Somewhere along the way, "frequency above 3 means fatigue" became a rule. It gets built into alerting, written into playbooks, and used to justify pausing ads that are doing fine.
The largest public study of frequency effects in online display says the rule doesn't exist.
2.8 billion impressions, and no single curve
The other 66% did show wear-out, but the range was enormous. Some campaigns got roughly 40× the impact from 40 ads that they got from one — near-linear returns. Others got only 3×. Same platform, same period, wildly different curves.
Two findings matter more than the headline:
- Marginal effects stayed positive. Within the ranges measured, additional exposures continued to add response. Diminishing, frequently — but not negative. The intuition that impressions past a threshold actively damage performance did not show up.
- Naïve estimates lied in half the cases. Simple before-and-after comparisons falsely detected increasing returns to frequency in 50% of campaigns. The effect vanished under proper experimental controls.
That second point is the one to take personally. If a straightforward read of the numbers produces a confident, wrong answer half the time, then the frequency chart in your dashboard is not a reliable basis for a decision.
This is display advertising, at scale, and it's now some years old. Social video in a ranked feed is a different environment with different creative and different auction dynamics. Take the structural finding — frequency effects vary enormously by campaign and the naïve read is often wrong — rather than the specific coefficients.
Why the threshold feels true anyway
Because something real is happening; it's just being measured with the wrong instrument.
Frequency rises when reach stops growing. Reach stops growing when the audience is saturated, or when the algorithm has narrowed delivery to the people most likely to respond — and that cohort is finite. So rising frequency reliably co-occurs with declining performance without being its cause. Pausing the ad resets the symptom and the apparent fix confirms the belief.
The distinction is not academic:
| If the real problem is | Pausing the creative | What actually helps |
|---|---|---|
| Genuine creative wear-out | Works | New creative with a different angle or hook |
| Audience saturation | Looks like it works, briefly | Expand the audience, or accept the ceiling |
| Auction pressure / seasonality | Does nothing | Bid and budget, not creative |
Swapping creative because frequency crossed an arbitrary line costs you a working ad and a production cycle, and leaves the actual constraint untouched.
How to tell them apart
Genuine creative fatigue has a signature that saturation doesn't: the response rate falls among people seeing it for the first time. If CTR is declining overall but holding steady for new reach, you have a saturation problem. If it's declining for everyone, including fresh eyes, that's the creative.
Most platform reporting won't hand you that split directly. A workable proxy: watch CTR trend against frequency trend over a rolling window, and treat a CTR decline that starts before frequency climbs as a creative problem, and one that starts after as a delivery problem.
And set the threshold from your own account, not from a blog post. Your account has a frequency at which response actually degrades. It is knowable. It is almost certainly not 3.
Knack doesn't fire a fatigue flag on frequency alone. Fatigue is flagged when a CTR decline and elevated frequency appear together, and a softer "fatigue risk" when frequency is elevated and CTR has fallen relative to the account's own average — not a fixed number carried in from somewhere else. Every flag names the condition that triggered it, so you can tell a delivery problem from a creative one instead of guessing.
What to change
- Delete the fixed frequency threshold from your alerting. It is producing confident false positives.
- Alert on response decline instead, with frequency as context rather than cause.
- Before replacing a creative, check whether reach is still growing. If it isn't, the creative probably isn't the problem.
Sources
- Lewis, R. A. Worn-Out or Just Getting Started? The Impact of Frequency in Online Display Advertising. 2.8 billion impressions, 30 campaigns, 609 million visitor-days of individual-level data.
- Advertising Research Foundation, Television Advertising Wearout (Knowledge at Hand) — the longer research tradition on wear-out, for context on how contested the effect has always been.