Ask the internet how many ads you should be testing and you’ll get every answer going. Four to six a week. Ten to twenty at a time. Across 17 top brands, the median launches 25 new creatives a week, and the busiest launches 287 (Adspo). Somewhere in that range, a business spending $10,000 a month on Meta is supposed to find its number.
Here’s the uncomfortable part. Most of that advice is written for budgets ten or a hundred times bigger than yours. And a good share of it comes from companies that sell ad production or ad-launching tools, so the answer leans one way. The maths isn’t wrong. The starting point just isn’t yours.
Why this matters
Every ad you add splits the same budget
The better guides agree on the method, and they’re right: take your weekly testing budget and divide it by the spend one ad needs before you can judge it (Adspo, AdManage, Aden’s Lab). A common bar for that spend is a few times your target cost per purchase or per lead (Adspo).
Meta points the same way. When you create many ads, its delivery system learns less about each one, and an ad set usually needs about 50 results in a week to leave the learning phase (Meta). We covered that floor in last week’s piece on the learning phase. Testing sits on top of it: every extra ad is another mouth to feed from the same plate.
Meta doesn’t share the plate fairly
Put five ads in an ad set and Meta won’t split the money five ways. It moves the budget towards the ads its engagement and conversion signals favour (Brkfst). Some advertisers report only one of three creatives in an ad set getting any spend at all (AdManage). So the more ads you load in, the more of them never get a fair go.
The maths at your budget
Here’s the sum at three budgets we see every week. We’ve assumed 20% of spend goes to testing and that each ad needs about $120 before you judge it, which is three times a $40 cost per purchase. Both are illustrative inputs, not benchmarks, so plug in your own. Even the vendors warn against treating any testing share as universal (AdManage).
At $5,000 a month, you’re spending about $1,150 a week. A fifth of that is about $230, enough to judge about two new ads a week.
At $15,000 a month, it’s about $3,460 a week and about $690 for testing. That judges about five new ads.
At $30,000 a month, it’s about $6,920 a week and about $1,385 for testing. That’s about eleven.
Compare that with what you read online. One guide suggests 10 to 20 creatives in active testing for anyone spending under US$1,000 a day (AdManage). That’s already above the top of our range. At $5,000 a month, the honest number is closer to two ads a week. Lead generation with a $20 cost per lead can judge more. A high-ticket store with a $150 cost per sale can judge far fewer.
Google is a different game
On Google Search the ad isn’t really the unit anymore. A responsive search ad takes up to 15 headlines and 4 descriptions and tests the combinations itself (Google). Google recommends at least two responsive search ads with Good or Excellent Ad Strength per ad group (Google).
Its own numbers show the curve flattening. Adding a second responsive search ad lifted conversions by 6.6% on average; adding a third lifted them by 3.7% (Google). Worth having. Not a reason to keep stacking.
A lot of “three ads per ad group” advice still floating around recommends two expanded text ads and one responsive search ad (Karooya). Expanded text ads haven’t been creatable since 30 June 2022 (Google). On smaller budgets, the bigger Google question is campaign count, not ad count: spread the budget too thin and no campaign gets enough conversions for Smart Bidding to work (Digital4Startups, AdControlCenter).
What people overestimate
That more ads means more winners
This is where the argument starts. Brkfst says only 2% of creatives become winners in its data, and that smaller budgets need more ads per dollar to find them (Brkfst). AdManage describes an advertiser spending $30,000 a day with a hit rate of about 6 to 7% (AdManage). At two tests a week, a 2% hit rate works out to roughly one winner every six months.
We read it differently. A hit rate is usually a count of files, and ten versions of one idea are still one idea (Aden’s Lab, AdManage). On a small budget you can’t buy your way to a winner with volume. You raise the hit rate instead: fewer tests, each one a genuinely different reason to buy, built from what your customers actually say in reviews, emails and sales calls. That is the only lever that doesn’t cost more budget.
That a quiet ad is a losing ad
An ad that barely spent hasn’t failed. It hasn’t been tested. Even the vendors who sell volume say not to count an inconclusive, low-spend ad as a failure (AdManage). Meta’s own built-in creative test, which compares two to seven versions, reports top performers without a confidence level (AdManage). Keep a simple log with three outcomes, not two: won, lost, and not yet tested. The third column is usually the longest.
That the winner is the one with the best click rate
Pick the metric before the test, and stick to it (Marketingblatt). Google says it plainly: a high click-through rate isn’t the end goal, and ad tests should be judged on incremental impressions, clicks and conversions for the whole ad group or campaign (Google). On Meta, the same logic applies. An ad that gets cheap clicks and expensive customers is a losing ad with a good-looking dashboard. Judge the test on the number the business banks: cost per qualified lead, profit per order. That only works if your tracking can see it, which is why we fix measurement before we scale testing.
That you can copy a big brand’s kill rate
AG1 launched 349 new creatives in one October week and switched off 202 of them within three days (Adspo). At its spend, three days is enough to read a result. At $5,000 a month, three days is noise. A test with 20 clicks is a snapshot, not proof, and most clean tests need two to four weeks (Marketingblatt). Borrow the discipline, not the timetable.
Where to start this week
Do the sum for your own account: monthly budget to weekly, your testing share, divided by about three times your cost per result. That’s how many ads you can judge, not how many you can make. Fill those slots with different ideas, not different files. Decide the winning metric and the spend you’ll wait for before you launch, and log every ad as won, lost or not yet tested. On Google, run two strong responsive search ads per ad group and judge the ad group, not the ad. And with Black Friday two weeks away, don’t launch your next test batch into it: volatile weeks distort results (Marketingblatt). Plan it for early December instead.
Dadek Digital sets the testing budget, the read and the definition of a winner on the business’s own numbers before a single test goes live, so every test teaches the next brief something useful. The full structure, from measurement at the base to creative testing on top, is laid out in our playbook, The Dadek Digital Pyramid.