A/B test ad formats systematically

A format test answers one question, such as: "Does a short video earn a lower cost per lead than a single image for our finance directors?" Write it down before you build anything. If you cannot phrase it as a question with two possible answers, you are not ready to test.

Decide which result you care about. Click-through rate is the quickest signal, but cost per lead tells you whether the clicks were worth having. Track both and let cost per lead decide when they disagree.

Keep everything else identical in the two ads: the audience, the headline and body text, the offer, the landing page, the budget and the start date. Only the format changes. If the video also has a new headline, you have learned nothing about video.

The usual formats to compare are single image, carousel, short video and text-only or document ads. Start with two. Add a third only when you have a result for the first pair. For guidance on which formats suit which objective, see Ad format selection.

Use the platform's own split-test option where it has one, in LinkedIn Ads or Meta Ads, because it divides the audience for you. If it does not, duplicate the campaign with identical targeting and give each version the same daily budget.

Check that the two audiences do not overlap, otherwise the same person sees both ads and your results mix. Set the same schedule, run both at the same time, and tag each ad with its own campaign name so reporting is clean.

Reading a test early is the most common way to fool yourself. As a rule of thumb, I would want at least 1,000 impressions and about 100 clicks per variant before looking at click-through rate, and enough leads to compare cost per lead, which usually means two to four weeks.

If your audience is small and you cannot reach those numbers, say so. Run the test for a longer period or combine two rounds, and treat the result as a direction rather than proof.

Do not pause one variant because it looks worse on day three. Do not change budgets, bids or targeting. Early numbers swing, and the platform needs time to settle. Look at the dashboard once a week, write the numbers down, and make no changes.

When the test ends, compare cost per lead and click-through rate. Call a winner only if the gap is at least 20 percent in relative terms and holds across the full period, not just one good week. If the gap is smaller, call it a tie and keep the format that is cheaper to produce.

Then check one level deeper: did the winner win in every segment, or only in one? The method for that is in Analyse click-through by audience segment.

Keep a test log with one row per test: the question, the formats, the dates, spend, impressions, clicks, leads, cost per lead, the decision and what you will test next. Add the winning ad to your swipe file.

Run one format test every two weeks at most. Between tests, apply the winner to your main campaigns. The wider principle, one variable at a time, is in Test one variable, or you learn nothing.

• Changing the format and the headline together.
• Calling a winner after two days or fifty clicks.
• Judging on click-through rate alone, then paying for cheap clicks that never become leads.
• Testing on a different audience for each variant.
• Running tests in a month that has a holiday, a product launch or a budget change.
• Forgetting to write down the result, so the same test runs again in six months.

Look at your test log after a quarter.

• You have at least four completed tests, each with a clear decision.
• Your main campaigns use the winning formats, and cost per lead has fallen compared with the start of the quarter.
• You can tell a new colleague which format works for which audience and show the numbers behind it.
• At least one test ended in "no difference", which shows you are reading results honestly instead of looking for a story.