Claude Fable 5.2 for Meta Ads: The Free Audit, Competitor Spy & Winning Angle Setup

If you commented "FABLE", this is the setup we promised.

Claude Fable 5.2 is the first model where "go audit my ad account" gives you an answer a senior media buyer would sign off on. That only happens if you give it the right inputs and the right instructions, though. Ask it to "make me an ad" and you'll get something generic.

This guide gives you the three workflows from the post, ready to copy and paste:

1. The Ad Account Audit: find what's wasting budget and exactly what to kill, fix and scale this week.
2. The Competitor Spy: pull your competitors' longest-running Meta ads and break down why they work.
3. The Winning Angle Finder: mine what your buyers actually say and turn it into ranked angles, hooks and briefs.

Each part ends with the step most guides skip: turning the insight into new ads you can launch.

TL;DR

• Use Claude Fable 5.2 at "High" effort inside a Claude Project with a one-page brand brief.
• Audit: export 30 days of ad-level data from Ads Manager, then run Prompt 1. You get a Kill / Fix / Scale table and a 5-step action plan.
• Spy: pull 10–20 of each competitor's longest-running ads from the Meta Ad Library, then run Prompt 2. You get their hooks, offers, angles and the gaps they're leaving open.
• Angles: feed in reviews, Reddit threads, ad comments and support tickets, then run Prompts 3A–3C. You get 5 ranked angles, 15 hooks and 3 creative briefs.
• Build: turn the briefs into finished static, video and UGC ads in , launch, and repeat weekly.

Anthropic released Claude Fable 5.2 in November 2026. Most of the coverage is about coding and science. These are the improvements that matter if you run ads:

• Reading charts and dashboards — The numbers (Anthropic-reported): Chartography benchmark: 44.8% → 73.4% vs. Fable 5.1 · What it means for your ads: It can read Ads Manager screenshots and performance charts properly, not just CSVs
• Using a computer / browser on its own — The numbers (Anthropic-reported): OSWorld 2.1: 83.5% → 89.3% · What it means for your ads: With a browser connected, it can browse the Meta Ad Library itself
• Overall reasoning — The numbers (Anthropic-reported): #1 on the Artificial Analysis Intelligence Index at launch · What it means for your ads: Better at connecting "CTR dropped" + "frequency rose" + "CPM spiked" into one diagnosis
• Price and speed — The numbers (Anthropic-reported): $8 / $40 per million input/output tokens (down from $10 / $50), up to 30% faster · What it means for your ads: Cheaper to run full audits every week

Be honest with yourself about when you need it. Anthropic still recommends Claude Opus 5.5 for most everyday work. Fable 5.2 earns its keep on the heavy-thinking steps: the audit diagnosis, synthesising dozens of competitor ads, and ranking angles. For rewriting 50 hook variations, Opus 5.5 or Sonnet 5.5 is cheaper and just as good. There's more on this under Cost and model tips .

Step 1: Get access and pick the right setting

• Fable 5.2 is available on paid Claude plans (Pro and above). It isn't on the free plan.
• In the model picker, choose Claude Fable 5.2.
• If you can set an effort or thinking level, use High. Independent testing found High and Extra-High score almost the same as Max while using far fewer tokens. Max is the most expensive way to get a very similar answer.

Step 2: Create a Claude Project called "Meta Ads Agent"

A Project lets Claude remember your brand across every chat, so you don't re-explain your business each time. Create one, then paste the brand brief below into the Project's instructions or knowledge.

Fill this in once. It's the single biggest quality lever in this whole guide:

Step 3: Collect your inputs

You'll need these as you go (each part explains how to get them):

• Audit — Inputs: Ads Manager export (ad level, last 30 days + previous 30 days) · Time: 5 min
• Spy — Inputs: Competitor names, Meta Ad Library access, optional browser connection · Time: 10 min
• Angles — Inputs: Reviews export, 3–5 Reddit threads, ad comments, support tickets · Time: 15 min

Step 4 (optional): Let Claude browse the Ad Library itself

The Meta Ad Library is an interactive page, so a plain "search the web" can't read it properly. You have two options:

• Option A, connect a browser: use Claude in Chrome (Claude's browser extension). Fable 5.2 can then open the Ad Library, search competitors and read the ads itself.
• Option B, screenshots: take screenshots of the ads yourself and paste them in. It's slower but works everywhere, and Fable 5.2's better image reading makes this reliable.

Goal: in one pass, find which ads are wasting budget, which are tired, which have a fixable weakness, and which deserve more money, plus a ranked action list for this week.

How to export the right data from Ads Manager

1. Open Meta Ads Manager and go to the Ads tab (ad level, not campaign level, because creative problems hide inside campaign averages).
2. Set the date range to the last 30 days. Turn on Compare and choose the previous 30 days.
3. Click Columns → Customize columns and add the following. Save the set as "Creative Audit" so next week takes 10 seconds.

• Campaign name, Ad set name, Ad name — Why the audit needs it: Structure and naming
• Delivery, Ad creation date / Starts — Why the audit needs it: Age of the ad, for spotting fatigue
• Amount spent — Why the audit needs it: Where the money goes
• Impressions, Reach, Frequency — Why the audit needs it: Saturation and fatigue
• CPM (cost per 1,000 impressions) — Why the audit needs it: Is Meta charging more to show it?
• CTR (link click-through rate), CPC (cost per link click) — Why the audit needs it: Is the ad earning the click?
• 3-second video plays, ThruPlays, Video plays at 25% / 50% / 75% / 95% — Why the audit needs it: Hook and hold for video ads
• Landing page views — Why the audit needs it: Do clicks actually load the page?
• Adds to cart, Checkouts initiated, Purchases (or your main result) — Why the audit needs it: Funnel drop-off
• Cost per result / Cost per purchase — Why the audit needs it: Efficiency
• Purchase ROAS (if e-commerce) — Why the audit needs it: Return

1. Click Reports → Export table data and download as CSV or XLSX.
2. Optional but powerful: screenshot your account-level Performance chart for the last 60 days. Fable 5.2 reads charts well, and the trend line often explains what the table can't.

Ads Manager exports contain performance numbers, not customer personal data. Remove anything else your company treats as confidential before uploading.

Prompt 1: The Ad Account Audit

Upload the export (and screenshot, if you have one), then paste:

Follow-up prompts that make the audit sharper

Show it the actual ads. Numbers tell you that an ad is failing; the creative tells you why.

Check your tracking. If the data check flagged anything, run this before acting on any verdicts:

Ask it to argue with itself. This is cheap insurance on big budget decisions:

How to read the results (and not wreck your account)

• Don't kill everything at once. Turn off the clearest losers, fix the rest and give replacements time to run.
• Respect the learning phase. Meta's delivery system needs roughly 50 optimization events per ad set within 7 days to stabilise. Big edits restart it.
• Fix the right stage. If the audit says "Weak CONVERT", a new ad won't save you; look at the landing page or offer.
• "Not enough data" is an answer. Leave those ads running until they've spent enough to judge.

🔧 Next step: replace the ads you just killed

Your audit gives you a list of FIX ads with briefs ("same offer, new hook", "same product, more proof up front") and a gap where every KILL ad used to be. That's where most audits stall, because someone has to make the new creative.

In HeyOz , paste your store URL and it pulls your products, images and brand identity into a workspace. Drop in the brief from your audit and generate fresh static, video and UGC-style ads. Need several hooks on the same body? Generate the variations side by side and send the best into your next test.

Turn your audit into new ads → HeyOz

Goal: find the ads your competitors keep paying for month after month, then understand exactly why they work: the hook, offer, angle, format and proof. Use that to make better ads, not copies.

Why "longest running" is the signal to look for

Meta doesn't publicly show how much a brand spends on most ads (outside political and some EU transparency data). It does show when an ad started running and whether it's still active. Brands rarely keep paying for an ad that loses money, so an ad running for 60, 90 or 150+ days is usually one that works.

Two more clues to look for:

• "This ad has multiple versions" or the same creative appearing many times usually means they're testing variations of something that works, or scaling it.
• The same concept in several formats (static, video, carousel) means it's a proven angle they're squeezing.

It isn't perfect (some brands leave ads on by accident), which is why you look for patterns across many ads, not one.

Step 1: Pick 3–5 competitors

Use the competitors in your brand brief. If you're not sure who competes for your buyers on Meta, ask Fable:

Step 2: Pull their ads from the Meta Ad Library

1. Go to the Meta Ad Library (facebook.com/ads/library).
2. Set country to where you advertise and ad category to All ads.
3. Search the competitor's brand name and select their Page from the dropdown, so you only see their ads.
4. Filter to Active ads and, if available, the platforms you run on (Facebook, Instagram).
5. For each ad, note the "Started running on" date. Sort your attention toward the oldest active ones.

Option A: let Fable do it (with Claude in Chrome connected):

Option B: screenshots. Screenshot the 10–20 longest-running active ads per competitor, including the start date, and upload them in batches.

Prompt 2: The Competitor Spy breakdown

A note on doing this ethically and safely: use competitors' ads to learn what works: structures, formats, angles. Don't copy their footage, wording, testimonials or trademarks, and don't make claims about your product that you can't prove. It's the right thing to do, and it's also what keeps your ads approved.

🧬 Next step: clone the structure, keep your brand

Prompt 2 gives you the five competitor ads worth learning from. With HeyOz Ad Cloning you upload a reference ad (image or video), and HeyOz recreates its structure, pacing and format for your product, with your branding: same proven skeleton, entirely your ad.

Want this whole loop run for you? The HeyOz Agent reviews your competitors' ads, picks out the top runners, extracts the principles behind them, and generates image and video ads built on those principles.

Clone a winning structure → HeyOz

Goal: stop guessing what to say in your ads. Find the problems, desires and objections your buyers mention most, in their own words, rank them, and turn the best ones into hooks and creative briefs.

An angle is the core reason someone should care: "saves 20 minutes every morning", "finally doesn't irritate sensitive skin", "the gift that doesn't look last-minute". The same product can be sold through many angles, and the angle usually moves results more than visuals or editing do.

Step 1: Gather what your buyers actually say

The more of these you include, the better. Aim for at least 100 reviews or comments in total.

• Your product reviews — How to get it: Export as CSV from your review app or store admin · What it's best for: Real outcomes and the words buyers use
• Competitor reviews (3-star ones especially) — How to get it: Copy from their site or marketplace listings · What it's best for: What buyers wish existed: your gaps
• Reddit threads — How to get it: Search "[your category] reddit", copy 3–5 relevant threads · What it's best for: Honest frustrations, objections, comparisons
• Ad comments — How to get it: Copy from your own (and competitors') Facebook/Instagram ads · What it's best for: Objections and questions at the moment of decision
• Support tickets / DMs / emails — How to get it: Export or paste the last 50–100 · What it's best for: Confusion, hesitation, pre-purchase doubts
• Post-purchase survey (if you have one) — How to get it: Export "why did you buy?" answers · What it's best for: The real trigger for buying

Remove customer names, emails and order numbers before uploading. You only need the words.

Prompt 3A: Mine the voice of the customer

Prompt 3B: Rank the angles

Prompt 3C: Turn angles into hooks and briefs

🎬 Next step: 15 hooks and 3 briefs → finished ads in minutes

You now have exactly what a creative team needs: ranked angles, tested-style hooks and briefs for a static, a UGC video and a demo. Producing them is usually the slow, expensive part.

In HeyOz , paste your store URL, pick the format and drop in your brief:

• Static ads for Brief A, in multiple sizes and variations
• UGC-style videos with AI creators for Brief B, using your hooks as the opening line
• Product and demo videos for Brief C

Short on ideas for the visual style? Start from a HeyOz Template . These are pre-made formats based on trending, high-performing ad styles, rebuilt around a single image of your product.

Generate your test batch → HeyOz

One audit helps once. The brands that win on Meta run this loop every week, because creative fatigue never stops.

• Monday — What to do: Export last 7 days (plus the 30-day view) and run Prompt 1. Kill, fix and reallocate. · Time: 20 min
• Tuesday — What to do: Re-check the Ad Library for competitors' new ads and anything that just crossed 30 days. Run Prompt 2 on what's new. · Time: 20 min
• Wednesday — What to do: Add the week's new reviews, comments and tickets. Re-run 3B and 3C for fresh hooks and briefs. · Time: 20 min
• Thursday — What to do: Produce the new batch in HeyOz: replacements for killed ads, variations of winners, 1–2 new angles. · Time: 30–60 min
• Friday — What to do: Launch the tests. One angle per ad set, consistent budgets. · Time: 20 min

Give your agent a memory

Create a document in your Claude Project called "Winners Log" and update it every Monday:

After a month, your audits stop being generic. Fable will know your account's history, which angles reliably work, and how long your ads typically last before fatiguing.

• Use Fable 5.2 where thinking matters: the audit (Prompt 1), the competitor synthesis (Prompt 2) and angle ranking (Prompt 3B).
• Use Opus 5.5 or Sonnet 5.5 for volume: rewriting hooks, producing 30 headline variations, reformatting briefs. They're cheaper and just as good for this.
• High effort is the sweet spot. At launch, independent testing measured Fable 5.2's average cost per task at max effort at roughly 2.6× that of Opus 5.5, while High and Extra-High scored nearly the same as Max.
• API users: Fable 5.2 is $8 per million input tokens and $40 per million output tokens. A full weekly audit on a typical export costs a small fraction of what one wasted ad day costs.
• Keep inputs lean. Export only the columns listed above, and trim review dumps to the last 6–12 months.

1. Auditing at campaign level. Creative problems hide in averages. Always export at ad level.
2. Trusting the audit when tracking is broken. If purchases don't match reality, every verdict is wrong. Run the data check first, every time.
3. Killing ads too early. An ad with $15 spent and no purchases hasn't failed yet. Let it reach a meaningful spend relative to your target CPA.
4. Copying competitor ads. Adapt structures and angles. Copying wording, footage or claims risks rejections and damages your brand.
5. Inventing proof. Only use claims, numbers and testimonials you can back up. Meta reviews ads, and so do your customers.
6. Testing five things at once. One angle per ad set, so each result teaches you something.
7. Stopping at the insight. An audit that says "you need new hooks" is worthless until the new ads are live. Insight without production is how accounts stall.

Is Claude Fable 5.2 free?

No. Fable 5.2 is available on paid Claude plans (Pro and above) and through the API at $8 / $40 per million input/output tokens. If you're on the free plan, you can run the same prompts with whichever model you have access to. Expect less sharp audits, especially when reading charts and screenshots.

Can Claude connect directly to my Meta Ads account?

The simplest and safest method is the one in this guide: export your data from Ads Manager and upload it. Some third-party connectors can link Claude to Meta's API directly. If you use one, give it read-only access and review any changes before they're made.

Can Claude browse the Meta Ad Library?

Yes, if you connect a browser (for example, Claude in Chrome). Fable 5.2 is much better at operating a browser on its own than earlier models. Without a browser connection, take screenshots of the ads and upload them, which works just as well for the analysis.

Is it legal to study competitors' ads?

The Meta Ad Library is public, and it exists for transparency. Studying it is standard practice. What you shouldn't do is copy competitors' creative, wording, testimonials or trademarks. Learn the structure, then make it your own.

Fable 5.2 vs Opus 5.5 for Meta ads: which should I use?

Use Fable 5.2 for heavy analysis: audits, synthesising dozens of competitor ads, ranking angles. Use Opus 5.5 (Anthropic's recommended everyday model) or Sonnet 5.5 for high-volume writing like hook and headline variations.

How often should I run the audit?

Weekly for most accounts. If you spend heavily or launch new creative every few days, run a lighter version twice a week. Keep a Winners Log so each audit builds on the last.

What's the difference between using Claude and using HeyOz?

They do different jobs and work best together. Claude is the strategist: it analyses your data, your competitors and your customers, and tells you what to make. HeyOz is the production studio: it turns your store URL and briefs into finished static, video and UGC-style ads, clones proven ad structures for your brand, and its Agent can run the competitor-research-to-creative loop for you.

You now have everything a senior media buyer would give you: a ranked fix list for your account, a breakdown of what's working for your competitors, and five angles backed by your customers' own words.

The only thing left is the part that usually takes a week and a production budget: making the ads.

Paste your store URL into HeyOz and turn this week's briefs into a full test batch of static, video and UGC ads in minutes.

Start creating with HeyOz →

Sources: benchmark, pricing and availability details are from Anthropic's Claude Fable 5.2 launch (November 2026), as reported by The Decoder , InfoWorld and MindStudio . Benchmarks are vendor-reported unless noted. The kill/fix/scale thresholds in this guide are starting points; adjust them to your account, margins and goals.