Most Creative Strategists Are Asking AI to Do the Wrong Job.

There's a manual version of this. You open Meta's Ad Library, paste a competitor's name, scroll through a few hundred active ads, sort by impressions when you can, count partnership ads on your fingers, screenshot the patterns, paste them into a doc, and write the brief. Four hours per brand if you're fast. Twenty hours if you do four brands a month.

The structural problem isn't the time. It's that the work is the part of creative strategy AI handles best, and the work most strategists ask AI to do least.

Ad copy is the thing strategists ask Claude to write first. Ad copy is the thing it writes worst, because the copy is downstream of the research, and Claude doesn't have the research. Hand it the research and the copy follows. Hand it nothing and the copy is generic.

The three workflows below run on Claude Cowork. Each one is a saved prompt that runs on demand or on /schedule. The structure of each prompt is the methodology... what to capture, in what order, what to drop. The prompts are an installation of how an agency that runs $1.6M+/month in managed Meta ad spend decides what the next campaign should say. Cowork is the engine. The Research Director, the role those prompts configure, is the layer.

If Cowork isn't installed yet, /claude-cowork is the setup walkthrough... Global Instructions, autonomous tasks, and saved Skills. The shorter version: install Claude Desktop on macOS or Windows, sign in to a paid Anthropic plan ($20/month entry tier), and enable the connectors these workflows depend on (browser/Chrome for everything below; Slack if you want scheduled deliveries). Connector availability and scheduled-task limits vary by plan tier and ship cadence; verify against current Anthropic plan documentation before relying on these workflows for client work. Sonnet handles the routine workflows; switch to Opus for persona extraction when the review dataset is large.

Paste it into Cowork. Cowork opens the browser through the connector, reads through the library, and writes the report. A publicly demonstrated Ridge teardown returned roughly 290 active ads across 4 product lines, a 50/50 video-to-image format split, single-digit partnership-ad share, and durability, toughness, lifetime-guarantee, and sales as the dominant messaging strategies. The exact numbers shift week to week as brands launch and pause; the structure of the deliverable holds. Run the prompt yourself for current numbers.

Some metrics Cowork derives reliably from what the Ad Library exposes (counts, dates, format mix). Others (partnership-percentage, inferred personas) involve interpretation; review them against your own read of the top ads before pasting into a brief.

Type /schedule to put it on a recurring cadence. Cowork walks you through the available frequency, day, time, and prompt configuration. Pick whatever cadence Cowork supports closest to monthly (the exact menu evolves; check the live UI). Verify whether scheduled tasks run server-side or require your local machine to be active in your plan tier before scheduling for off-hours. The answer determines whether you can schedule overnight or need to align the run with your normal laptop time.

Run it on direct competitors. Run it on your client's competitors. Run it on brands operating at the spend level your client wants to reach. The compute cost on the entry plan is small against the four-hour-per-brand teardown it replaces; verify your plan's usage caps before scaling beyond a handful of brands per cycle.

Before running these workflows on review platforms, note that some platforms (Trustpilot, Yotpo, Amazon, others) restrict automated extraction in their terms; review the source platform's terms before pulling at scale or for commercial use.

The personas document is the part that matters across time. Loaded back into a Claude project, it becomes the upstream context for every ad creative brief, every landing page rewrite, every email rewrite. The brand's voice becomes a saved file the strategist edits once and uses for the next twelve months.

David Ogilvy ran research before he ran an agency. He spent the years before founding Hewitt, Ogilvy, Benson & Mather working under George Gallup at Audience Research Inc., where the work was statistical sampling, audience studies, and direct-response measurement. He carried that worldview into the agency. The thing he wrote about most... and stayed disciplined about across his career... was that creative work without research is decoration.

The story is not new. The implication is.

The majority of creative strategist time is research. Ad library teardowns. Reading reviews. Tracking competitors. Building personas. Writing the brief. The “creative” part... the headlines, the visuals, the production direction... is downstream of all of it, and most strategists never get a research director on the team because the budget belongs to creative production.

The role most strategists never had a budget to hire is now a configured Cowork session. The Research Director is the role. Cowork is the substrate. The prompts above are the install. The methodology that decides what to capture, in what order, with what output structure, is the work that doesn't change when Anthropic ships a new UI next quarter.

The research compresses. The strategy stays. The strategist still edits.

The Research Director is one layer in the stack a creative strategist running Meta ads operates on top of.

If Cowork isn't installed yet, /claude-cowork is the setup page. Once the research is in hand, the strategic frame for what each ad should say lives at /meta-ads-creative-framework... the awareness-stage map that decides what each variant addresses.

The variant production pipeline sits at /ai-ad-pipeline. The manual entry point is /ai-motion-control. The account-management layer is /run-meta-ads-from-claude. The billing context for any account spending real money is /meta-ads-billing-deadline.

The full strategist stack runs on one chat surface: research with The Research Director, render with Higgsfield, manage with the Meta Ads MCP, optimize with the Meta Ads Agent. The landing page that closes the lead the ads send sits at /how-we-build-websites... the cognitive-bias build constraint Undeniable applies to every site.

Browse the rest of the lab at /resources.