In short — There's no magic number, and the platforms don't claim one. TikTok recommends 3–5 creatives per ad group and adding fresh ones to the same ad group when results decline. Meta gives no fixed count: it asks for genuinely different concepts rather than twenty variations of one ad. The number that actually limits you is budget: each creative needs enough spend to produce a signal. A workable rule is to run as many creatives as your weekly budget can fund at 2–3× your target CPA each, and to refresh the losers weekly. With AI generation at roughly $5–$9 per finished ad, production is no longer the bottleneck. Your media budget is.
Disclosure: Klaivo is our product, an AI ad generator. Platform guidance below is quoted from Meta's and TikTok's own documentation. The budget rule and cadence table are our working method, labelled as such, not platform rules.
What the platforms actually say
Search this question and you'll find confident numbers that contradict each other: "3–5 per ad set", "6 or fewer", "15–25 minimum since Andromeda". Here is what the two platforms themselves document.
TikTok is the most specific. Its creative best practices recommend "between 3-5 different creatives per ad group and 3-5 diversified ad groups per campaign." It advises refreshing when "delivery results exhibit a consistently declining trend," and to "add new creatives to an existing ad group instead of creating a new ad group to extend its lifetime" (TikTok Ads Manager).
Meta deliberately avoids a number. Its guidance on creative diversification asks for distinct assets tailored to different personas or use cases: different in visuals, messaging and format. It says that "simply iterating on a single concept isn't enough," because ads that share key visual and thematic attributes are grouped and optimized together as variations (Meta for Business). The reason is Andromeda, Meta's ads retrieval system, which Meta's engineering team describes as narrowing tens of millions of ads to the few thousand considered for each person. More distinct creative gives it more to match.
So the honest synthesis is:
- TikTok: 3–5 per ad group, refreshed inside the same ad group.
- Meta: no fixed count; quantity of concepts matters, not quantity of files.
Concepts vs variations: why 20 ads can count as one
This is where most creative testing goes wrong. On Meta, these are one concept, not five:
- the same video with five different opening lines;
- the same ad in five colors;
- the same script read by five voices.
These are five concepts:
- A problem-first hook for a skeptical buyer ("Your afternoon crash isn't about sleep").
- A person-to-camera testimonial for a social-proof buyer.
- A product demo for a feature-driven buyer.
- A comparison with the old way for a switcher.
- An offer-led ad for a deal-driven buyer.
Variations still have a job: once a concept wins, hook and format variations help you find its best version. But they're a second step, not a substitute for testing different angles. The four-beat structure behind each concept is in the video ad structure that sells.
How many creatives your budget can actually test
A creative that never gets enough spend teaches you nothing. The limit is arithmetic, not the platform.
Our working rule: give each creative in a test enough budget to reach 2–3 conversions' worth of spend before you judge it. Then:
Creatives you can test per week ≈ weekly test budget ÷ (2.5 × target CPA)
Worked examples:
| Weekly test budget | Target CPA | Creatives you can judge per week |
|---|---|---|
| $500 | $40 | ~5 |
| $1,500 | $40 | ~15 |
| $1,500 | $100 | ~6 |
| $5,000 | $50 | ~40 |
Two consequences follow:
- A small account running 25 creatives isn't testing; it's diluting. Most ads never get enough spend to be judged, and the algorithm concentrates delivery on one or two anyway.
- A large account running 4 creatives is starving the algorithm. It has budget for far more angles than it gives the system.
This rule is a starting point for deciding volume, not a statistical significance test. Adjust it once you see how fast your account typically concentrates spend.
How to spot creative fatigue
Fatigue is when a creative's performance declines because the same people have seen it too often, not because the offer or audience changed. Watch these together, over a few days rather than one:
| Signal | What it looks like |
|---|---|
| Frequency rising | The same people see the ad again and again |
| Hook rate / thumb-stop falling | Fewer viewers get past the first seconds |
| CTR falling | Fewer clicks at the same reach |
| CPA or CPM rising | Each result costs more for the same creative |
One bad day is noise. A consistent decline across two or more of these signals is fatigue, which matches TikTok's own trigger of a "consistently declining trend."
A weekly refresh cadence by spend
A cadence turns fatigue from an emergency into a routine. This is the one we suggest as a starting point:
| Monthly spend per platform | New concepts per week | Variations of winners per week | Retire |
|---|---|---|---|
| Under $3k | 2–3 | 2–3 | Anything fatigued or below target after its test budget |
| $3k–$15k | 4–6 | 4–6 | Bottom third weekly |
| $15k+ | 8–15 | 10+ | Bottom third weekly, winners reviewed twice a week |
On TikTok, add the new creatives to the existing ad group (TikTok's recommendation) rather than restarting learning in a new one. On Meta, make sure each new batch contains different concepts, not only hook swaps of last week's winner.
What that volume costs to produce
This is what changed. When a creator video cost $100–$300 (Influee), a cadence of 5–10 new creatives a week was an agency-sized budget line. With AI generation, a finished 3-shot vertical ad costs roughly:
| Production route | Cost per finished ad | 10 new ads a week |
|---|---|---|
| Human UGC creator | $100–$300 | $1,000–$3,000 |
| AI actor tool billed per video (e.g. ~$11/video) | ~$11 | ~$110 |
| AI generation, budget model (Seedance 2.0 Mini, 480p) | $5.32 | ~$53 |
| AI generation, mid model (Seedance 2.0 or Kling 3.0) | $9.31 | ~$93 |
AI generation figures: Klaivo retail at $0.19/credit, 3 shots with AI voiceover and music, September 2026. Budget one extra shot per ad for retries.
Hook tests are cheaper still: regenerating only the first shot of a winning ad costs one shot, about $1.52–$2.85 depending on the model. That's why hook variations are the cheapest way to extend a winner's life. Model-by-model costs are in Seedance 2.0 vs Kling 3.0 vs Veo 3.1.
The practical consequence: production stops capping your testing, and media budget becomes the real limit. That's exactly why the budget rule above matters more than any "magic number."
Five mistakes that waste creative budget
- Counting variations as concepts. Twenty hook swaps of one idea is one test on Meta.
- Launching more creatives than the budget can judge. Use the formula; don't let most ads die unjudged.
- Restarting ad groups to refresh. On TikTok, add to the existing ad group.
- Killing on one day's data. Wait for a consistent decline across several signals.
- Scaling the test version. Once a concept wins, regenerate it at a higher quality tier before putting real spend behind it; see what an AI video ad really costs.
FAQ
How many creatives per ad set on Meta?
Meta doesn't set a number. It recommends a diverse set of genuinely different concepts per audience. In practice, size it to your budget: roughly your weekly test budget divided by 2.5× your target CPA.
How many creatives per ad group on TikTok?
TikTok recommends 3–5 creatives per ad group and 3–5 ad groups per campaign, adding new creatives to the existing ad group when results start to decline.
How often should I refresh ad creatives?
Refresh when performance declines consistently: frequency up, CTR or hook rate down, CPA up. For most accounts spending actively, that means new creatives every week. TikTok itself frames refresh as a regular supply rather than a one-off.
Does Meta's Andromeda mean I need more ads?
It means you need more distinct ads. Andromeda matches ads to people from a very large pool, and Meta groups near-identical variations together, so extra copies of one concept add little.
What's the cheapest way to produce enough creatives?
Test concepts with AI-generated ads on a budget model (around $5 per finished 3-shot ad), then produce variations of winners by regenerating only the hook shot. Tools that bill per credit rather than per finished video make this kind of testing much cheaper; see our TikTok ad generator comparison.