Frequency hits 3.2, cost per lead climbs, and someone asks Claude which Meta ads to kill. I want one question answered first: did repeated exposure weaken the response, or did something else get more expensive?

Frequency is average exposure: impressions divided by reach. It does not measure boredom or lost sales. A rising frequency number is a reason to investigate, not a verdict on the creative. Everything else follows from that distinction.

An ad does not expire at frequency 3.0

For a running example, say I’m looking at a home-services prospecting ad set spending around $150 a day. One video has carried it for five weeks. Frequency climbs from 1.8 to 3.1, and cost per lead climbs from $42 to $61.

That looks suspicious. It is not yet a diagnosis.

Frequency is an average exposure count for Accounts Center accounts, estimated from sampled data. An average of 3.0 could mean everyone saw the ad three times. It could also mean 5,000 people saw it once, 2,000 saw it four times and 1,000 saw it eleven times. Same average, very different exposure patterns.

The heavy viewers could be tiring while most of the audience barely remembers the ad. The average cannot tell me which is happening.

Rules such as “refresh prospecting above 3.0” or “retargeting is fine until 8” are not universal Meta thresholds. I used to tell clients 3.5 meant refresh. I was wrong. I had turned a useful warning light into an expiration date.

There is another complication: separate ad IDs do not necessarily mean separate creative experiences. Meta’s fatigue tracking can consider recent exposures across campaigns from the same Page and visually similar creative. Three UGC ads with the same opening can look different in a report and feel identical to the person watching them.

I use frequency to decide where to look, not what to pause.

Make Claude establish the comparison before diagnosing it

Before asking for a fatigue verdict, I ask Claude to show its working conditions:

Read this Meta account through groas. Show the reporting settings and available data before diagnosing creative fatigue. Make no changes.

The response should identify:

  • Account name and ID.
  • Campaign objective and optimization event.
  • Reporting dates and comparison windows.
  • Attribution window.
  • Available fields and missing data.

Claude can read the account’s real numbers through groas once connected. That access is useful only if the comparison makes sense. A leads campaign and a purchases campaign are not interchangeable baselines.

No comparison windows, no verdict.

Follow the mechanism from exposure to acquisition cost

For fatigue to explain the home-services ad’s higher CPL, something has to connect repeated exposure to weaker performance.

The proposed chain is straightforward: repeated exposure weakens click response, so the same thousand impressions produce fewer clicks. If CPM and post-click conversion stay flat, fewer clicks mean fewer leads for the same spend. CPL rises.

That is a plausible mechanism. It is not an inevitable sequence in which the fourth impression suddenly stops working.

I compare the ad with its own early baseline, such as its first two weeks, while keeping the objective and conversion event consistent. The denominators matter:

LayerMetricWhat I’m checking
ExposureFrequency and reachIs average exposure rising, and how is reach changing?
Click responseOutbound CTRAre fewer impressions producing outbound clicks?
Post-click responseLanding page conversion rateAre clicks still becoming leads at a similar rate?
Delivery priceCPMHas the price of impressions changed?
Acquisition costCost per leadWhere does the combined effect land?

Practitioner descriptions of fatigue commonly point to this pattern: frequency up, click-through down, CPM roughly flat and cost per result up. Stable post-click conversion makes the explanation more specific: the deterioration appears before the landing page, not after it.

The useful signal is a connected chain, not a red number in isolation.

Put all three response layers on one timeline

For the home-services video, I ask Claude for a weekly table covering exposure, click response and conversion response, with CPM and CPL alongside them.

Compare this ad with its first two weeks. Show weekly frequency, reach, outbound CTR, landing page conversion rate, CPM and cost per lead. Use consistent reporting settings. Mark unavailable fields as missing; do not reconstruct rates from mismatched windows.

Meta’s own Creative limited or Creative fatigue labels can provide another signal, as described here. I treat them as a second opinion, not a substitute for the timeline.

A commuter beside a row of repeated subway advertisements

Then I read the table in order:

  1. Exposure: did frequency rise?
  2. Click response: did outbound CTR weaken?
  3. Conversion response: did the clicks that remained convert at a similar rate?
  4. Cost: did CPM stay steady while CPL rose?

If outbound CTR falls while landing page conversion stays roughly flat, the evidence points toward the ad. If CTR holds while conversion rate falls, the decline appears after the click. That distinction changes the diagnosis.

Same ad, same windows, all three layers. Otherwise, I’m guessing.

Three lookalikes that should change Claude’s answer

The home-services ad’s move from $42 to $61 per lead could fit fatigue. It could also fit a more expensive auction, a different delivery mix or a weaker landing page. Each explanation predicts a different pattern.

Auction costs rose while response held steady

If CPM rises across campaigns while outbound CTR and landing page conversion stay flat, the account is paying more for the same response. Replacing the creative does not address the cause shown in that comparison.

I ask Claude to look for contradictory evidence before recommending a refresh:

  • Did CPM rise across other active campaigns?
  • Did the increase begin on the same date?
  • Did ads with unrelated creative experience the same change?
  • Did click and conversion response remain steady?

The wider market average cannot settle this. One aggregate market comparison describes falling overall CPM alongside increases for many individual advertisers. An average auction price is not the price this account pays.

If delivery became more expensive but response held, investigate the auction before blaming the video.

Auction paddles illustrating rising bid costs while click response stays flat

Placement or audience composition changed

Delivery can concentrate without the creative wearing out. Spend shifts toward Reels, or delivery finds a smaller pocket inside a broad audience. Average frequency rises. Aggregate response changes because the audience or placement mix changed.

That does not rule out fatigue. It gives fatigue a competing explanation.

For the running example, I ask Claude to compare the period before and after the CPL increase using placement, platform, age and gender breakdowns, with spend share beside frequency and response metrics where available.

A large shift in placement share matters. So does an audience mix that tilts toward one bracket. Before calling the creative tired, I want to know whether comparable parts of delivery weakened or whether the overall mix simply changed.

Those breakdowns are not a view of each person’s exposure history. Meta’s breakdown reporting includes estimates and may include modeled conversions. It can reveal composition changes; it cannot recover the exposure distribution hidden inside an average.

A frequency rise following a delivery shift needs a mix check before a creative verdict.

Click response held, but post-click conversion weakened

Now suppose outbound CTR holds at 1.1%, CPM stays flat and frequency moves from 2.4 to 2.9. Meanwhile, landing page conversion falls from 8% to 4.5%, and CPL nearly doubles.

The ad still earns the click. The deterioration comes afterward. The page, offer, price, stock status or checkout belongs in the investigation before the creative does.

Tracking problems can produce a similar report. Normal clicks and CPM alongside an overnight conversion drop can point toward a pixel or Conversions API issue rather than fatigue, as this breakdown explains.

I ask Claude three questions, with dates:

  1. Did the destination URL, page, price or form change?
  2. Did reported conversions diverge from Shopify or CRM numbers?
  3. Did the decline arrive suddenly or develop over several weeks?

A gradual slide can fit fatigue. A cliff sends me looking for a sudden change, including broken measurement. Neither shape proves the cause alone.

When CTR holds and conversion rate falls, investigate the destination and measurement before replacing the ad.

Say what the account cannot distinguish

Even a neat timeline has limits. Aggregate reporting cannot show that the same person saw an ad repeatedly and then stopped responding.

Overlapping changes make the comparison harder. If the home-services account raises its budget, shifts placements and edits its landing page in the same fortnight, Claude cannot cleanly assign the CPL increase among those changes. Fluent explanation is not additional evidence.

Timing matters too. Spend records before all attributed conversions arrive, so a fresh reporting window can make performance look worse before it matures.

Small samples add another problem. Two purchases versus four on the same spend halves cost per purchase from $500 to $250. The arithmetic is dramatic. The evidence may still be thin.

I do not want Claude to turn a minimum number of days or conversions into another magic threshold. I want it to state whether the available volume, reporting delay and overlapping changes support a useful comparison.

A cracked magnifying glass over indistinct advertising charts

Competitor ads sit outside that diagnosis. groas collects competitor ads on Google and Meta, their hooks, angles and offers, and the landing pages and funnels behind them. That is useful material for deciding what to test next.

It does not reveal a rival’s spend, targeting or cost per result. A competitor’s long-running ad is not proof that it beats the one under review.

When the account cannot distinguish fatigue from other causes, Claude should say so and propose monitoring or a test, not a confident kill list.

Turn the diagnosis into a brief that can be wrong

Suppose the home-services video does show the full chain against its first two weeks: frequency rises, outbound CTR falls 28%, CPM stays within 5% of baseline, landing page conversion stays flat and CPL rises 41%.

That supports a fatigue hypothesis. It earns a creative test, not a declaration that the case is closed.

My next prompt is specific:

For ads with evidence consistent with fatigue, propose two genuinely different creative concepts using our offer and competitor examples. Show the evidence supporting the diagnosis, the competing explanations and what would make us revise it. Make no changes.

“Genuinely different” means a new hook and proof structure, not the same UGC script shot in a different kitchen. I use the competitor ads and funnels stored in groas for possible angles, not borrowed claims of performance.

The brief also needs a disproof condition. If the old creative’s outbound CTR recovers while the delivery conditions change, the claim that the creative itself wore out needs another look. If replacement ads encounter the same post-click decline, the destination remains unresolved.

A brief without a disproof condition is an opinion with formatting.

Replace deliberately, then apply the model again

Before publishing anything, I require Claude to preview:

  • Exact ad names proposed for pausing.
  • New creative files, headlines and primary text.
  • Destination URL for each ad.
  • Affected ad sets.
  • Budget and audience settings that will remain untouched.

I approve that list in plain English first. I leave autopilot off unless I explicitly want it running. groas can operate around the clock when enabled, with changes logged and undoable, but that does not remove the need to define the task. Nor does undoing a change recover spent budget.

After publishing, I check the action log and evaluate delivery with the same windows and response layers. The replacement has to survive the same scrutiny as the original.

Now apply the model to three ads at frequency 3.2, 3.4 and 3.8:

  • The 3.2 ad: CPM rises 31% across the account while CTR stays steady. I investigate delivery cost, not creative fatigue.
  • The 3.4 ad: CTR holds, but conversion rate falls after a form change. I investigate the form and measurement.
  • The 3.8 ad: outbound CTR falls 28%, CPM stays within 5% of baseline, post-click conversion holds and CPL rises 41%. I propose a replacement test, subject to the reporting limits already checked.

Same frequency band. Three different answers.

The useful outcome is not “Claude found three tired ads.” It is keeping two ads out of an unnecessary replacement queue, investigating a form and testing one credible fatigue hypothesis. Frequency starts the inquiry. It does not get to end it.