Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

LinkedIn Marketing › LinkedIn Ads

Testing LinkedIn Ad Creative

How to test LinkedIn ad creative when audiences are small and clicks costly: organic pre-tests, rotation, controlled tests, what to vary and how to read it.
Guideintermediate12 min readUpdated

Why testing is harder on LinkedIn

The LinkedIn Ads hub explains the cost logic: audiences are narrow, the advertisers competing for them sell high-value products, and members spend less time in the feed, so each impression and each click costs more than on consumer platforms. The same facts make testing harder. A test needs enough people in each arm to tell a real difference from noise, and on a precise B2B audience that volume takes longer to arrive and costs more to buy. The temptation, which the campaign objectives and formats guide names in its creative habits, is to call a test early because waiting is expensive.

The hub also describes how delivery works inside the auction: ads are ranked on a combination of bid and a relevancy score built from predicted engagement, so creative members respond to lowers the price of reaching them. Creative is therefore a cost lever as well as a conversion lever, which makes testing worth doing. But it also means the delivery system is not neutral between ads, and any test that leaves the choice of who sees what to the system is measuring the system's preferences as much as the creative.

Two further constraints shape the programme. Small audiences tire quickly: the same members see the same ad often, Creative Fatigue arrives sooner than on a broad audience, and a test that runs too long ends up comparing a tired control with a fresh variant. And the outcome that matters, a lead sales will work, arrives in small numbers and late. The guide to reading small numbers explains why rates built on a handful of events jump around and why a range says more than a single figure; nearly every LinkedIn test is a small-numbers problem.

The Meta and TikTok guides to creative testing on Meta and creative testing on TikTok describe programmes built on volume. The method underneath is the same, a hypothesis, one change and a decision rule, as the experimentation loop sets out; what changes on LinkedIn is how sparingly each paid test has to be spent.

Which question each method answers

Most testing waste comes from using an expensive method for a cheap question, or a cheap method for a question that needs proof. Match the method to the question before anything is built.

QuestionMethodWhat it can tell youWhat it cannot
Which angles does the audience care about at all?People's own postsWhich problems and claims draw comments from the right rolesWhether the angle produces leads in a paid campaign
Which of a few ready ads should carry this campaign?Rotation inside a campaignWhich ad the delivery system and the audience favour nowWhether one ad is better than another for the same people
Does this offer or angle produce better leads than that one?Controlled comparison on split audiencesA fair difference in accepted leads between armsAnything when the arms are too small to separate
Is the creative tiring?Weekly read of frequency and click-through by adWhen to rotate in the next variationWhy it tired
Should the lead be captured on LinkedIn or on a page?A designed experimentThe difference in qualified leads between destinationsCreative questions, which it holds constant

The last row belongs to the experiments library rather than to creative testing. The native lead form test, the lead form length test and the gated content test each design one such test with its hypothesis, metrics and sample reasoning; the video ad length test and the frequency cap test do the same for two execution questions. Use those designs as they are written rather than improvising a version inside a creative test.

People's own posts as a free pre-test

The cheapest test on LinkedIn costs nothing in media. Before an angle goes into a paid campaign, the people who know the subject (a founder, a specialist, a seller) write about it in their own posts. The personal branding guide describes the analytics LinkedIn provides for individual posts: impressions, reactions, comments, reposts and the demographics of who engaged. Those demographics are the useful part. A post that drew comments from the job functions and seniority the campaign targets has shown that the angle matters to the right people; a post that drew applause from peers has not.

Run it as a deliberate pre-test, not as hope. Write down the angles you are considering, give each to the person best placed to write about it, and record for each post who engaged and what they said. The comments are often more useful than the counts, because they show the words buyers use for the problem, which become the first line of the ad. Treat the result as a filter, not a verdict: organic reach depends on the writer's network and on the day, so a weak post may be a weak post rather than a weak angle.

A post that does well can also become the ad itself. LinkedIn's thought leader ads sponsor a post written by a named person, an employee or another member who has given permission, as the campaign objectives and formats guide describes, which lets a pre-tested post carry its comments and its author's face into a paid campaign. The guide to LinkedIn thought leader ads covers choosing the people and the posts, permission and disclosure, and how to measure what a person's post adds.

Rotation inside a campaign: choosing, not proving

Putting a few ads in one campaign and letting delivery run is the commonest LinkedIn test, and it is useful for one job: choosing which of several reasonable ads should carry the campaign now. Because delivery favours the ads members respond to, the system quickly concentrates impressions on the ad it predicts will do best. That is what you want when the aim is to run the campaign well. It is not what you want when the aim is to learn, because the ads end up reaching different people in different amounts, and a weaker ad may never get enough delivery to show what it could do. The ad creative hook test explains the same problem from the experiment side and why a split test gives each version its own audience.

Treat rotation accordingly. Put in a small number of ads that differ in a way you can describe, read which ones the campaign settles on, and keep the result as a lead rather than a finding. Carousel ads offer a small bonus here: each card is reported separately, which shows where attention falls away inside one ad without needing a test. Rotation is also the fatigue defence the objectives guide asks for: have the next variation ready before the current one tires, and use the weekly read of frequency and CTR per ad that the diagnosis page for rising LinkedIn Ads costs recommends to decide when to swap it in.

A controlled comparison on split audiences

When a question changes what the business offers, or where a large share of budget will go, it deserves a controlled comparison: an A/B Test in which each person can see only one version. The experiments library describes the two routes. Use the platform's own test tool where it supports the comparison; where it does not, run two campaigns on mutually exclusive audiences, as the native lead form test does. LinkedIn's help centre is the reference for whether the tool is available to an account, what it can compare and how it is set up, and this guide does not describe those settings.

The split-campaign route depends on the audiences staying apart. The targeting and Lead Gen Forms guide explains that campaigns with overlapping audiences bid against each other, and that exclusions keep each member in one campaign. Divide the audience on something that does not itself change the result: halves of an uploaded company list split at random, for example, rather than one region against another. Exclude each half from the other campaign, and keep audience, budget, schedule, bidding and objective identical so the creative is the only difference.

Decide before launch what result would change the plan, how long the test will run and what happens if it is inconclusive. The experiment hypothesis template holds those decisions on one page. The Sample Size a test needs and whether a difference reaches Statistical Significance are reasoned in the experiment designs, not here; the honest reading on a small LinkedIn audience is often that the arms cannot be separated yet, and the right response is to keep running, pool with a later test or accept that the difference is too small to matter.

What to vary first

The order of tests matters more on LinkedIn than on a cheaper platform, because each test costs a period of budget. Start with the variables most likely to change the result and work down.

OrderVariableExamplesWhy this order
FirstThe offerA checklist, a report, a calculator, a briefing, a conversationDecides who raises a hand and why
SecondThe angleThe problem led with, the role named, the claim madeDecides whether the right people stop
ThirdThe formatSingle image, document, video, carousel, a named person's postChanges how much the member gets before the ask
FourthThe executionFirst line, headline, image, call to actionRefines a combination that already works

A brief for each concept, written with the creative brief template, keeps the variable under test explicit and the rest of the ad steady. The creative production playbook covers producing variations in batches, which is what a rotation habit needs.

Reading a test on accepted leads, not clicks

Click-through is the figure LinkedIn reports first, and it is the wrong place to end. An ad that names a broad pain draws more clicks than one that names a specific role, and the extra clicks are frequently from people sales will not work. Read every test down the chain: click-through as a mechanical check (the CTR calculator gives the rate on the same basis for each arm), then leads, then the leads sales accepted as a Sales-Qualified Lead or at whatever stage the business uses, and cost per accepted lead through the CPA calculator. The arm that wins on accepted leads wins the test, even if it lost on click-through. When click-through is low on every arm from launch, the diagnosis for a low LinkedIn click-through rate checks the objective, the audience and its delivery settings, the opening line and the offer before any arm is judged.

That reading takes longer, because sales need time to work the leads. Agree with sales before launch how leads from each arm will be marked and how quickly they will be judged, and make sure both arms enter the same queue, as the experiments library advises. Compare the result with the account's own history rather than with any figure from elsewhere, and quote the counts beside the rates, so that a difference built on a handful of leads is read as the small number it is.

Naming concepts so reports can group them

A single test answers a single question; a programme learns across many. That only works if results can be grouped by concept over months, which depends on names. Give every ad a name that records the offer, the angle, the format and the version, in a fixed order, following a Naming Convention the whole account uses. The guide to structuring a LinkedIn Ads account sets out names that let reports roll up from ad to campaign to campaign group. A report can then show every ad that used the renewal angle, or every document ad, across campaigns and quarters, rather than a list of ads called "new image final". Keep a short test log beside the names: the question, the arms, the dates, the result on accepted leads and the decision, so the next person does not rerun a test the account has already paid for.

An insurance broker's testing quarter (illustrative)

Illustrative example — not a real result

Plumvexa Insurance Brokers tests offers on a company list

Plumvexa Insurance Brokers, an invented commercial insurance broker selling liability and property cover to manufacturers, has a target company list of 1,800 firms and a quarter's LinkedIn budget of £11,000.

In the first three weeks its three directors each write two posts on one angle: claims that were refused because of missing records, renewals that arrived too late to shop around, and new contract terms that demand higher cover. The claims posts draw 23 comments from finance and operations managers; the renewal posts draw 17, mostly from finance directors; the contract-terms posts draw 9, largely from other brokers. The contract-terms angle is dropped.

In weeks four to six the lead campaign rotates two single image ads on the two surviving angles. Delivery settles on the claims ad, which takes most of the impressions, so the broker treats that as a choice for the running campaign, not as proof.

For weeks seven to twelve it splits the company list at random into two halves of 900 firms, excludes each half from the other campaign, and gives each campaign £3,600 with identical settings.

ArmOfferClick-throughLeadsAccepted by salesCost per leadCost per accepted lead
AClaims checklist0.62%4112£88£300
BRenewal calendar0.47%2915£124£240

The checklist wins on click-through and cost per lead; the calendar wins on what sales accepted. With 12 and 15 accepted leads the difference is small enough to be noise, so the broker keeps both offers running, names every ad by offer, angle, format and version, and repeats the comparison next quarter before moving budget.

Common mistakes

  • Calling a test from rotation. Delivery favours the ads members respond to early, so the ads in one campaign never had a fair comparison.
  • Overlapping arms. Split campaigns without mutual exclusions bid against each other and show some members both versions.
  • Testing the button colour first. The offer and the angle decide most of the result; execution details refine a combination that already works.
  • Judging on click-through. The ad that draws the most clicks is often the one that draws the wrong people.
  • Stopping early because waiting is costly. A small audience produces small counts; read them as a range against the account's own history.
  • Running a test long enough for fatigue to decide it. A tired control and a fresh variant are not a fair pair.
  • Ignoring free evidence. People's own posts show which angles the right roles care about before any budget is spent.
  • Names nobody can group. Without a convention, a quarter of tests becomes a list of ads nobody can learn from.

Frequently asked questions

Can I A/B test LinkedIn ads by putting several ads in one campaign?

You can rotate several ads in one campaign, but that is a way of choosing, not a fair test. Delivery inside a campaign favours the ads members respond to early, so the ads end up reaching different people in different amounts. Use rotation to pick working creative from a few candidates, and keep a controlled comparison, through LinkedIn's own test tool where it supports the comparison or two campaigns on mutually exclusive audiences, for questions you need answered properly.

What should I test first in LinkedIn ads?

Test the offer and the angle before the execution. Whether the ad offers a checklist, a report, a calculator or a conversation, and which problem it leads with, usually moves results more than the colour of the image or the wording of the headline. Once an offer and angle hold up against accepted leads, test format and then the smaller details inside the winning combination.

Why judge a LinkedIn creative test on accepted leads rather than click-through rate?

Because a creative can attract clicks and form fills from people sales will never work. An ad that names a broad pain draws more clicks than one that names a specific role, and the cheaper clicks often come from the wrong people. Click-through and cost per lead are useful guardrails that show the ad is working mechanically; the decision belongs to the cost per lead sales accepted, read against the account's own history.

Terms used on this page

Tools to try

Frameworks

Where this page is used

Sources

  1. LinkedIn Marketing Solutions Help — LinkedIn
  2. LinkedIn Ads — LinkedIn Marketing Solutions

Related