Testing audiences vs testing creatives: what to test first, and why

A campaign has gone two weeks without taking off. The automatic reaction of almost any ads manager, novice or experienced, is the same: "let's try a different audience". They change the targeting, wait a few days, it still doesn't work, try another audience. Three audiences later, the campaign still isn't delivering results and the wrong conclusion starts to settle in: "advertising just doesn't work for my business". The problem was almost never the audience. It was the ad.

There's a practical hierarchy, fairly well established from the experience of thousands of campaigns, about which variable to test first when something isn't working: first the creative (image, video, copy), then the offer or message angle, and only third the audience. Going in the opposite order is the most common way to spend budget without learning anything useful.

Why the creative comes first

The reason is statistical, not a matter of opinion. Today's ad platforms (Google, Meta, TikTok) have very powerful automatic targeting algorithms: when you give them a broad audience and several creatives, the system itself finds, within that audience, the people most receptive to each ad. In practice, this means a large part of the "find the right audience" work is already being done for you by the algorithm, as long as you feed it quality material to work with. If the creative is weak, it doesn't matter how well-chosen the audience is: nobody clicks an ad that doesn't catch their attention, whoever they are.

In a test we ran with a cosmetics brand, we kept the exact same audience and changed only the creative (from a generic product video to a video with a real person explaining the problem it solves) and cost per sale dropped 46%. The audience was never the problem; the problem was that the video didn't make anyone feel that product was for them.

When the audience genuinely is the problem

There are fairly clear signals that the problem lies in targeting, not the creative. The first is when the ad has a good click-through rate (people find it appealing, stop, click) but then a very low conversion rate on the landing page: that usually indicates you're attracting people who don't match what you offer, even though the ad catches their eye for other reasons. The second signal is an abnormally high cost per click compared with sector averages, which can indicate you're competing for an audience that's either too contested or poorly matched to your product.

The recommended testing order, step by step

First, test three to five clearly distinct creatives (not minor variations on the same idea, but genuinely different message angles: one focused on price, another on a specific problem, another on social proof) with the same broad audience. Second, once you've identified the winning creative, test different offer or message angles within that same creative line (for example, "free shipping" versus "15% discount" versus "free first consultation"). Only third, with the creative and message already validated, do you move on to testing audience variations: ages, interests, lookalikes, geographies.

This sequence isn't absolute dogma, there are businesses where the audience is so clearly critical (for example, a product only relevant to a very specific type of professional) that it's worth fixing it from the start and not leaving it open to the algorithm. But as a general rule for most small businesses with a mass or semi-mass consumer product, starting with the creative gets results faster and cheaper.

How to test without splintering the budget into uselessness

The most common execution mistake when testing is creating too many variants at once with too little budget each, leaving each variant without enough data for the algorithm to learn anything. A practical rule: each variant under test should have enough budget to generate at least 20-30 conversion events (or, if the objective is higher up the funnel, at least 1,000-2,000 impressions) before drawing conclusions. With less than that, what looks like a clear winner can just be statistical noise.

Use the platforms' native A/B test tools when available: they split budget and audience in a controlled way between variants, something that doing manually (by creating separate campaigns) tends to introduce bias into, because campaigns don't always compete on equal footing within the auction.

An important nuance: fatigue and freshness

Even a winning creative stops performing over time, because the same audience sees it again and again and stops reacting (this is known as ad fatigue). That's why creative testing isn't a one-off exercise done at launch, it's something worth repeating every four to eight weeks, especially in campaigns with a high budget and a relatively small audience, where fatigue arrives sooner.

A case where the testing order changed the final conclusion

A sports physiotherapy clinic had spent two months testing different audiences to attract new patients: amateur athletes, people over 50 with chronic pain, people who had recently suffered an injury. No audience produced a cost per lead below 40 euros, a figure the business couldn't sustain. Before testing more audiences, they decided to apply the reverse hierarchy: fix a single broad audience (adults aged 25 to 65 within a 15-kilometre radius) and test three completely different creatives.

The winning creative, a 20-second video with a real patient explaining how they'd gone back to running after an injury they thought would stay with them forever, brought cost per lead down to 14 euros, with the same broad audience they'd been dismissing as "not specific enough". The problem had never been targeting: it was that none of the previous creatives (generic clinic photos, copy focused on equipment technology) connected with the real fear or hope of someone looking for physiotherapy. Only after fixing that winning creative did testing audience variations deliver an additional improvement, though far more modest than the leap achieved by changing the creative.

How to document results so you don't repeat tests already run

A common problem in businesses that test constantly is keeping no record of what's already been tried and with what result, which leads to repeating tests months later without realising they'd already been done. A simple spreadsheet with four columns (what was tested, date, result, conclusion) avoids that waste and, over time, lets you identify patterns specific to your business that go beyond any single test: for example, that creatives featuring a real person always outperform product-only ones, a pattern that only becomes visible if there's an organised history to look back on.

What to do when neither the creative nor the audience seems to be the problem

There's a third scenario, less common but real, where neither the creative nor the audience explains poor performance: the problem is the offer itself. If different creatives with different audiences generate a reasonable click rate but none convert well on the landing page, it's worth also testing the price itself, the shipping terms or the guarantees offered, before continuing to tweak the communication around something that, fundamentally, isn't attractive enough as currently framed.

An inexpensive way to test this hypothesis without rebuilding the whole product: test two identical landing pages differing in only one offer element (for example, free shipping versus paid shipping with an equivalent discount), while keeping the creative and audience that are already known to work fixed. If that variable does clearly move the needle, it confirms the problem was the offer, not how it was being communicated.

Why this hierarchy sometimes flips on very mature accounts

On advertising accounts with years of history and very high budgets, where the creative has already been extensively optimised and almost every obvious message angle has already been tested, there can come a point where testing new audiences (entering a different country, targeting an unexplored age bracket) delivers more room for improvement than continuing to polish an already very refined creative. This exception doesn't invalidate the general rule for the vast majority of businesses, which still have plenty of untapped room in their creative, but it's worth knowing the hierarchy isn't an immutable law of physics, it's a guide based on where the biggest room for improvement usually sits at each stage of an account's maturity.

Step by step for setting up your first test following this hierarchy

For anyone who has never deliberately structured a test, here's a concrete four-week process. Week 1: define a reasonable broad audience (it doesn't need to be perfect, just coherent with your product) and prepare three to five creatives with clearly distinct message angles, not minor variations on the same idea. Week 2: launch the test with enough budget for each creative to receive at least 1,000-2,000 impressions or, if the goal is conversion, at least 15-20 events, and don't touch anything during those first few days except obvious setup errors. Week 3: identify the winning creative with the data already gathered, and launch a second test on that same creative testing two or three offer or message variations. Week 4: with creative and offer already validated, start testing audience variations (ages, interests, lookalikes) on top of that already-winning base.

This four-week process isn't a rigid formula (some businesses need more time due to low volume, others can speed it up with more budget), but it gives a concrete structure to follow instead of testing haphazardly, which is what wastes the most time and budget in practice.

Frequently asked questions