Early signs of creative fatigue: spotting it 7 days before CPA spikes

By the time cost-per-acquisition rises on Meta, Google Ads, or TikTok, the creative has been fatiguing for over a week. The lag is structural. Conversions sit inside attribution windows. Delivery engines re-explore. CPA is the last metric to move because it is downstream of every other signal in the funnel.

That structural lag is the single most expensive thing about most ad accounts. The signals that mean "this creative is dying" arrive first as click-through-rate decay, then as frequency creep, then as engagement drop, and only then as CPA spike. Reading the early signals correctly buys back roughly seven days of budget that would otherwise be spent training a delivery system on a creative that has already lost the audience.

This guide covers why CPA arrives last, the four leading indicators in the order they move, the diagnostic sequence for reading them, the platform-specific differences across Meta, Google Ads, and TikTok, and how to decide between refreshing and killing.

Why CPA arrives last

Three structural delays sit between a creative starting to fatigue and CPA reflecting it on a dashboard.

Attribution windows. Meta's default is 7-day click + 1-day view. Google Ads default click attribution runs to 30 days. TikTok defaults to 28-day click. A conversion that fires today may have started its journey from an impression a week ago, and the conversion is reported under the creative active at the click, not the creative active when the dashboard is being read. CPA on day seven of a creative's fatigue still shows you a blend that includes pre-fatigue clicks.

Learning-phase requirements. Each new creative needs roughly $50-$150 of spend and seven days to exit the learning phase on Meta. Google Ads similarly recommends 50 conversions per ad group per week. Until that threshold is crossed, delivery is exploratory and CPA is artificially elevated. So the first week of any creative's life is noisy by design — and the platform's automatic fatigue labels are not active yet.

Cohort recycling. Delivery engines re-show ads to the same audiences in cycles. The same creative shown to the same cohort for the third time is structurally less interesting than the first; auction prices rise, click-through-rate drops, and conversion follows. By the time CPA reflects this on a dashboard, the auction prices have already been rising for several days.

The combined effect is a roughly 5-7 day lag between the moment a creative meaningfully starts fatiguing and the moment CPA confirms it. That window is what early indicators recover.

The four leading indicators in the order they move

These four metrics move in a predictable sequence ahead of CPA. Watch them in this order.

1. Frequency creep

Frequency is the average number of times a unique user has seen the creative. It moves first because it is mechanical — the platform decides who sees the ad, and the ratio of impressions to reached users rises whenever the audience pool starts to saturate.

Healthy frequency in a 7-day window for most consumer advertisers is between 1.0 and 2.5. Above 3.0, audiences are tuning out. Above 5.0 in a 7-day window, even strong creatives usually degrade. The number itself is less diagnostic than the slope: frequency rising 30% week-over-week without a corresponding rise in reach is the cleanest leading signal of fatigue available.

Brand advertisers run at higher frequencies intentionally — frequency caps of 7 or 8 over a 30-day window are common in brand campaigns. Direct-response advertisers should not.

2. CTR decay

Click-through rate is the earliest objective signal that fatigue has reached the user, not just the auction. It moves faster than ROAS or CPA because clicks are the same-session response to the impression — no attribution window stretches it.

The slope matters more than the level. A creative running at 2.0% CTR that has dropped from 3.0% over ten days is fatiguing faster than one stable at 1.2% for a month. A 25% drop in CTR over a rolling 7-day window is the conventional fatigue threshold; a 50% drop is unambiguous.

CTR decay also moves before frequency creep is severe enough to be alarming on its own. The two together — CTR down 25% and frequency rising — narrow the diagnosis sharply.

3. Engagement drop

Engagement covers the qualitative reactions to the creative: video view-through rate at 25%, 50%, 75% completion; comment volume; share count; reaction mix. These metrics matter for a different reason than CTR: they describe how the audience is reacting to the creative given the click, not whether the creative is winning the auction.

Engagement drop is the bridge between "the audience has seen this enough" (CTR) and "the audience now actively prefers to scroll past" (CPA). When 25%-completion video view rate drops 20% and reaction mix shifts negative, that is the audience telling the algorithm to demote the creative — and the algorithm will, by raising auction prices.

4. CPM rise

Cost-per-thousand-impressions is partly an auction signal and partly a fatigue signal. It rises mechanically when delivery engines decide the creative is less likely to convert, because they have to pay more to place an ad that engages less. CPM trailing CTR by a day or two is the platform's price response to the audience's behavioural response.

CPM is also noisy — auction-level effects (competitor budgets, broader market activity) move it independently. Use it as confirmation rather than primary diagnosis. A CPM rise without CTR decay is a market signal, not a fatigue signal. A CPM rise with CTR decay is fatigue, with high confidence.

The diagnostic order: read them as a sequence

The four signals are most diagnostic when read in the order they move, not in isolation. The sequence is:

1. Frequency climbing 30%+ WoW with stable reach — flag, not diagnostic on its own
2. CTR drop of 25%+ over a 7-day rolling window — diagnostic when paired with the frequency flag
3. Engagement drop of 20%+ on 25%-completion video view rate or reaction mix — confirmation
4. CPM rise without external auction-market explanation — final confirmation

A creative showing all four is fatiguing with very high confidence. A creative showing only one or two is either entering fatigue (early enough to refresh) or being misread.

The most common diagnostic error is reading CPA in isolation, decision-making on the spike, and concluding the creative was fundamentally weak. It probably was not — it probably exhausted its audience and would have been the right creative to refresh and re-launch with adjusted hook or audience widening.

Platform variance: Meta, Google Ads, and TikTok

Each platform's fatigue signal pattern is slightly different.

Meta. Meta's automatic fatigue labels (rolled out late 2024) catch fatigue at roughly the CPM-rise stage — useful as a final confirmation but late enough that the seven-day window has already been spent. Meta is the platform where reading frequency and CTR ahead of the labels matters most. Andromeda delivery clustering also means similar creatives within an account compete with each other for impression share, which compresses fatigue cycles.

Google Ads. Search and Performance Max have lower frequency exposure (fewer impressions per user per week) so frequency creep is less diagnostic. CTR decay matters most. Google's quality score also encodes a long-running fatigue signal; quality score dropping a tier across an ad group is a strong delayed indicator that the creatives are tiring.

TikTok. TikTok's audience cycles fastest of the three because of its For-You-Page-driven discovery loop. A creative that fatigues on TikTok in five days might run on Meta for two weeks before showing the same signals. Hook retention at the 3-second mark is the dominant TikTok fatigue signal — when 3-second view rate drops 25%, the creative is functionally dead even if CTR has not moved yet.

For more on reading the same creative across the three platforms together, see reading cross-platform ROAS.

Refresh or kill: the decision

Spotting fatigue early opens up one decision the post-CPA-spike timeline does not: refresh versus kill.

Refresh when the creative had a strong opening week, the audience has saturated, but the underlying angle and hook still fit the brand and product. A refresh changes the surface — a different opening frame, a swapped voice-over, a re-cut for vertical, fresh end-card — while keeping the angle. Refreshes are cheap and re-set the audience clock.

Kill when the creative had a weak opening week (CTR already below benchmark before fatigue signals), the angle is fundamentally not working, or you have a stronger replacement ready. Killing locks in the budget already spent and stops feeding a delivery engine training data on a losing creative.

The distinction matters because most accounts under-refresh and over-kill. A killed creative loses all its accumulated audience-learning and starts cold. A refreshed creative inherits the targeting the platform learned during its first run, often producing a stronger second wave than a brand-new creative would.

The full decision tree, and the brief structure for writing the next variation, is in what to test next: closed-loop creative iteration.

What Omniscia reads on this

Lens scores creatives across six dimensions before they spend, which is upstream of all of this — see how to score a video ad. Cortex aggregates fatigue cohorts across the customer base, so the fatigue thresholds in this guide are continuously re-fit against actual delivery data on Meta, Google Ads, and TikTok rather than relying on platform-published rules of thumb. The combination flags the four indicators above as they move, in the order they move, against current cross-account benchmarks.

For a deeper read on what fatigue is and how it shows up in dashboards, see the field guide on creative fatigue.

Frequently asked questions

What frequency is too high on Meta in 2026?

For direct-response advertisers, frequency above 3.0 in a 7-day window is the conventional ceiling, with degradation accelerating above 5.0. Brand campaigns run higher (7-8 in a 30-day window is common). The slope matters more than the level: frequency rising 30% week-over-week without matching reach growth is a stronger fatigue signal than any specific number, because it isolates audience saturation from intentional brand exposure.

How fast does CTR drop before CPA does?

CTR typically moves 5-7 days ahead of CPA on Meta and TikTok, and 7-14 days ahead on Google Ads. The lag is driven by attribution windows (Meta 7-day click default, Google Ads 30-day click default, TikTok 28-day click default) plus learning-phase smoothing. A 25% rolling-7-day CTR drop is the most reliable single early signal across all three platforms.

Does creative fatigue show up the same way on TikTok?

No. TikTok's audience cycles fastest because of its For-You-Page recommendation loop, so fatigue compresses into roughly half the timeline of Meta. Hook retention at the 3-second view mark is the dominant TikTok signal — when 3-second view rate drops 25%, the creative is functionally dead even if CTR has not moved. Frequency creep is also less diagnostic on TikTok because audience pools refresh faster.

Should I refresh the creative or refresh the audience?

Refresh the creative when the audience is saturated but the angle is sound. Refresh the audience (widen targeting, rotate lookalike percentages) when the creative is strong but the seed cohort is exhausted. The diagnostic question is whether CTR was high in the first week. High first-week CTR with later fatigue points to audience saturation — refresh the audience or refresh the surface of the creative. Low first-week CTR points to angle weakness — refresh the angle entirely or kill.

Can I rely on Meta's automatic fatigue labels alone?

No. Meta's automatic fatigue labels (rolled out late 2024) trigger at roughly the CPM-rise stage, which is the fourth of the four leading indicators in this guide. By the time the label fires, CPA is typically within 1-2 days of moving. Useful as final confirmation, not as primary diagnosis. Reading frequency and CTR directly catches fatigue 5-7 days earlier.