What 35,031 B2B Software Ads Reveal About Ad Formats

Between 28 and 30 July 2026 we collected every ad we could reach from the public ad libraries of 267 B2B software companies. The result is 35,031 individual ad records.

Each record carries the advertiser, the platform, the placement, the format, the ad text, and the capture timestamp. Nothing here is modelled or estimated. Every number below is a count of records we hold.

The companies span 33 categories. Security is the largest at 12,550 ads, followed by general B2B SaaS at 2,757, observability at 2,728, and DevSecOps at 2,138. These are engineering-heavy categories, which matters when you read the results.

The headline finding is how ordinary the winning format is.

Nearly three quarters of what B2B software companies actually run is a single still image with text on it. Video, the format most marketing advice tells you to prioritise, is one ad in eight.

This is not a claim about what performs best. We cannot see performance. It is a claim about what companies with real budgets are choosing to keep in market, which is a different and in some ways more honest signal.

A single number can hide a lopsided distribution, so we split the format mix by category. It barely moves.

Every category lands between 70 and 81 percent static. Observability is the most image-heavy at 81.1 percent and the least video-heavy at 6.3 percent. No category comes close to a video-first mix.

If you are building a first campaign and wondering whether you need video before you can launch, the companies already spending money in your category have answered that question.

We downloaded 279 of the video ads as real files and probed each one with ffprobe for duration, resolution, frame rate and audio. This is a smaller and narrower sample than the format counts above, and we say so plainly in the limits section, but it is the only frame-level look at B2B video ads we know of.

Of those 279 files, 211 are motion graphics and 68 are filmed footage. Motion graphics here means flat vector colour, clean edges, screen recordings and animated type rather than a camera pointed at a person.

On shape, 208 of 279 are vertical 9:16, 27 are square, 25 are 4:5 portrait and 17 are 16:9 landscape.

On length, the median is 28 seconds. 156 files run 30 seconds or under, 96 run between 31 and 90 seconds, and 27 run longer than 90 seconds. One outlier runs over 30 minutes, which is almost certainly a webinar recording placed as an ad rather than a designed video ad.

The practical read is that a B2B video ad is usually an animated explainer of about half a minute, built in software rather than filmed.

Every number above is a real count. That does not make every number a market truth, and the limits matter more than the headline.

The channel mix is a collection artefact, not a market share. 34,418 of the 35,031 records are LinkedIn, 588 are Meta and 25 are Google. That is 98.3 percent LinkedIn, and it is not evidence that B2B software runs 98 percent of its advertising on LinkedIn. LinkedIn publishes a more accessible ad library than the others. Read this as a LinkedIn dataset with a small Meta comparison attached.

Per-company counts are capped by collection, so do not compute a market average from them. 86 companies stop at exactly 192 ads and 41 stop at exactly 120. Those are pagination boundaries in our collection, not advertiser behaviour. The gap is large: we hold 197 ads for Wiz where the live library reports 55,791, and 255 for Datadog against 8,227. Any median ads per advertiser calculated from our capture would measure our scraper.

The video sample is narrow. The 279 probed files come from 19 companies and represent 6.5 percent of the motion ads in the dataset. Several of the largest motion advertisers we captured are not in that sample at all. Treat the 76 percent motion graphics figure as a description of those 279 files, not of B2B video advertising.

The categories skew engineering. Security alone is over a third of the dataset. A marketing tool or an HR product may behave differently.

It is a snapshot. Everything was captured across three days in July 2026. Ad libraries show what is live at a moment, not what ran all year.

Three things follow from the numbers rather than from opinion.

Start with static. It is what three quarters of your category is running, it is the fastest thing to produce, and it lets you test the argument before you spend on production. If the message does not work as an image, animating it will not save it.

When you do make video, make an animated explainer of roughly 30 seconds rather than a filmed piece. That is what the files we examined actually are, and it is the cheaper thing to iterate.

Build vertical first if you run on Meta. 208 of 279 files were 9:16. On LinkedIn the safe default is still 1:1 or 4:5 for feed placement.

None of this tells you what to say. Format is the cheapest decision in a campaign and the one people spend the most time on. The argument you make and the audience you make it to still matter more.