How the AdPeekly dataset is built

As of September 20, 2026, the library holds 253,610 published LinkedIn ads from 649 curated B2B companies, collapsing to 95,893 distinct creatives after deduplication. In prose across the site we round these to 250,000+ ads and 95,000+ distinct creatives, because the pipeline adds ads continuously and exact figures drift daily. Observation began on June 10, 2026.

Two different counts answer two different questions. The published count (253,610) is every ad that passed review: it measures library size. The distinct-creative count (95,893) is what search returns: one row per creative template, duplicates collapsed. Any number a reader could sit down and verify uses the deduplicated count.

A collection pipeline monitors the public LinkedIn Ad Library around the clock and revisits every company on the curated list on a rolling cycle. Each visit records the ads currently running, captures the creative, and timestamps the observation. Ads enter the public library only after passing automated validation (creative present, minimum dimensions, required fields) and review.

The company list is curated by hand: B2B companies selected for advertising activity, with a bias toward SaaS and venture-backed firms. Companies are not added by advertisers and cannot pay to be included or excluded.

Advertisers routinely run one creative in many near-identical variants. Counting each variant as a separate ad inflates every statistic, so AdPeekly groups visually matching creatives into template groups using perceptual hashing, and elects one representative per group. Searches and gallery counts use representatives only; the variant count itself is kept as a metric, because how many variants a company tests is a signal in its own right.

For every ad we record the first date we saw it and the last date we saw it running. Observed duration is the span between the two. It is observed, not reported: LinkedIn does not publish campaign schedules, so this is a lower bound. An ad already running when observation started on June 10, 2026 has an unknown earlier start, and an ad still running today has an open end. Duration statistics on this site state these bounds and the sample size next to every figure.

Company-side facets, researched and maintained per company: industry, subcategory, company size, employee range, revenue range, funding stage, company age, country, city, region cluster, and buyer ICP (who the company sells to).

Ad-side facets, recorded per ad: creative type and format (single image, video, document, message, carousel, text, event), call-to-action text, landing page domain, first and last seen dates, observed duration, variant count, and targeting attributes (countries, locations, languages) where LinkedIn exposes them.

LinkedIn only: the library contains zero Meta, Google, or TikTok ads, and no copy on this site should imply otherwise. Curated, not exhaustive: 649 companies is a deliberate editorial sample of active B2B advertisers, not a census of LinkedIn advertising. The selection skews toward SaaS, English-language creatives, and venture-backed companies, so aggregates describe this population, not all of LinkedIn. Durations are bounded by the observation window described above. Message ads are recorded but under-observed relative to feed formats, because the Ad Library surfaces them differently.

September 20, 2026: first public edition of this methodology. Corpus verified at 253,610 published ads, 95,893 distinct creatives, 649 companies. Material changes to collection, deduplication, or facet definitions will be dated here.

Something unclear, or a number that does not add up? Tell us and we will either fix it or explain it here.