# Claude by Anthropic Meta Ads and Creative Strategy

Claude by Anthropic is a Productivity app by Anthropic PBC. This is the plain-text rendition of HyperScale Radar's teardown of its live Meta advertising: the ads that survive 30 or more days of continuous spend, the decoded creative strategy, and what App Store reviewers say.

- Canonical page: https://www.hyperscale.net/apps/claude/6473753684/
- App Store listing: https://apps.apple.com/us/app/id6473753684
- First published: 2026-07-21 · Data refreshed: 2026-07-27
- Source data: public Meta Ad Library, Apple App Store listing and customer reviews

## Key facts

- As of July 27, 2026, Claude by Anthropic runs 503 active ads in Meta's Ad Library across Facebook and Instagram; 36 of them have survived 30 or more days of continuous delivery.
- Its longest-running ad has been live for 52 days.
- The winning set splits 69% video, 28% static, 3% carousel.
- The most common message angle is problem callout, carried by 36% of the winning set.
- The winning ads speak mainly to Late-Night Software Engineer (36% of ads) and Startup Team Lead (22%).
- All winning copy runs in English.
- In App Store reviews, the loudest praise goes to response quality & intelligence (11% of mentions) and the loudest complaint is usage & rate limits (25%).
- The App Store listing holds 4.71739 stars across 211,009 ratings.

## Creative strategy: the dissection

Of 503 active Meta ads for Claude, only 36 survive past 30 days, and this week added 11 new winners while retiring 8. The longest-runner, 52 days, comes from build your own tool, an angle with just 3 ads total, while the real workhorse is a ~5-second, no-people product clip opening 13 of 36 winners with a blunt problem callout.

### Five Seconds of App, Zero People

Problem callout is the account's workhorse: 13 of 36 winning ads (9 x-rayed creatives) run this exact shot, a roughly 5-second video with no presenter at all, just product and app footage on screen. That's a deliberate break from the account's own casting norm: presenters appear on camera in 65% of the 26 x-rayed creatives, meaning this angle is the one place Claude lets the interface do the talking instead of a person.

### Two Runtimes, One Casting Split

Slow thinking transformation (6 ads, 4 creatives) runs long at ~58 seconds behind a female 25-34 presenter, while F1 partnership (5 ads, 4 creatives) runs tight at ~29 seconds behind a male 25-34 presenter, the same two demographics making up 15 of the 26 x-rayed creatives' on-camera casts (9 male 25-34, 6 female 25-34). Runtime bends per angle; who's cast barely does.

### One Ad, One Angle, 52 Days

Build your own tool carries just 3 ads, yet it holds the corpus's single longest-runner at 52 days: the 'three years of email chaos' Gmail story built with Google Apps Script and Claude Opus. Its adjacent personas are thin too, DIY Non-Coder Automator at 2 ads and Eng-Adjacent Automator at 4 ads, proof one execution can carry an angle far past what its ad count would suggest.

### Sixteen Ads Still Say Nothing

Of 13 unique primary texts in the whole corpus, 16 ads run literally '(no body video)', plus 3 '(no body dco)' and 2 '(no body image)', 21 blank-copy ads before counting anything else. Only 5 ads carry the Claude Code 'leveled up' partnership line, and just 2 run the F1 line 'Debuting the Pattern of Thought.' Copy is nearly absent from most of what's winning.

### What Reviewers Say, Ads Don't

Across 2,342 reviews (4.72★ lifetime), usage/rate limits draw 85 mentions and reliability/crashes draw 75, with pricing value close behind at 49. Reviewers describe caps hitting fast 'even on paid Pro/Max tiers' and outages that undercut any reliability claim. None of the 36 winning ads mention limits, crashes, or price at all.

### What HyperScale would test next

- Produce a ~5-second, no-people product-footage video using a pattern interrupt hook (the format behind problem callout's 13 winning ads) for the cross-team adoption angle (3 ads), targeting the Startup Team Lead persona (8 ads, the largest persona pool without its own signature execution).
- Recast the slow thinking transformation angle's ~58-second testimonial format with a male 25-34 presenter, currently cast only in F1 partnership, instead of the female 25-34 lead, testing the account's underused casting split against the Reflective Deep Thinker persona (6 ads).
- Test a side-by-side before/after comparison cut, a mechanic absent from the six hook types logged (pattern interrupt, testimonial, relatable developer pain point, celebrity endorsement, text-based problem statement, philosophical question), built for the build your own tool angle (3 ads, holder of the 52-day longest-runner) targeting the Eng-Adjacent Automator persona (4 ads).
- Test a carousel execution, just 3% of the 26 x-rayed creatives, paired with the testimonial mechanic (tied top at 6) for the creative partner testimonial angle (3 ads), targeting the Creative Partner Seeker persona (3 ads) that has zero carousel presence today.

## Angle mix (share of the winning set)

- problem callout: 36% (13 ads)
- slow thinking transformation: 17% (6 ads)
- f1 partnership: 14% (5 ads)
- build your own tool: 8% (3 ads)
- creative partner testimonial: 8% (3 ads)
- cross-team adoption: 8% (3 ads)
- central tool testimonial: 8% (3 ads)

## Persona coverage (who the winning ads speak to)

- Late-Night Software Engineer: 36% (13 ads)
- Startup Team Lead: 22% (8 ads)
- Reflective Deep Thinker: 17% (6 ads)
- Eng-Adjacent Automator: 11% (4 ads)
- Creative Partner Seeker: 8% (3 ads)
- DIY Non-Coder Automator: 6% (2 ads)

## Language mix

- English: 100% (36 ads)

## Winning-set x-ray: the longest-surviving creatives

Top 25 of 36 analyzed winners, by days live. The full set is on the HTML page.

| Days live | Format | Lang | Angle | Persona | Opening hook | On-screen headline |
| --- | --- | --- | --- | --- | --- | --- |
| 52 | video | en | build your own tool | DIY Non-Coder Automator | What's the difference between this and this? My inbox was three years of complete chaos. | - |
| 52 | video | en | build your own tool | Eng-Adjacent Automator | It's cool like it's Google search, but the problem is it doesn't work while I sleep. | Jeremiah returns! |
| 52 | video | en | build your own tool | DIY Non-Coder Automator | I built a tool that roasts your thesis statement like a brutally honest Oxford professor. | - |
| 47 | video | en | slow thinking transformation | Reflective Deep Thinker | A woman holds a small, black radio with a sticky note labeled 'THE VOICE' and turns the dial, looking confused. | The AI for problem solvers. |
| 47 | video | en | slow thinking transformation | Reflective Deep Thinker | We need to rebel against time by thinking slower. | The AI for problem solvers. |
| 47 | video | en | creative partner testimonial | Creative Partner Seeker | The most important thing I have learned as a creator is to slow down. | The AI for problem solvers. |
| 46 | video | en | slow thinking transformation | Reflective Deep Thinker | Can you hear me? Can you hear me? | The AI for problem solvers. |
| 45 | dco | en | problem callout | Late-Night Software Engineer | Here for the 3am deploy | Here for the 3am deploy |
| 45 | video | en | creative partner testimonial | Creative Partner Seeker | The most important thing I have learned as a creator is to slow down. | The AI for problem solvers. |
| 45 | dco | en | problem callout | Late-Night Software Engineer | Here for the last refactor | Here for the last refactor |
| 45 | video | en | problem callout | Late-Night Software Engineer | Here for Hashing... | Here for Hashing... |
| 45 | video | en | problem callout | Late-Night Software Engineer | Here for the merge conflict | Here for the merge conflict |
| 44 | image | en | problem callout | Late-Night Software Engineer | Here for the last refactor | Here for the last refactor |
| 44 | video | en | slow thinking transformation | Reflective Deep Thinker | A woman holds a small, black radio with a sticky note labeled 'THE VOICE' and turns the dial, looking confused. | The AI for problem solvers. |
| 43 | image | en | problem callout | Late-Night Software Engineer | Here for the 3am deploy | Here for the 3am deploy |
| 43 | video | en | slow thinking transformation | Reflective Deep Thinker | We need to rebel against time by thinking slower. | The AI for problem solvers. |
| 43 | image | en | problem callout | Late-Night Software Engineer | Here for the last refactor | Here for the last refactor |
| 42 | video | en | slow thinking transformation | Reflective Deep Thinker | We need to rebel against time by thinking slower. | The AI for problem solvers. |
| 41 | video | en | creative partner testimonial | Creative Partner Seeker | The most important thing I have learned as a creator is to slow down. | The AI for problem solvers. |
| 40 | video | en | problem callout | Late-Night Software Engineer | Here for Hashing... | Here for Hashing... |
| 40 | video | en | problem callout | Late-Night Software Engineer | Here for Hashing... | Here for Hashing... |
| 39 | video | en | cross-team adoption | Eng-Adjacent Automator | The engineers adopt Claude Code first, and then the eng adjacent roles look over their shoulder and they're like, whoa, this thing is very powerful. Let me try | Meet your new thinking partner |
| 39 | video | en | central tool testimonial | Startup Team Lead | For us Claude is just at the center of everything. Whenever I have a question, I ask Claude. | Meet your new thinking partner |
| 39 | video | en | cross-team adoption | Eng-Adjacent Automator | Essentially Claude writes the code. | Meet your new thinking partner |
| 38 | video | en | central tool testimonial | Startup Team Lead | For us Claude is just at the center of everything. | Meet your new thinking partner |

## Customer voice (2224 App Store reviews analyzed)

Objections the ads must pre-empt:

- Rate/usage limits often hit within minutes, even on paid Pro/Max plans
- Server outages and crashes make the app unusable for hours at a time
- Accounts can be suspended or age-verified/banned without warning or support response
- Pricing is seen as high relative to the limited number of messages allowed
- Support is slow or automated, with users waiting days for a human reply

Review-theme mix (praise / complaint / mixed):

- usage & rate limits (complaint, 25% of mentions): Even paying customers hit message caps within minutes and must wait hours to continue, which feels stingy for the price.
- reliability & crashes (complaint, 22% of mentions): Users report frequent crashes, glitches, errors sending messages, and server downtime that breaks workflows.
- pricing value (complaint, 14% of mentions): Many feel they're paying premium prices ($20-$200/mo) for far less usage than expected.
- response quality & intelligence (praise, 11% of mentions): Praised for clear, concise, well-thought-out, and accurate answers that feel like talking to a smart partner.
- competitor comparison (mixed, 10% of mentions): Many switched from ChatGPT, Gemini, or Grok, saying Claude is smarter, more honest, and less 'parasitic.'
- account bans & support (complaint, 7% of mentions): Users describe sudden account suspensions or bans with no explanation and unresponsive support.
- productivity impact (praise, 7% of mentions): Customers say Claude multiplies their output, handling 80% of workload so they can focus on strategy.
- voice input & app bugs (complaint, 5% of mentions): Voice transcription and input bar issues frustrate users trying to dictate longer prompts.

> "Claude changed my career by multiplying my hourly rate by 20x and empowering me to accomplish previously impossible tasks" (5 stars, App Store review, CA)

> "I'm able to outsource 80% of my projects to an intelligent thought partner" (5 stars, App Store review, AU)

> "I'm able to outsource 80% of my projects to an intelligent thought partner" (5 stars, App Store review, AU)

> "Claude changed my career by multiplying my hourly rate by 20x and empowering me to accomplish previously impossible tasks" (5 stars, App Store review, CA)

> "I basically have a super intelligent PA 24 hours of the day!" (5 stars, App Store review, AU)

> "realizar el 80% del trabajo pesado u operativo y tú dedicarte a la estrategia" (5 stars, App Store review, CO)

> "won me one tax review and a 34k deductible, and a small court case" (5 stars, App Store review, CA)

> "realizar el 80% del trabajo pesado u operativo y tú dedicarte a la estrategia" (5 stars, App Store review, CO)

## Buyer personas (read from the winning set)

- Late-Night Software Engineer. pain: stuck debugging alone at 3am, drowning in refactors and merge conflicts. desire: a coding partner that ships fixes and handles the grind at any hour. awareness: solution-aware: already trials AI coding tools daily and compares them on real tasks.
- Eng-Adjacent Automator. pain: manual workflows pile up while technical teammates move faster, and coding feels out of reach. desire: build agents and automations for own work without becoming a developer. awareness: problem-aware: watches engineers on the team already win with Claude Code and wants the same edge.
- Startup Team Lead. pain: problem solving is fragmented across people and apps, slowing decisions down. desire: a shared AI collaborator that speeds up how the team thinks and builds together. awareness: solution-aware: hunting for one AI the whole team can rally around instead of scattered tools.
- Reflective Deep Thinker. pain: feels rushed, unheard, and shallow in her own work, chasing output instead of clarity. desire: an AI that helps her slow down and actually think a problem through. awareness: pain-aware: exhausted by tools that reward speed over real thinking.
- Creative Partner Seeker. pain: burned out from fast content churn with no one to challenge or sharpen the work. desire: a creative partner that pushes back and raises the quality of the final piece. awareness: solution-aware: already making content and wants a collaborator, not a shortcut generator.
- DIY Non-Coder Automator. pain: inbox and workflow chaos that off-the-shelf apps never fixed, no budget for a developer. desire: build her own custom system herself with zero coding background. awareness: problem-aware: buried in years of manual repetitive tasks and unaware AI can build tools without code.

## FAQ

### How many Claude ads are running on Meta, and how many are actually winning?

503 active ads sit in Meta's Ad Library right now; only 36 have survived past the 30-day mark Radar treats as a winner. This week added 11 new winners and retired 8 since 2026-07-21.

### What angle and format is working best for Claude right now?

Problem callout leads with 13 of 36 winning ads (9 x-rayed creatives): a ~5-second, no-people video of pure product/app footage. Slow thinking transformation (6 ads, ~58s, female-led) and F1 partnership (5 ads, ~29s, male-led) trail behind.

### What languages and platforms do the winning Claude ads run on?

All 36 winning ads run in English only, none in other languages yet. They span video, dco, image, and carousel formats across Facebook, Instagram, Messenger, Threads, and Audience Network.

### What do Claude reviewers complain about that the ads don't address?

Across 2,342 reviews (4.72★ lifetime), usage/rate limits draw 85 mentions and reliability/crashes draw 75, with pricing value at 49. None of the 36 winning ads mention limits, crashes, or price at all.

## The market around it

- Perplexity (369+ active ads, Productivity). Meta Ad Library: https://www.facebook.com/ads/library/?active_status=active&view_all_page_id=109760398882363
- ElevenLabs (600+ active ads, Photo & Video). Meta Ad Library: https://www.facebook.com/ads/library/?active_status=active&view_all_page_id=108361132071329

## About this data

Independent editorial analysis by HyperScale (https://www.hyperscale.net), the AI performance-creative team for mobile apps. Not affiliated with or endorsed by Anthropic PBC. Ad creatives, start dates, and active status come from Meta's public Ad Library; listing data and verbatim review quotes come from Apple's public App Store. When citing, name "HyperScale Radar" and link https://www.hyperscale.net/apps/claude/6473753684/

All teardowns: https://www.hyperscale.net/apps/ · Plain-text index: https://www.hyperscale.net/llms.txt