Changelog

Three features that make findings actionable and shareable.

• Insight cards (finding → why → fix → confidence): every analysis finding is now structured as a brief — not just "weak hook" but what's causing it, exactly what to change, and how confident we are. Severity-coded and scannable in seconds.
• Portfolio pulse: analyse 3+ ads and the portfolio view now surfaces your most consistent weakness across the library — the one pattern appearing in 60%+ of your ads. Catches what single-ad analysis misses.
• Share to X / LinkedIn: every analysis report now has one-click social sharing with your score, hook rating, and biggest lever pre-populated. The brief includes a link back to your result.
• Cost tracking: each analysis now logs Claude API token usage so you can see the actual AI cost of your reports in the account page.

Analysis results now stream to the browser as each stage completes — you see the hook score, acoustic fingerprint, and misalignment list the moment they're ready, rather than waiting for the full pipeline. Typical time-to-first-insight dropped from ~90s to ~20s.

• Partial result SSE stream: each pipeline stage writes its output immediately; the browser polls and renders progressively
• Regression significance: benchmark comparisons now show statistical significance (p-value) so you know whether a percentile rank is signal or noise

Three changes that move you from a score to a decision faster.

• Benchmarks on every report: each ad's hook, structure, and Voice IQ now show how they rank against your own library — and, once there's enough data, against other ads in your industry. A 7/10 hook means more when you know it's your top 20%.
• Your creative signature: the dashboard now shows the signal you're consistently strongest at and the one holding you back — with a clear next test. If structure is your gap, that's where the next iteration should go, not the hook that's already working.
• Faster analysis: the deep-analysis stage now runs its passes in parallel, cutting the time to your report roughly in half.
• Sharper recommendations: the proprietary metrics (Persuasion Half-Life, Voice Trust Index, Attention Stability, Hook Decay) now tell you what to change, not just what they measured.

Every analysis now includes a cross-signal alignment score that measures whether the acoustic, visual, and caption layers are working together or against each other.

• Silent viewer score (0–100): How well your ad works with sound off — the default for 85% of social video viewers
• Dual-channel CTA detection: Does the call to action appear both spoken and on screen? If not, you're losing the CTA for half your viewers
• Hook visual-audio sync: Is there text in the first 3 seconds AND a strong spoken hook? If not, silent viewers miss your opening entirely
• Caption density analysis: Coverage by thirds — is captioning consistent or front-loaded?

Cross-signal data now factors into the AdZhi Score composite. A dual-channel ad scores higher than an acoustically identical audio-only delivery.

Two new pre-spend predictions added to every analysis report:

• Fatigue velocity: Slow / Moderate / Fast — how quickly this creative will fatigue in a running campaign, estimated from acoustic entropy and persuasion half-life before you spend a pound
• Downward counterfactuals: "If you'd used one filler word in the hook, thumb-stop would drop 18 points." Shows how close a high-scoring ad is to failure — where the fragility is

Three new ways to access AdZhi:

• Analysis packs: One-time purchase of 25, 100, or 250 analyses. Never expire. Buy when you need a burst without upgrading your plan.
• Extra seats: Growth and Enterprise customers can add team seats at £19/seat/month — same analyses, more logins.
• API tier (£99/month): Self-serve REST API access with 500 analyses/month. For developers and tools teams building acoustic analysis into their workflows.

The acoustic brief now closes the loop. Set acoustic targets when you brief a creator (target CTA Momentum above 80, no filler words in the hook, energy arc building to the close). Upload the delivered ad. See exactly which targets were hit and which were missed — with specific re-recording direction for every miss.

AdZhi is live. Upload any video ad or paste a YouTube URL. Get a full acoustic analysis in under 3 minutes — no ad account, no platform data, no spend required.

• 11-stage analysis pipeline: Whisper transcription → acoustic analysis (librosa pYIN pitch, RMS energy, HNR, disfluency) → Claude act structure scoring → intelligence layer → 8 proprietary metrics
• 8 named proprietary metrics: Voice IQ™, CTA Momentum™, Hook Decay Rate™, Persuasion Half-Life™, Voice-Script Alignment™, Creative Entropy™, Attention Stability™, Specificity Index™
• Voice-Script Misalignment detection: 7 misalignment types, timestamped, with specific re-recording direction for each
• Persuasion DNA: 20-dimension acoustic fingerprint enabling similarity search across your ad library
• Missed opportunity detection: 5 absence patterns — what's not there but should be
• PDF export and shareable links
• Brand memory: Acoustic DNA accumulates with every analysis — your brand's winning patterns become searchable

Free tier: 1 analysis (no card required). Starter: £49/month. Growth: £159/month.