Claude for Meta Ads: Prompts for Fatigue, Frequency, Audiences and Copy

Most Meta ads work is pattern-finding across many small numbers: which ad's click-through rate broke trend, which ad set has been in learning for two weeks, which angle keeps winning. Claude is good at that, and good at the writing that follows. This guide is the Meta half of a pair; the search half is Claude for Google Ads.

What to give Claude

Include your target cost per result or target ROAS in every prompt. Without a target, Claude can rank ads, but it cannot tell you which ones are failing.

Creative fatigue

Fatigue shows up per ad before it shows up per campaign. The prompt asks Claude to find the day each ad broke trend, not just the ads that are bad today:

Here is a daily ad-level export for the last 30 days: spend, impressions,
frequency, CTR, CPM, results and cost per result.
Target cost per result: $[X].

For each ad with more than $[Y] spend:
1. Find the first day CTR fell 25% or more below its own first-7-day average
   and stayed there for 3+ days.
2. Report frequency on that day.
3. Classify: healthy, fatiguing, or fatigued.
Return a table sorted by spend, then list the 3 ads to replace first and why.

A typical result for an ecommerce account:

The thresholds in the prompt (a 25% drop held for three days) are a starting point. Change them to match how fast your account usually burns through creative.

Frequency

Here is ad set data by day with reach, impressions and frequency.
For each ad set:
- Weekly frequency trend and the audience size if shown.
- Whether cost per result rises as frequency rises (state the correlation
  plainly and say if the sample is too small to tell).
- A recommendation: new creative, broader audience, lower budget, or leave it.

The useful part of this answer is the "too small to tell" line. Claude will happily draw a curve through eight data points; asking it to say when the sample is thin keeps the recommendation honest.

Audiences and ad sets

Two prompts cover most audience questions. The first finds ad sets that compete with each other or never left learning:

Here are my active ad sets with targeting, budget, optimization event,
learning status and results for 14 days.
1. Flag ad sets whose targeting overlaps heavily (same interests, same
   lookalike source, or same broad audience in the same country).
2. Flag ad sets still in learning or learning limited, and suggest a
   consolidation that gives each remaining ad set more budget per result.
3. Show the before and after structure as two short lists.

The second plans new audiences from what you already own:

Our sources: [purchasers list size], [website visitors 180 days],
[engaged Instagram followers]. Our average order value is $[X].
Propose a test plan of 3 ad sets: which custom audiences and lookalikes
to build, the lookalike percentage for each and why, and the exclusions
each ad set needs so they do not bid against each other.

Reading Advantage+ campaigns

Advantage+ hands placement and audience decisions to Meta, so the reporting question changes from "what did I target" to "where did Meta spend it". Export the placement and platform breakdown and ask:

This is an Advantage+ campaign with a placement and platform breakdown
for 30 days.
1. Where did spend go (Feed, Reels, Stories, other) and what did each
   placement return per result?
2. Which creatives carried each placement? Do vertical videos dominate Reels?
3. Is anything spending at 2x target cost per result with no improvement
   over the period?
4. What creative is missing, by format and aspect ratio, given where Meta
   is choosing to spend?

The fourth question is the one that pays. If most spend lands in Reels and you have two vertical videos, the fix is not a targeting change. It is more vertical creative.

Ad copy and hooks

Paste your three best ads with their results, then ask for variations by angle. Claude writes better copy against proven winners than from a blank brief.

Here are our 3 best Meta ads by cost per purchase, with primary text,
headline and result. Product facts and reviews are pasted below.
Write 9 new ads: 3 on the strongest angle above, 3 on an angle we have not
tested (pick from objection handling, social proof, comparison), and 3
UGC-style scripts with a spoken first line under 8 words.
Primary text: lead with the hook in the first line. Headline: 40 characters
or fewer. Only use claims that appear in the pasted facts or reviews.

Sample hooks for a sleep supplement, all from customer reviews:

• I stopped waking up at 3am.
• My sleep score went from 61 to 84.
• No grogginess. That was the surprise.

Look at these competitor ads [paste ad text or descriptions].
List the angles they repeat, the offers they lead with, and one angle
none of them uses that our product facts support.

For a deeper look at how ads are structured across the funnel, the breakdown of Anthropic's own Claude ads is a useful reference on one advertiser's choices.

Where Claude stops

• Exports go stale. A fatigue read from last Tuesday's export is about last Tuesday.
• It cannot see the creative. From an export Claude sees names and numbers, not the video. Name ads descriptively, or describe each one, or the analysis stays generic.
• Attribution is Meta's. Claude reports what Ads Manager reports. If your pixel or Conversions API is broken, every answer inherits the error.
• Nothing changes until someone changes it. Without a writing connector, every recommendation becomes manual work in Ads Manager.

For a full account review, pair these prompts with the Meta ads audit checklist. More prompts by job, including reporting, are in Claude marketing prompts.

Where Sprites fits

Sprites is an official MCP connector for Claude. Connected, it reads the Meta account itself: ROAS, CPA, frequency and placement breakdowns, per-creative performance, learning status, pixel and Conversions API health, and public competitor ads from the Meta ad library. New ads, ad sets, custom audiences and lookalikes, budget moves and status changes arrive in the chat as approval cards, with the creative previewed before anything ships.

See the Meta Ads MCP page for setup in Claude. Plans start at $99 a month, with a 7-day trial for $1.

Claude for Meta ads questions

Can Claude run Facebook ads?

Claude can analyze Facebook and Instagram ad data and write ads, audiences and test plans. To create or change ads in the account it needs a connector with write access. Sprites connects Claude to Meta and shows every change as an approval card before it applies.

How do I use Claude for Meta ads?

Export a daily, ad-level report from Ads Manager, or connect Claude to the account through an MCP connector. Then ask specific questions with a stated target cost per result: which ads are fatiguing, which ad sets overlap, where Advantage+ spent the budget, and what to write next.

Can Claude detect creative fatigue?

Yes, with daily ad-level data. Ask it to find the first day each ad's click-through rate fell well below its own early average and stayed there, and to report frequency on that day. Campaign totals hide fatigue, so the per-ad breakdown is required.

Can Claude write Facebook ad copy?

Yes. It writes better copy when you paste your best-performing ads with their results and the product facts it may use. Ask for variations by angle, a hook in the first line of primary text, and no claims that are not in the facts you gave it.

Can Claude analyze Advantage+ campaigns?

Yes, from a placement and platform breakdown. Claude can show where Meta spent the budget, what each placement returned and which creative carried it. The most useful output is the missing creative: formats and aspect ratios for the placements Meta favors.

Is there an official Meta Ads connector for Claude?

Meta publishes an Ads MCP server that Claude can connect to, and third-party connectors such as Sprites also connect Claude to Meta ad accounts. Check what each one can read and change, and whether writes wait for your approval, before you connect it.