# How to pull selects from hours of client footage > Pulling selects from hours of client footage means getting from 'somewhere in these files' to a short list of moments an editor can open and trust. Canonical URL: https://vivu.ai/guide/how-to-pull-selects-from-hours-of-client-footage Pulling selects from hours of client footage means getting from "somewhere in these files" to a short list of moments an editor can open and trust. There are four ways to do it. You can watch and log everything, cut from a transcript, filter on the tags in an asset manager, or upload the footage to a search tool that indexes it once and lets you describe the shot you want. Which one fits depends on two things: whether the shots in the brief are spoken or only visible, and how often this client's archive gets reopened. ## Turn the brief into findable moments first A selects pass goes faster when the brief becomes a list of findable moments before anyone opens a file. "Energetic lifestyle b-roll" isn't findable. "The product on a kitchen counter in daylight" is, and so is "a hand unlocking the phone to show a notification". Write the list, then mark which items are spoken lines and which are purely visual. That split tells you which method will carry most of the load. Keep each client's footage in its own library or project from the start. When clients share one pool, every result list gets longer, and someone eventually has to ask whose material ended up in whose cut. ## Route 1: watch and log An assistant editor scrubs the footage at speed, drops markers, and writes a log with timecode, a description and a rating. This is the most reliable way to judge performance and framing, because a person looks at every frame. It costs hours of viewing for every hour of footage, and the log only covers what the logger thought mattered that day. When the next brief asks for something different, you're back to scrubbing. ## Route 2: cut from the transcript Transcribe the interviews, testimonials and talking-head footage, search the text for the line, and jump to it. For anything spoken this is fast and cheap, and [finding clips through a transcript](https://vivu.ai/guide/how-do-i-use-a-transcript-to-find-good-video) goes through it step by step. The limit is simple: a transcript doesn't know what's on screen. A product close-up, a reaction shot, or a screen recording with no voiceover will never turn up in a text search. ## Route 3: tags in an asset manager If the client's footage already sits in an asset manager with automatic tagging, you can filter by labels like "outdoor", "person" or "phone". Tags narrow a pile of files. They're generic, though, and usually attached to a whole clip, so you still open each file to find where the shot actually happens. ## Route 4: upload to a search layer and describe the shot The fourth route treats the footage as something you query. You upload the client's files to a search tool, it indexes them once, and after that you type a description of the shot and get back time ranges to open instead of whole files. What it costs you is the upload itself, a usage allowance on most tools, and a review step, because a result matched to a description is a candidate and not yet a select. How you search matters here. With short ads and product demos, a quick search can return whole clips that probably contain the shot. That answers "is it in here at all?" but still leaves you dragging a playhead. In [Vivu](https://vivu.ai/solutions/marketing) the fast mode behaves this way, while the precise mode runs as a job you wait for and returns short stretches with a one-line reason for each, such as the time and notifications shown on a lock screen. A fast pass to see which files matter, then a precise pass to land on the moment, is a sensible order, and the results you pick can be exported as the original clip for the editor. ## Handing selects to the editor Whatever route built the list, the editor needs the same things: the file, the time range, one line on why it's there, and ideally the clip itself so nobody searches twice. Leave some handle on both sides of each range. Writing [the brief around those clips](https://vivu.ai/guide/how-to-brief-an-editor-with-clips-from-existing) is a separate step. If the list was put together in an AI chat, [moving it from the chat to the edit](https://vivu.ai/guide/how-to-hand-a-selects-list-from-claude-to) has its own pitfalls. ## When you don't need a search tool If the job is one shoot you logged on set, the log already is your search tool. If the footage is mostly interviews, a transcript covers nearly everything. If a client sends two hours of footage once and never comes back, uploading and indexing buys you little over an afternoon of scrubbing. Search starts to pay off when the same client archive keeps getting reopened for new briefs and the shots people ask for are visual. ## Which side you're on Look at the last three briefs for this client. If most of the moments they asked for were things someone said, stay with transcripts. If they were things someone had to see, and your team has scrubbed the same folders more than once, a search layer is probably worth the upload. ## FAQ ### Can AI pick the best take for a client edit? No. A tool that matches descriptions can find every take where the product is held up to camera, but choosing between those takes is an editorial call about performance, framing and what the client approved last time. Treat search results as a candidate list for a person to review. ### Is it okay to upload client footage to a search tool? Check the contract first. Many agency agreements say where client material may be stored and who may process it. If uploading is allowed, give each client a separate project, so every search runs inside one client's footage. ### Why does a search return the whole clip instead of the moment? Some quick search modes match whole files, so with short ads the "result" is the entire clip. That tells you the shot is in there somewhere. To get the stretch itself, use a slower, more precise mode if the tool has one, or scrub only the few files the quick search turned up. ### What should a selects list include? The source file name, the time range, a one-line description of what's in it, and why it matches the brief. Add a little handle before and after each range. If the tool can export the original clip, attach it so the editor doesn't have to find it again.