SYSTEM · ONLINE CD.MODEL · CLAUDE_OPUS_4.7 BUILD · 0.2.β --:--:-- UTC
POWERED BY ANTHROPIC'S CLAUDE API · TUNED BY A SENIOR CREATIVE DIRECTOR

Score your ads
before you spend.

A senior creative director's read on every ad you upload — running on Anthropic's Claude API, tuned with 30+ years of agency craft. Honest, opinionated notes. No hedging. No corporate fluff. Independent product. Not affiliated with or endorsed by Anthropic.

// HOW_IT_WORKS

Three steps. About fifteen seconds.

Drop in any ad creative and get a senior-CD's read on whether it'll do its job — before you spend a dollar of media on it.

01

Upload your ad

JPG, PNG, GIF, or WebP up to 5 MB. Pick the channel from the dropdown — feed, billboard, print, OOH — so the work is judged on the right rules. Or leave it on Auto-detect.

02

AI tuned by a senior CD

This is not a human review. We call Anthropic's Claude API (Opus 4.7 with vision) and feed it a rubric a senior creative director with 30+ years across every channel built into the system. The AI does the analysis; the CD set the standards. We are not affiliated with Anthropic — we use their commercial API like any other developer.

03

Get notes you can act on

One overall score, five category scores, and a short list of specific, opinionated fixes. Cited to actual elements in your creative — no generic feedback, no hedging.

// WHAT_WE_SCORE

Five categories. One overall verdict.

Weighted toward the things that actually move conversion. CTA clarity matters more than typography contrast. Policy risk matters most of all — rejected ads spend zero.

20%

Readability

Does the eye know where to land in the time the medium gives you? Is there one big idea, clearly the hero — or is the audience working to decode it?

10%

Contrast

WCAG-style contrast between text and background. Headline and CTA need to clear AA at intended size — no hiding behind text shadows.

30%

Policy compliance

Meta and Google ad-policy risk. Unsubstantiated superlatives, miracle claims, personal attributes, fake UI, sensitive categories without disclaimers. The biggest weight, because rejected ads can't sell.

25%

CTA clarity

One specific, action-verb CTA — visible, contrasting, unambiguous. "Learn More" is fine, never great. Multiple competing CTAs split intent and tank conversion.

15%

Brand safety

Native to its channel, professionally produced, distinctive enough to escape banner-blindness. Stock-photo soup and template-y feeds get penalized here.

// YOUR_WORK · YOUR_DATA

We don't keep your work. The AI never trains on it.

Performance ads are competitive. Here's exactly what happens to anything you upload — written plainly, no legalese.

Your creative is not stored.

Images you upload are sent to Anthropic for analysis and discarded the moment the response comes back. Nothing is written to disk. Nothing is kept on our servers. Refresh the page and the file is gone.

Anthropic does not train on it.

Per Anthropic's commercial API terms, your ad is processed only to answer the request you made. It is never used to train or improve their models. Zero retention beyond the call itself.

We don't sell or share your data.

We collect an email only when you choose to (Deep-dive unlock or Human review request) — used to deliver what you asked for and any follow-ups you opt into. Never sold. Never shared. Unsubscribe any time and your record is removed.

Privacy or data questions → legal@theadbench.ai