# Current creative handoff: Round 47 Start with the 40-second **All your ecommerce data. In your favorite AI.** film. The promise is simple: SourceMedium connects commerce data to the AI you already use. The logo opening represents 60+ ecommerce integrations, SourceMedium in the middle, ChatGPT and Claude in the first client row, and Codex, Claude Code and Cursor in the second. The film shows ChatGPT analysis and a PowerPoint, a Claude dashboard/app, then the real recurring-workflow action in ChatGPT. ## Round 47: keep the middle moving At 11–17 seconds the real ChatGPT retrieval continues on screen. Copy: “Everything your AI can do. Now with your data.” Supporting line: “In ChatGPT, Claude and other MCP-compatible agents.” This is part of the demo, not a centered brand card. Preserve the genuine app motion and camera continuity. A short lift in the persistent agent rail replaces the standalone logo animation. Grok Bot is in the shared lineup. Portrait keeps agent names above the destination; landscape keeps both in the bottom margin. Muse remains pending compatibility clarification. Grok Bot's MCP/plugin policy is documented by xAI; a SourceMedium connection in that app was not exercised during this edit. Every demonstrated workflow remains ChatGPT or Claude. Big Idea and all film durations are preserved. Round 47 is under `.traffic/creatives/story-47/`; Round 46 is the rollback gallery. Publication status and complete verification are in Round 47's review note. ## What each asset is for | Asset | Duration / formats | Job | | --- | --- | --- | | Choice | 40s / landscape, portrait | Recommended introduction to the product promise | | Acquisition | 30s / landscape, portrait, feed | Explore data and create a PowerPoint in ChatGPT | | Paid, paid-chatgpt, paid-review | 20s each / portrait, feed | Trusted data, growth review and familiar-tool hooks; identical body | | Claude | 35s / landscape, portrait | Separate affiliate question, answer and reporting site | | Workflow | 60s / landscape | Inspect the full ChatGPT process, not an acquisition ad | | Customer, customer-intro | 30s, 60s / landscape | Customer setup invitation; longer narrated release pending | | Analyst | 15s / landscape, portrait | Separate Slack brand teaser; no recorded Slack workflow | The gallery puts current work first and preserves superseded studies in a collapsed archive. Its supporting page contains all three paid hooks, customer edits, both workflows and the existing narrated setup walkthrough. The old carousel is explicitly archived and is not a current distribution asset. ## Round 46: maintained data and the full AI workspace The lead now says “Unified, accurate, documented, fresh,” then “Accessible to all your AIs via MCP.” At 11–17 seconds, “Everything you do in ChatGPT, Claude, or any MCP-compatible agent. Now with your ecommerce data.” All five agent logos float up with a soft stagger, then hold. Real exploration starts at 17 seconds, PowerPoint at 22, dashboard/apps at 26, weekly review at 31, ending at 36. Only the two lead exports are new in story-46. All supporting films retain the verified Round 45 gallery bytes and actual original receipts. The Big Idea music source and mix are unchanged; the lead excerpt now runs 40 seconds with its fade at the new ending. This does not claim identical features in every AI, an actual external-file retrieval, or a completed recurring job. The gallery publication was explicitly requested; use the same Site and current audience. Actual deploy status is recorded in story-46/review/ROUND-46.md. ## Round 45: copy you would actually say Replace “Talk to unified, trusted data” with **Talk to your data.** The lead continues “Right in ChatGPT and Claude,” then “Keep asking. Go deeper.” Its output sequence is “Create a PowerPoint,” “Build a dashboard. Or an app,” and “Schedule your weekly review.” Keep the existing possibility-led ending. Focused films use “Ask in your own words,” “SourceMedium brings your data together,” and “Now show your team.” The longer explanation keeps trust explicit with “Start with data you can trust.” Claude says “Build the dashboard you need.” The three paid opening hypotheses stay distinct and share the same revised body. This is a copy-only revision: titles and chapter labels change together. Footage, source evidence, durations, cameras, recurring-action treatment, layout, 60+ integrations, all five agents, destinations and Big Idea remain unchanged. Analyst is unchanged. Avoid infrastructure adjectives and generic “put your data to work” filler; speak to the viewer, name the useful thing and let the real app show the mechanics. Natural wording must still make SourceMedium's data role clear. Local outputs: `.traffic/creatives/story-45/`. Round 44 remains the verified baseline, with pre-edit source in story-45/review/before. See that round's review report for completed export checks. No publication or commit is included. ## Round 44: a clear weekly-review action Remove the generic “Demo data · Edited for time” footer from current MCP films. Keep SourceMedium.com/mcp and all five named agent marks. The separate Analyst teaser retains its own identity and product-workflow label. The lead says **Schedule your weekly review.** Its real ChatGPT offered action gets five seconds, at 25–30s in the 34-second film. Establish the full captured app, then move once into the intact weekly-assessment button. Dim surrounding text during the close-up so cropped analysis does not compete with the action. Use a taller proof stage in portrait. The PowerPoint payoff is four seconds; the dashboard/app remains five. Overall length and Big Idea mix are unchanged. This is an offered scheduling action, not evidence of an activated or completed job. No browser action was submitted; the source screenshot remains unchanged. Current local outputs: `.traffic/creatives/story-44/`. Previous verified exports remain in story-43; pre-edit source is in story-44/review/before. No publication, commit, audience change or ad launch is included. ## Round 43: capabilities, not editorial commentary Lead: create a PowerPoint; build dashboards and apps; set up recurring workflows; build anything you can think of. The lead stays 34 seconds. The five-second Claude ranking/app replaces its request-and-output detour, freeing four seconds for the actual ChatGPT weekly-assessment action screen. That supplied screenshot is a static capture, with one deliberate zoom to the button. It shows an offered recurring action, not a completed or newly scheduled job. Original capture and provenance are in story-15/captures/sources.json. The original was a JPEG named PNG; the prepared PNG preserves every decoded RGB pixel. Remove “Another example, in Claude.” and editorial date/analysis footnotes from all current MCP edits. The actual app and deck text remains untouched. The later user correction supersedes older instructions to repeat periods in overlays. All MCP cuts keep ChatGPT, Claude, Codex, Claude Code and Cursor in the bottom margin, with names to distinguish shared marks. No extra client badge sits above “Create a PowerPoint” or other capability captions. The separate Analyst teaser retains its own identity. Native client UI still identifies each real example. Keep the opening, 60+ ecommerce integrations, SourceMedium hub, simple arrows, all supporting scene durations, and the selected Big Idea mix. Current outputs: `.traffic/creatives/story-43/`; full baseline: story-42 and story-43/review/before. Local gallery only. Existing release and audience limits below still apply. ## Round 42: simpler connection, lower destination The lead opens “All your ecommerce data. In your favorite AI.” This follows the current website MCP hero. The opening uses straight, fine arrow lines with one quick travelling signal, informed by HomeFlowConnector.astro. No looping dots or double arrowheads. Keep all nine source marks, 60+ integrations, the SourceMedium hub and both client rows. The persistent prospect URL now sits in the lower-left margin. Landscape puts the disclosure on the same row at the right; portrait stacks the two below the app. This clears the longer workflow’s two-line caveat. The Claude footer stays below its evidence notes. Customer setup endings and the separate Analyst teaser retain their own destination behavior. Footage, cameras, timing and Big Idea are unchanged. All 18 current formats are refreshed under story-42; the prior lead is in story-41 and supporting exports in story-40. Local gallery only. ## Round 41: complete the opening promise The lead opens with “All your ecommerce data. Ready for your AI.” Keep the second sentence in the existing supporting type so the integration and agent logos retain their space. Split the next sentence across two captions: “Talk to unified, trusted data.” then “Right in ChatGPT and Claude.” Follow with “Dive deeper.” The real ChatGPT PowerPoint and distinct Claude dashboard examples remain sequential. Keep their native client labels and the separate-example marker. No split screen. Only the lead's landscape and portrait exports change. Round 40 supplies every supporting film and its original receipts/source snapshot. Timing, framing, proof, caveats, client lineup, CTA and Big Idea remain unchanged. The opening promises choice of supported AI tools without claiming universal compatibility. ## Round 40: the capability is the headline Viewers should understand the benefit without reading the native chat history. Use “Talk to data you can trust” and “Dive deep into your data,” not headlines about the most valuable customers or winning affiliates. The existing analyses remain truthful illustrative proof. Name the finished outputs directly: “Create a PowerPoint” and “Build a custom dashboard.” Close with “What will you create?” to invite broader uses without asserting that every imagined task is supported. The lead keeps its source-to-SourceMedium-to-AI opening, then trust, exploration, PowerPoint and a distinct Claude dashboard example. The dashboard headline holds across request and output. Other cuts show only the output they actually contain. Trust refers to SourceMedium's maintained data, checks and documented definitions; it does not guarantee AI answers. Keep SourceMedium named in focused context scenes. Footage, timing, caveats, Big Idea and destinations are unchanged. Round 39 remains the source, export and gallery baseline. ## Round 39: one punchy headline per beat Remove secondary analysis recaps such as “Here, subscribers showed higher value over 12 months.” Keep “Find your best customers.” or “Understand your customers.” as the single benefit-led headline. Claude's numeric recaps and the longer workflow's metric list are also removed. The actual app/deck evidence, period labels and material caveats remain unchanged. Keep “Another example, in Claude.” to distinguish the lead's separate demonstrations. The short Slack identity labels remain useful. Round 38 is the full source/export/gallery baseline. ## Round 38: name what becomes available Lead with “All your ecommerce data. In the AI you already use.” Use “data” for what SourceMedium connects and “business” for the questions and benefits. End with “Your data. Your AI.” The broader product story and all real proof remain unchanged. “All” means data connected through SourceMedium, not universal source coverage. Round 37 remains the complete source, export and gallery baseline. ## Round 37: product introduction plus focused proof The lead film establishes the broader change: your ecommerce data inside the AI tools you already use. It no longer inherits the customer-value ad. Its own introduction.json supplies a 30-second body; Choice adds the existing four-second source-to-SourceMedium-to-client opening. The total remains 34 seconds. ChatGPT demonstrates customer analysis and a useful deck. A visibly separate Claude example demonstrates a reporting request and the generated affiliate site. They are not presented as one analysis or a cross-client handoff. SourceMedium connects the data; the AI tools investigate it and create the outputs. The presentation is one possibility, not the whole product promise. The focused acquisition, customer, longer workflow and Claude cuts retain their own question, finding and output. Their copy now frames these as examples of using business data inside a familiar AI. The paid openings compare a customer question, preparing a growth review and bringing business data into ChatGPT; the rest of their footage and copy are identical. The Analyst remains a clearly labeled separate Slack teaser, with the same data-in-your-workspace idea. No added proof of external connected-tool use or successful recurring execution is claimed. Those require actual matching demonstrations. “All” refers to data connected through SourceMedium; do not call the advertising lineup exclusively first-party data. Keep the 60+ count and full client lineup. Preserve all material evidence caveats, the selected Big Idea music, and existing audience destinations. ## Round 35 correction The prior audit missed the older two-logo ending used outside Choice. Every MCP ending now shows ChatGPT and Claude on the first row, with Codex, Claude Code and Cursor on the second. This includes the paid hooks, customer variants and Claude workflow. The separate Analyst teaser stays specific to Slack. The logo identities live in src/clients.ts and the shared ClientChoices renderer serves both ending paths, so future updates cannot silently reach only the logo-led cut. ## Accepted creative choices - Original moving ChatGPT and Claude apps, one screen at a time. Establish the complete app, zoom deliberately for analysis, return for controls. No permanent top crop, doubled screens, fabricated product UI or synthetic proof. - Original PowerPoint slide exports for crisp output. The final test-plan slide follows the customer-value finding; do not reintroduce the unrelated FLASH25 headline. Keep original screen resolution limits honest. - Short, value-led captions. No standalone dollar-versus-dollar headlines. Preserve the cohort period, observation window, incomplete-cost and non-causal caveats. Claude explicitly changes from January–August conversions to July–August orders, including repeat orders. - The selected **Big Idea · breakbeats + brass** Suno recording is unchanged. MUSIC-RIGHTS.md records the exact file, hash and commercial-use provenance. Automated loudness and hash checks do not replace listening on actual speakers. - SourceMedium.com/mcp persists in the lower margin in MCP prospect films. Customer films keep the setup CTA; the separate Analyst teaser points to /ai. - Website typography, lavender actions and graphite phone frames. Preserve the full screen, balanced line breaks and both native export orientations. ## Release state These are reviewed creative candidates, not a launched campaign. On 15 September 2026 the /mcp destination returns **404**. The AI page, connection guide and help page return 200. Do not substitute a different landing page or deploy the website without authorization. A final phone-speaker/headphone listen is still required. Narration scripts are not recordings. The supplied phone footage has no usable narration; the existing setup walkthrough retains its own original narration. Paid exports also require preview in the actual placement before distribution. The existing review Site is configured **public**, despite its historical private label. Updating the local gallery does not change its audience. Confirm publication scope before replacing the live version. No ads or customer distribution are part of this editing task. CapCut is optional finishing, not an assembled project. ## Continue editing Use tools/demo-video in Traffic. The JSON edit maps are canonical. Render to a new output folder, mix, validate, then use scripts/refresh_gallery.py to update both review pages together. Keep a baseline. Render with one worker: parallel workers have produced missing native text despite passing decode checks. Review actual scene and boundary frames, plus the smallest format, not only an empty corner. Keep the headline, client attribution, proof, qualification and next action as a single readable composition. A one-second scene cannot carry a paragraph. Keep the paid ending and its CTA grouped within the useful viewing area. Checks of static branding cannot establish that animated native text rendered correctly. Current local outputs and complete gallery: .traffic/creatives/story-42. Historical decisions remain in projects/mcp/HANDOFF.md and earlier round reviews. The tool README has the commands; review/ROUND-42.md records actual verification.