On the one open dataset that publishes both formats with the same method, static images beat video on cost per lead: $184 against $238. The direction holds independently on all three social platforms in the data.

That is not what the industry says, and it is worth being careful about what it does and does not establish. It is observational, not experimental. It is one panel. And it is still better evidence than almost everything else published on this question.

This page gives the figures, the caveats they need, and what to do with them.

What the dataset is

Before the numbers, the provenance, because it is the only reason to take them seriously.

The scope. A B2B advertising benchmark covering 153 advertisers and $57.6M of 2025 ad spend, released as a downloadable file under a Creative Commons attribution licence.

The suppression rules, which matter. No cut is published with fewer than five advertisers, under $50,000 of spend, or with any single advertiser above half of it. Suppressed cuts are absent from the file rather than shown as zero, which prevents the usual trick of a category that looks empty when it was simply too thin to report.

What it is not. An experiment. Advertisers chose which format to run for which purpose, so any difference between formats includes the difference between the jobs they were given. This is the central caveat and everything below should be read through it.

What makes it unusual anyway. Most format comparisons come from companies selling one of the formats, with no sample, no period and no suppression rule. This one publishes all three and lets you check the arithmetic yourself.

The limit to state alongside every figure. The panel skews toward B2B technology and does not disclose its geography. It describes a population of advertisers, not a market, and certainly not your account.

Static image and video advertising compared on a single business-to-business panelTable comparing static image creative with video creative on a single business-to-business advertising panel covering one hundred and fifty-three advertisers and fifty-seven point six million dollars of two thousand and twenty-five advertising spend, released under an open licence with declared suppression rules. Static image creative, measured on one hundred and forty advertisers, produced a cost per lead of one hundred and eighty-four dollars, a cost per click of five dollars twenty-seven, a cost per thousand impressions of thirty-seven dollars sixty-eight, a click-through rate of zero point seven one percent and a click-to-lead rate of four point four percent. Video creative, measured on eighty-nine advertisers, produced a cost per lead of two hundred and thirty-eight dollars, a cost per click of seven dollars sixty-nine, a cost per thousand impressions of thirty-seven dollars twenty-two, a click-through rate of zero point four eight percent and a click-to-lead rate of three point one percent. The cost per thousand impressions figures are almost identical between the two formats, which removes inventory price as an explanation and shows that video underperforms after the impression rather than before it: fewer people click, and fewer of those who click become leads, the two effects compounding into the cost per lead gap. The dataset is observational rather than experimental, since advertisers chose which format to use for which purpose, so any difference between formats includes the difference between the jobs those formats were given.Static and video, same panel, same year153 advertisers, $57.6M of 2025 spend, identical definitions.METRICIMAGE (n=140)VIDEO (n=89)Cost per lead$184$238CPM$37.68$37.22Click-through rate0.71%0.48%Click-to-lead rate4.4%3.1%Cost per click$5.27$7.69Almost identical CPM. Video does not lose on inventory price.Fewer clicks per impression, fewer leads per click. The two compound.
The CPM is almost identical, so video is not losing on inventory price. It loses after the impression. Source : Metadata B2B Ad Spend Benchmark (2026)

The headline comparison

Two rows, same panel, same year, same definitions.

Static image, on 140 advertisers: cost per lead $184, cost per click $5.27, CPM $37.68, click-through rate 0.71%, click-to-lead 4.4%.

Video, on 89 advertisers: cost per lead $238, cost per click $7.69, CPM $37.22, click-through rate 0.48%, click-to-lead 3.1%.

Read the CPM row first, because it removes one explanation. The two formats cost almost exactly the same per thousand impressions, $37.68 against $37.22. Video is not losing because the inventory is more expensive. It is losing after the impression.

Where it actually loses. Fewer people click, 0.48% against 0.71%, and fewer of those who click become leads, 3.1% against 4.4%. The two effects compound into the cost per lead gap.

What that pattern suggests, carefully. Video in this dataset attracts less action per impression and less conversion per click than a still image. Whether that is a property of video or of how these particular advertisers used video is exactly what an observational dataset cannot separate.

One honest alternative explanation. Video may be carrying awareness work that does not end in a form, and being measured here on a job it was not doing. That is plausible, and it is also the argument that lets any underperforming format off the hook indefinitely, so it should be tested rather than assumed.

The same result on three platforms independently

A single aggregate can hide a composition effect. This one does not, and that is what makes it interesting.

On the professional network, image at $200 per lead against video at $265, on 127 and 81 advertisers.

On the first social platform, image at $136 against video at $225, on 64 and 43 advertisers.

On the second social platform, image at $131 against video at $201, on 45 and 31 advertisers.

Why three independent replications matter. If the aggregate gap came from a mix effect, for example video being concentrated on an expensive platform, splitting by platform would dissolve it. It does not. The same direction appears three times, on three different auction environments and three different audiences.

The click-through rates tell the same story each time. 0.66% against 0.47%, 0.83% against 0.52%, 0.67% against 0.39%. Static creative earned more clicks per impression on every platform in the data.

What this does not rule out. A common cause across all three: for instance, that the advertisers producing video in this panel were systematically doing something else differently. Observational data cannot exclude it, and no honest reading pretends otherwise.

Cost per lead by creative format across three advertising platformsTable presenting cost per lead by creative format across the three social advertising platforms represented in a business-to-business benchmark dataset, showing that static image creative outperformed video on each platform independently. On the professional network, image creative produced a cost per lead of two hundred dollars across one hundred and twenty-seven advertisers against two hundred and sixty-five dollars for video across eighty-one advertisers. On the first social platform, image produced one hundred and thirty-six dollars across sixty-four advertisers against two hundred and twenty-five dollars for video across forty-three. On the second social platform, image produced one hundred and thirty-one dollars across forty-five advertisers against two hundred and one dollars for video across thirty-one. Click-through rates followed the same pattern on every platform, zero point six six percent against zero point four seven, zero point eight three against zero point five two, and zero point six seven against zero point three nine. Because splitting the aggregate by platform does not dissolve the gap, the difference cannot be attributed to a composition effect such as video being concentrated on a more expensive platform. The cheapest format in the entire dataset is neither video nor a conventional image but the document carousel on the professional network, at one hundred and forty-two dollars per lead across fifty-one advertisers, with a click-through rate of one point two three percent and a click-to-lead rate of eleven point nine percent against a panel average of four point three percent.Cost per lead by format, platform by platformPLATFORMIMAGEVIDEOGAPProfessional network$200$265n = 127 / 81Social platform one$136$225n = 64 / 43Social platform two$131$201n = 45 / 31And the cheapest lead in the file is neitherDocument carousel: $142 per lead, 1.23% click-through, 11.9% click-to-leadPanel average click-to-lead is 4.3%. This format qualifies before the click, not after it.Three independent replications. A composition effect would have dissolved the gap.
Splitting by platform does not dissolve the gap. And the cheapest lead in the file comes from a static, unglamorous format. Source : Metadata B2B Ad Spend Benchmark (2026)

The format nobody talks about wins

The most interesting row in the file is neither video nor image.

The document carousel, on 51 advertisers: cost per lead $142, cost per click $5.87, click-through rate 1.23%, click-to-lead 11.9%.

Put the click-to-lead figure in context. The panel average across all formats is 4.3%. This format converts clicks to leads at 11.9%, roughly two and a half times that, and it is the cheapest lead in the dataset.

Why the mechanism is plausible rather than mysterious. A document ad is a multi-page carousel that people read inside the feed. By the time somebody has swiped through several pages and then clicks, they have self-selected far harder than someone who clicked a headline. The format does its qualification before the click rather than after it.

Why it is under-used anyway. It is unglamorous, it exists on one platform, it does not showcase a production budget, and no agency has ever won an award for one. Those are not performance arguments.

The caveat this row needs as much as the others. Fifty-one advertisers, self-selected into a format that suits explanatory content. A company whose proposition cannot be explained in six slides will not replicate this.

What to take from it practically. Before commissioning a video, find out whether your proposition survives being explained in six still frames. If it does, you have a cheaper test available, and on this data a better-performing one.

Why the document advertising format converts clicks to leads at a higher rateDiagram explaining why the document carousel advertising format converts clicks into leads at a substantially higher rate than other formats on the benchmark panel examined. A document advertisement is a multi-page carousel that a reader swipes through inside the feed without leaving the platform, which means that by the time somebody has read several pages and then clicks through, they have already self-selected far more strongly than somebody who clicked on a headline after a single impression. The format therefore performs its qualification before the click rather than after it, which is the mechanism behind its measured results on the panel: a cost per lead of one hundred and forty-two dollars across fifty-one advertisers, a click-through rate of one point two three percent and a click-to-lead rate of eleven point nine percent, against a panel average click-to-lead rate of four point three percent, making it roughly two and a half times the average on that measure and the cheapest lead in the dataset. The format is nonetheless under-used, for reasons that are not performance arguments: it is unglamorous, it exists on one platform only, it does not showcase a production budget, and no agency has won an award for one. The caveat it requires is the same as every other row, namely that fifty-one advertisers self-selected into a format suited to explanatory content, so a company whose proposition cannot be explained in six slides will not replicate the result.Qualification before the clickHeadline adOne impression, one click→Qualification happens afterOn your landing page, or by phone→Panel average click-to-lead4.3%Document adSix slides, swiped in-feed→Qualification happens hereBefore anybody clicks anything→Document click-to-lead11.9%Cheapest lead in the file, at $142, on a format nobody wins awards for.Before commissioning a video, find out whether six still frames explain your proposition.Caveat: 51 advertisers, self-selected into a format that suits explanatory content.
Somebody who swiped through six slides and then clicked has self-selected harder than somebody who clicked a headline. Source : MASTRATOS from Metadata benchmark (2026)

The row you must never put in the same table

One line in the file looks spectacular and means nothing, and it is a useful lesson in reading benchmark data.

The figures. Conversation ads report a click-through rate of 42.36%, a cost per click of $325.22 and a CPM of $137,773.

What is happening. Those are not comparable denominators. A conversation ad is a messaging format, so an impression, a click and a conversion count entirely different events from a feed ad. The metric names are the same and the objects behind them are not.

Why this matters beyond one row. Every benchmark table you will ever be shown contains at least one row like this, and it will usually be the one somebody wants to build a recommendation on. A 42% click-through rate should not impress you. It should make you ask what a click is.

The check that catches it. Look for figures that are implausible by an order of magnitude rather than by a margin. A CPM in six figures is not a good result, it is a different unit.

And the general rule. Never compare two rows of a benchmark without asking whether they count the same events. That question disqualifies more comparisons than any statistical test.

Why one row of a benchmark table cannot be compared with the othersDiagram warning about a row in a benchmark dataset whose figures appear spectacular but are uninterpretable because its denominators count different events. Conversation advertisements in the dataset report a click-through rate of forty-two point three six percent, a cost per click of three hundred and twenty-five dollars twenty-two and a cost per thousand impressions of one hundred and thirty-seven thousand seven hundred and seventy-three dollars. These are not comparable denominators, because a conversation advertisement is a messaging format in which an impression, a click and a conversion count entirely different events from those counted by a feed advertisement, so the metric names are identical while the objects behind them are not. This matters beyond a single row, because every benchmark table contains at least one line of this kind and it is usually the line somebody wants to build a recommendation on. A click-through rate of forty-two percent should not impress a reader, it should prompt the question of what a click is in that context. The check that catches this failure is to look for figures implausible by an order of magnitude rather than by a margin, a cost per thousand impressions in six figures being a different unit rather than a good result. The general rule is never to compare two rows of a benchmark without first asking whether they count the same events, a question that disqualifies more comparisons than any statistical test.The row that means nothingMETRICFEED FORMATSCONVERSATION ADSClick-through rate0.5% to 1.2%42.36%Cost per click$4 to $9$325.22CPM$26 to $72$137,773A six-figure CPM is not a good result. It is a different unit.An impression, a click and a conversion count different events in a messaging format.Never compare two benchmark rows without asking whether they count the same events.
Every benchmark table contains a row like this, and it is usually the one somebody wants to build a recommendation on. Source : MASTRATOS from Metadata benchmark (2026)

Why the video-first default took hold anyway

The gap between what this data shows and what the industry recommends is worth explaining, because the explanation is not that everyone is wrong.

Platforms sell video inventory. Video carries higher engagement signals, occupies more feed real estate and supports formats that command higher prices. Recommendations from a party that benefits from the recommendation are not automatically wrong and they are not neutral either.

Agencies price video higher. A video costs more to produce than a still image, and production is a revenue line. Again, not dishonest and not neutral.

Video is measurable in ways that flatter it. Video-specific metrics such as view counts and completion rates produce impressive-looking numbers that no static format can generate, because static formats have nothing equivalent to report. A dashboard full of video metrics looks like performance.

And consumer marketing genuinely favours it. Most of the evidence people cite for video comes from consumer categories with mass audiences and emotional propositions. B2B propositions are frequently explanatory, and explanation reads better than it watches.

What that combination produces. A default nobody chose, defended with evidence from a different context, sold by parties with an interest in it, and measured with instruments that only one side of the comparison can use.

The cheap correction. Not to reject video, but to stop treating it as the starting point. Make the still version first, because it is faster, and let the video earn its budget against it.

What this does and does not license you to conclude

The honest reading is narrower than the headline and more useful than it.

What it supports. That video is not automatically the better format in B2B, that the default assumption in most creative briefs is unevidenced, and that a static test is worth running before a video budget is committed.

What it does not support. That video does not work. The data cannot separate the format from the job it was given, and video may be doing work that never ends in a form submission.

The selection problem, stated plainly. Advertisers who produce video are not a random subset. They tend to have larger budgets, agency involvement and brand objectives, all of which change the outcome independently of the format.

Why the result is still worth acting on. Because the cost of testing the alternative is close to zero. A static version of your current message takes an afternoon. The video it would replace takes a production cycle and a budget line.

The asymmetry that decides it. If static works as well, you have saved a production budget. If it does not, you have lost an afternoon. Very few tests in advertising have that payoff structure.

What to test, and how to judge it

Four steps, and none requires new budget.

One static against your current video. Same audience, same offer, same landing page, running concurrently rather than sequentially so seasonality cannot explain the difference.

Judge on cost per meeting held, not cost per lead. The whole argument for video is usually that it produces better-quality attention. If that is true, it will appear in the meeting rate, and if it does not appear there it was not true.

Give it enough volume to conclude, or admit you cannot. Detecting a realistic difference between two creative variants takes thousands of impressions per variant. On a small account, run the test for a quarter or accept that you are comparing noise.

Test the document format if your platform offers it. On this data it is the cheapest lead available and it is under-used. Six slides explaining one problem is a lower production commitment than any video.

And keep the log. Date every creative change. Without it, the next person to look at this account will attribute the difference to the season, the algorithm or the audience, and the test you ran will have taught nobody anything.

Design and payoff structure of a static against video creative testDiagram setting out how to test static creative against video creative on an advertising account, and why the payoff structure of that test is unusually favourable. The design requires one static creative run against the current video, on the same audience, the same offer and the same landing page, running concurrently rather than sequentially so that seasonality cannot explain any difference observed. The judgement should be made on cost per meeting held rather than cost per lead, because the usual argument for video is that it produces better quality attention, which if true will appear in the meeting rate and if absent from the meeting rate was not true. The test needs enough volume to conclude, since detecting a realistic difference between two creative variants requires thousands of impressions per variant, so on a small account the test should run for a quarter or the advertiser should accept that they are comparing noise. Where the platform offers it, the document format should be tested as well, since on the panel examined it produced the cheapest lead available and remains under-used, six slides explaining one problem being a lower production commitment than any video. Every creative change should be dated in a log, since without one the next person to examine the account will attribute the difference to the season, the algorithm or the audience, and the test will have taught nobody anything. The payoff structure is asymmetric: if the static creative performs as well, a production budget is saved, whereas if it does not, an afternoon is lost.The test, and why you should run itOne static against your current videoSame audience, same offer, same page, running concurrently rather than one after the other.Judge on cost per meeting heldIf video produces better attention, it shows there. If it does not show there, it was not true.If static wins or tiesYou saved a production budget, permanently.If video winsYou lost an afternoon, and you now know.Very few tests in advertising have that payoff structure. Date every creative change.
If static works as well, you saved a production budget. If it does not, you lost an afternoon. Source : MASTRATOS (2026)

Where to go next

You want the buying mechanics behind all this. Media buying explained.

Your creative has been running a long time. Creative fatigue metrics.

You are choosing between platforms. LinkedIn Ads vs Google Ads, and Meta Ads or Google Ads.

Your cost per lead is the number in dispute. Cost per lead.

You want to know what an impression costs. CPM, cost per thousand impressions.

Your click-through rate is moving and you do not know if it is real. Click-through rate benchmarks.

In short

  • On one open panel of 153 B2B advertisers, static images produced leads at $184 against $238 for video.
  • The CPM was almost identical at $37.68 and $37.22, so video does not lose on inventory price. It loses after the impression.
  • Fewer clicks and fewer conversions per click: 0.71% against 0.48%, and 4.4% against 3.1%.
  • The same direction held on three platforms independently, which rules out a composition effect.
  • The cheapest format was neither: document carousels at $142 per lead, converting clicks to leads at 11.9% against a panel average of 4.3%.
  • One row in that file is uninterpretable. Conversation ads report a 42% click-through rate because their denominators count different events entirely.
  • The data is observational. Advertisers chose their formats, so the comparison includes the jobs those formats were given.
  • Test it anyway, because a static version of your message costs an afternoon and the video it might replace costs a production cycle.

Stop assuming video is the default and find out on your own account. Book a diagnostic, or see how we approach B2B paid acquisition.