An AI ad workflow for small brands: start from the footage you already have

We made the ads for SmartSolar first, and built a library around them. In October 2026 we ran the same workflow by hand for a brand outside our own. This page is that workflow in plain steps, for a small brand with ad history. Turntake offers it as Bring your own content Early access. Early access: we set it up with you. Coming: we’re building it.

Who this fits

It fits you if you have run ads for a while, so the ad account has history. It fits if the footage sits in a drive, spread across folders no one can keep track of, and you would rather not sort it by hand. And it fits if you use Claude or are happy to start: in Turntake’s version, your own AI does the looking.

You do not need new footage. If you have no ads yet, start with new concepts instead, as described in how to make AI UGC ads. The short version of this flow is on the Turntake Bring your own content page.

Source: our own method, run by hand on one brand outside our own, October 2026.

Step 1: connect, read-only

Start with what anyone can see. A read-only connection lets someone look at your ad history and your folder tree without changing either: no posting, editing, moving, renaming or deleting. In early access we set up the connection with you, and Connect buttons Coming is how you will connect your sources without us.

Read-only still covers a lot, so read the consent screen. Google’s Drive scopes word the read-only file scope as “View and download all your Drive files”, and a lighter one as “View metadata for files in your Drive”. Meta’s permissions reference describes ads_read as access to ad report information for accounts you own or have been granted access to, and it ties sending web events from a server to the same permission. Know what you are granting.

Read-only lets someoneIt never lets anyone

• Read-only lets someoneSee your ad history and resultsIt never lets anyonePost or edit an ad
• Read-only lets someoneList your folders and filesIt never lets anyoneMove or rename a file
• Read-only lets someoneView and download filesIt never lets anyoneDelete anything

Before anything is sorted, we ask four things:

• Your products, listed with one photo each. Agreement was low when we guessed the list ourselves.
• Rights. Whether creator footage may be cut into new ads is your call, and a question for your own adviser if you are unsure. We give no legal advice.
• Editors’ project files, if you have them.
• Your AI policy: what AI may and may not make for your brand.

Steps 2 and 3: best ads, then the clean footage behind them

We start from the ads that already worked, ranked by what they spent and what they delivered. Starting there keeps you from guessing: the ads that ran are the strongest evidence you have of what works for your audience. The same video often runs in many ad sets, so we merge the copies. Catalog and dynamic ads have no single file, so they stay out of the footage match. You get a short list, and you approve it before anything else happens.

Then we look for the footage behind each ad. Finished ads make poor sources, because in our work most of the winners had captions burned in. Clean footage sits in raw folders, in project files and in the large master files left in delivery folders.

File names lie: “final_v3_FINAL” says nothing about what is inside. So we map the whole drive with a plain listing of names, sizes and hashes (Drive itself stores an md5Checksum for files with binary content), then match footage to ads by picture and length, not by name. Even a file named “no titles” proves nothing: end cards and watermarks slipped through in our work, so we confirm cleanliness by eye. The full method is in how to repurpose video ads.

Editors’ project files are a bonus. A Premiere Pro project file records the in and out points a person chose, which makes them a source of hooks and selects. Premiere Pro project files Early access is optional, and few winners may be covered by them.

Step 4: clips sorted by job

Next the footage is cut into clips, and each clip is filed by the job it does in an ad, not by what it shows: hook, explanation, demo, result, reaction, call to action, end card. How a clip looks (phone, produced, AI), which product it shows, the text in the frame and whether anyone speaks are separate fields. Our labelers agreed on the job far more often than on the look or the product, so the owner confirms the look in review.

1. Hook
2. Demo
3. Before and after

The cutting rules are simple:

• Cut at scene changes, never inside a transition or a flash.
• Cut talking heads by sentence, from word times, and never mid-word. Keep them in their own folder.
• Drop black frames and fades. Voiceover edits use only b-roll where nobody speaks.
• Nothing under one second, and b-roll about five seconds at most.
• A clip must make sense alone, with a clear purpose and the product in frame within a second or two.

Naming and fields are covered in how to organize video footage for ads.

Step 5: you rate

Rating has three choices and a comment: keep, favorite or archive. We chose that over five stars because the difference between three and four stars is never used for anything. The doubtful clips come first, “delete” means archive, and the whole step works from the keyboard. If a clip needs fixing, you leave a comment such as “shorter” or “wrong tag”, and your AI carries it out and answers each one. Inside Turntake this is Rating inside Turntake Coming. Until then we run the rating step with you.

Step 6: enough, or missing?

A useful score answers one question: how many new ads can I build, and what is missing? It does not answer how many clips you have. In our work, many green clips gave few ads, because hooks were missing. So measure the library in unique hooks, and cap how many come from the same creator. As a rule of thumb we keep any one person or place to about a third of a bucket. Our targets are proposals, and we have not calibrated them.

The plan for the gaps has an order:

1. Find it in your own footage first.
2. Film it second.
3. Use AI only where that is honest: b-roll without the product, text cards made from your brand’s own files, a slow camera move on a real packshot where the product does not change shape.

When you fill a gap in a library of your real footage, never use AI for talking heads, reactions, hands demonstrating your product or claims about what it can do. See AI b-roll for product ads for where the line sits. AI fill-ins Coming is the plan for fill-ins in that spirit.

What stays human

No fake customers, no invented reviews, no AI person claiming they used your product. A synthetic person can describe a typical situation, but never a personal result. The same goes for numbers: use only the ones you can document.

The honest promise of this workflow is “you can build this many new ads, and here is where you lack hooks”, not a count of clips. You press go. Everything is created paused, your ad account stays read-only, and you decide what goes live. Once ads are live, plan for fatigue early with creative fatigue on Meta.

Which footage problem should you solve next?

Use the detailed guide for the task in front of you. Each owns one question, so you can work through a specific production decision and return to this overview for the wider workflow.

• Controlled clip tagging
• Footage coverage gaps
• Project-file selects
• Read-only library research
• Log-footage transforms
• Product proof shot lists
• Useful phone footage
• Variable frame rates
• Library deduplication
• Stable asset names
• Traceable review proxies
• Restorable asset archives

Who wrote this We made the ads first, for SmartSolar in Denmark, Sweden and Ireland. Then we built Turntake. Where Turntake comes from