Build an AI ad copy generator web app with these features:

1. Input modes: paste a landing page URL (the app fetches and summarizes the page to extract the offer, audience, and proof points) or paste a raw product description. Also collect offer details: discount, deadline, target audience, and main objection to overcome.

2. Platform-native output: generate correctly formatted copy per platform. Google Ads: 15 headlines (max 30 characters) and 4 descriptions (max 90 characters). Meta Ads: primary text (short and long versions), headline (max 40 characters), and CTA suggestions. LinkedIn and TikTok variants as additional modes.

3. Campaign goal modes: direct response, brand awareness, retargeting, and lead gen, each changing the copy structure (pain-first, proof-first, offer-first).

4. A/B variant sets: generate 3 distinct angles per generation (problem-solution, social proof, urgency) so users get genuinely different tests, not paraphrases. Tag each variant with its angle.

5. Brand voice: save a voice profile (tone, banned words, must-include phrases, brand facts) and apply it to every generation. One-click tone presets: authoritative, playful, luxury, founder-to-founder.

6. Policy checker: flag risky claims (guarantees, health promises, 'you' targeting quirks per platform) before the user wastes an ad review on a rejection.

7. Character counters and live truncation preview showing exactly how each line renders on desktop and mobile placements, including how Google truncates descriptions.

8. Performance feedback loop: users paste or connect final results (CTR, conversions per variant) and the app ranks winning angles, then biases future generations toward what worked for that brand voice.

9. Export: copy per line, CSV download, and Google Ads Editor compatible format so variants can be imported directly.

10. History: every generation saved per project with search, so teams can reuse and remix past winners.