data is pulled at scale — moved to a tiered availability schedule effective 12 January 2026, announced 16 October 2025. Three tiers: 37 months for total aggregate values, 13 months for unique-count field breakdowns and hourly breakdowns, and 6 months for frequency breakdowns. The same change removed the 7-day view and 28-day view attribution windows from the API entirely.

That followed an earlier tightening. From 10 June 2025, reach is no longer returned for standard queries that apply breakdowns and use start dates more than 13 months old, and the use_unified_attribution_setting and action_report_time parameters are disregarded so that API responses mirror Ads Manager settings. A further restriction is already scheduled: from 3 August 2026, certain accounts must opt in for the impression_device, hourly_stats_aggregated_by_audience_time_zone and frequency_value breakdowns.

Two operational points follow for counsel. First, the granular splits that decide performance disputes — frequency, hour of day, device — are on the shortest clocks, so a dispute that surfaces eight months after a flight has already lost some of them. Second, Meta states that Insights refresh every fifteen minutes and do not change after 28 days of being reported. A pull taken more than 28 days after the period should be reproducible; a pull taken inside 28 days may not match a later pull of the same dates. Every extract should therefore carry the date and time it was taken, and any two extracts that disagree should be checked against that rule before anyone alleges anything. Meta's developer announcement states the tiers.

January 2021 broke comparability, and the break is still in most exhibits

On 19 January 2021 Meta told developers that it would no longer support 28-day click, 28-day view and 7-day view attribution windows. The account-level attribution setting was replaced by a per-ad-set setting created at campaign creation, defaulting to 7-day click and 1-day view — which Meta itself said "may result in a decrease in the number of reported conversions" — and, once Apple began enforcing its tracking prompt, defaulting to 7-day click alone.

A pre-2021 conversion count and a post-2021 conversion count are therefore not comparable. The measured window shrank from 28 days to 7 on the click side, and view-through was largely removed. Any year-over-year comparison spanning January 2021 is measuring a definitional change at least as much as a performance change, and an agency-performance report that draws a line across that date without saying so is impeachable on Meta's own developer blog. Historical data for the removed windows survived only in the API — and, per the January 2026 change, that residual path is now closed too.

Aggregated Event Measurement arrived at the same time, introduced to support measurement of web events once the tracking prompt applied. Three of its features matter evidentially. Advertisers were limited to eight conversion events per domain for optimization. Meta stated plainly that "statistical modeling will be used for certain attribution windows and/or metrics." And delivery and action breakdowns are not supported for offsite conversion events — which removes age, gender, region and placement splits for precisely the events a performance dispute turns on.

Instagram, and what a user-side archive is good for

Instagram campaigns are bought and reported through the same Ads Manager surface, so the advertiser-side record is the one described above, with placement identifying where delivery occurred. Where Instagram matters separately is on the consumer side, and the artifact is different: Meta's Download Your Information export, which since 17 October 2023 can pull Facebook and Instagram account data together in a single request.

It is worth being exact about what that archive is, because it is often produced into advertising matters on the assumption that it says something about a campaign. It is a user's export of a user's own account. It is generated from inside the account, which is why it is the workhorse of social media discovery — where the account holder is a party or a cooperating witness, no subpoena is needed. It is not an advertiser-side export and it contains no campaign performance data. In a dispute about what a campaign delivered, it is not responsive. In a dispute about what a particular individual was shown, targeted with, or interacted with, it can be central.

Whether an engagement calls for an expert witness in Instagram marketing to examine a produced archive or an advertiser's own export is therefore not a stylistic choice. The two answer different questions, and putting the wrong question to the wrong artifact is how reports get excluded on relevance. Where the question is click quality on a social buy, the honest starting point is to check what filtered-activity reporting the account actually exposes rather than to assume a counterpart to another platform's column exists.

The platform's numbers are the platform's claims about its own product

Meta has published corrections to its own metrics. On 16 November 2016 it disclosed, among other errors, that 7-day and 28-day organic Page reach had been miscalculated as a simple sum, with corrected 7-day reach roughly 33 percent lower and 28-day reach roughly 55 percent lower, an error live since May of that year; that time spent on Instant Articles had been over-reported by 7 to 8 percent on average since August 2015; and that video completions had been undercounted, an issue identified by a third-party auditor. A second disclosure on 9 December 2016 covered estimated reach methodology, a streaming-reactions misallocation and a Graph API count discrepancy.

A court has also permitted advertisers to litigate over what a metric meant. In DZ Reserve v. Meta Platforms, Inc., No. 22-15916, decided 21 March 2024, the Ninth Circuit affirmed certification of a damages class of advertisers alleging that Meta represented Potential Reach as an estimate of people when it was an estimate of accounts, and vacated certification of the injunctive-relief class, remanding on standing. The court decided class certification under Rule 23. It did not decide the merits, did not find fraud, and did not determine liability. The opinion is short enough to read before anyone characterizes it.

The use for that history is narrow and should be stated narrowly. It supports the proposition that platform-reported metrics have, on the platforms' own admissions, been materially wrong for extended periods, and that a court has allowed advertisers to litigate a metric's meaning as a class. It does not support an assertion that any particular figure in any particular account is wrong.

What the paid social record does not settle

The Ad Library cannot establish a commercial advertiser's spend or reach, because those fields exist only for political and issue ads. It cannot establish that an ad did not run, because non-political ads are searchable only while active — absence from the library today says nothing about last quarter. Anyone planning to make a competitor's ad spend an element of a claim should know that before the pleading is drafted, not after.

Reach is a deduplicated estimate of accounts, not a count of people, and impressions are not people either. No ad platform record identifies a natural person. An export is a snapshot at the time it was taken and cannot establish that nothing was edited or deleted beforehand, which is why the extraction date belongs in every report and why a preservation demand issued early is worth more than a broad subpoena issued late.

On the conversion side, some proportion of the reported figures is modeled rather than observed, Meta said so in its own words, and the proportion is not broken out. Delivery and action breakdowns are unavailable for offsite conversion events, so the splits that would let an expert test a conversion pattern against demographics or placement are frequently absent by design rather than withheld. And none of it reaches the question counsel most often wants answered, which is whether the advertising caused the business outcome. Every record here is observational: it establishes what was configured and what was delivered.

Frequently Asked Questions

Published 2026-08-12