AI for creative strategy: what to hand off to LLMs

When it comes to creative work, AI is very good at working from something real and noticeably worse at inventing something human from scratch. Give an LLM a performing script, a set of customer reviews or last week's numbers, and it'll save you hours. Ask it for a net new script from a blank prompt, and you'll usually get copy your audience scrolls straight past.

Most creative strategists we talk to have reached the same conclusion through trial and error. Here's where AI truly ads value for creative work, where it still falls short and how to set it up so the output is worth using.

Parts of creative strategy where AI works

1. Scheduled reporting

The weekly creative report is the most obvious win. One in-house strategist we spoke with has hers scheduled in Claude for 9am every Monday. By the time she opens her laptop, the week's numbers are summarised: what scaled in the testing campaign, how the top ads moved from last week and what's taking spend without being efficient. The analysis still needs her judgment, but the data pull and first-pass summary don't.

2. Flagging creatives that need iteration

The same strategist runs a second scheduled task that looks at the right metrics and suggests which creatives that need iterating. She reviews the list, picks what's worth acting on and re-briefs with a link to the existing asset.

3. Iterating on real material

This is where LLMs perform best in creative work. Take a transcript from a video that's performing and ask for a version written for a different motivator, or rewritten for another archetype. Because the model starts from something human that already works, the output tends to be much higher quality.

The same applies to images and motion. Starting from a real creative ("here's a static that works, make a version of it this way") gives much better results than generating from scratch. A practical example: turning a performing static in Figma into a motion version with Figma's AI tools. You get a new format for different Meta placements with very little risk of the AI getting it wrong.

4. Turning meeting notes into draft briefs

After a monthly creative retro or sprint review, Claude can read the call transcript, pull every idea and suggestion that came up, turn them into a list of briefs and, after a human review, create them as drafts in Focal through the Focal MCP.

5. Research synthesis

LLMs are good at reading more than any person has time to. Give it a few hundred customer reviews and ask it to identify 10 recurring motivators, or to summarize what's changing in your market. It's a strong starting point for new concepts, as long as a human decides which findings are worth briefing.

6. Reviewing and critiquing

In our own testing, the newest models such as Claude Opus 5.5 do their best work as reviewers. Load them with your tone of voice guidelines, brand context and ICP, and they're great at spotting logic gaps, weak claims and small copy fixes in a brief or script.

Where AI for creative strategy still falls short

AI is still fairly terrible at producing net new scripts from scratch. The output tends to sound plausible and generic at the same time, missing the specific, human detail that makes UGC or founder-led creative work.

It's also audience-dependent. Younger audiences, and marketers themselves, are quick to spot copy that reads as AI-written. The more your audience values authenticity, the less you should lean on generation.

We covered where Claude helps and where it falls short in a recent Focal Point session with a creative strategist at a fast-growing consumer app. Watch that section here.

LLMs also tend to be verbose. The fix is building separate, specific instructions (in Claude, skills) for different jobs, such as one for static briefs and one for video briefs.

If your brand has strict brand guidelines or rules on AI-generated content, avoid AI-generated ads entirely and use AI to adapt real creative instead: resizing, reformatting, turning a static into motion. The creative stays yours, and AI handles the production work around it.

Always keep a human review step

Every one of the workflows above follows the same principles:

• AI produces a draft

• A person reviews it

• Only after review does it become a brief, a report or a creative.

• At least when starting out, build a review step into every scheduled task, check the output in Claude and make tweaks before anything gets into a ready brief.

This applies to using AI in performance creative as well. Real creativity, original ideas and taste stay human. What AI changes is how much context sits behind each decision and how much of the admin around it disappears.

AI is only as good as the context you give it

Most of what goes wrong with AI in creative strategy comes down to context. An LLM can't tell you which hooks already failed if it can't see what was tested. It can't brief from last quarter's winners if the winners are scattered across Drive, a project tool, ad names and a spreadsheet.

Connecting an AI tool to each of those systems separately gets you faster access to fragments. It still has to guess which asset came from which brief and got which result, because those connections were never recorded. Better context decides which advertisers win.

The Focal MCP gives Claude and other AI tools the full creative context: the brief, the asset, its tags and its performance together. That's what makes the scheduled reports, iteration suggestions and draft briefs above accurate enough to be worth reviewing. You can read more about what the Focal MCP does.

Test, don't settle

Don't stick to one setup. Models have changed dramatically in the space of a single year, and a workflow that didn't work in January might work well now. Keep testing, change things up and don't be precious about the tools you've built.

Give your AI tools the full creative context. Get a demo of Focal.