Start by turning the request into a description of what is visible on screen, because that is the only form every search method can act on. A client asking for "the shot of their CEO at the conference" is describing a person and an event, and neither is something your footage knows about. What the footage contains is a man in a grey jacket standing at a podium, a wide shot of four people seated with a backdrop behind them, a close shot of someone gesturing while they talk. Translating the request into that vocabulary is the step that decides how fast the rest goes.
Work backwards from the delivered cut
The fastest route is usually the project file, not the archive. If the shot was in something you delivered, the sequence tells you exactly which source clip it came from and where, and an exported edit list gives you the same information without opening the editor. This handles a large share of client requests, because clients mostly ask for things they have already seen. It fails on the other case, which is a request for something that was shot and never used, and that case is where the archive search actually begins.
Ask whoever cut it, too. An editor who spent a week in those rushes can often name the card and the rough timecode from memory, and that is worth five minutes of someone's time before you start anything systematic.
Bins, selects, and the limits of tagging
Well-run projects have organized bins and a selects reel, and that structure is why you can find the hero shots. Tagging at ingest extends this, and automatic tagging tools will label objects and scenes without anyone typing. Both approaches share a ceiling: they find what someone, or something, thought to label at the time. The client request that arrives eighteen months later is usually the one nobody anticipated, phrased around a detail that seemed unimportant while you were logging.
There is also a cost question. Tagging an archive properly is an ingest discipline applied to every job, including the ones nobody ever revisits, and that cost is paid up front against uncertain future value.
Transcripts, if anyone said it
If the request is about content rather than composition, transcript search is direct and cheap. This covers interviews, panels, and anything with a speaker. It covers nothing visual, which is most shot requests, and it only finds the exact words used. A client asking for the bit where their founder talked about the early days will be findable. A client asking for the shot with the empty conference room will not.
Searching by visual description
The other route is indexing the footage itself so it can be searched by describing what the frame looks like. This is the category built for exactly the request in this article's title, and it handles specific compositional descriptions well: a wide shot of several people seated with a sponsor backdrop behind them comes back as a set of time ranges, each with a line explaining what it matched on. Descriptions of posture and action work too, which is how you locate the same person repeatedly across a long recording without any face recognition being involved.
Vivu fits here, with each client's footage in its own project and every search running inside a single project rather than across all of them. Selected results export as the original clip, which is what actually goes back to the client or into the edit.
Two limits are worth knowing before you rely on this. The match reasons describe how people look rather than who they are, so identifying a specific person means writing down what is visible about them. And an empty result is genuinely ambiguous: on a recording that was shot mostly in wide, a request for a close single may come back with nothing, and you cannot tell from that whether the shot does not exist or your description was too tight. Loosen one constraint at a time rather than rewriting the query.
When none of this is worth setting up
If the footage is from one shoot, on one day, and you were there, open the folder and scrub. If the client request is for something you delivered last month, the project file answers it in under a minute. Archive search becomes worth building when requests arrive for material that is years old, shot by people who have since left, or spread across enough jobs that nobody can say with confidence whether the shot exists. That threshold is about turnover and volume rather than about how much footage you have in terabytes.
One last thing to settle before indexing anything: confirm you have the right to hold and reuse a former client's footage. That is a contract question and it is easier to answer before the material is sitting in a searchable system than after.
So the judgment is about which half of the problem you actually have. If the shots you get asked for are mostly ones the client has seen, invest in keeping project files and edit lists retrievable, and skip the rest. If you are regularly asked for material that was never delivered, no amount of project hygiene will reach it, and you are choosing between logging everything at ingest and indexing the footage so it can be described later. Finding the finished video is a much easier version of this problem than finding a shot inside it.
FAQ
What do I do when the client cannot describe the shot?
Get them to name something you can act on: the campaign it was for, the shoot date, the city, or the deliverable they think they saw it in. Any of those narrows the archive to a manageable set, and from there you can send a contact sheet or a handful of stills and let them point. Asking a client to describe a shot precisely tends to produce a description of what they remember feeling rather than what was on screen, which is why anchoring to a job or a date works better than pressing for detail.
Can I search footage for a specific person by name?
Not by name, unless someone has already tagged that person in your system. Content search works from what is visible, so the practical approach is to describe appearance and action: a man in a dark suit standing and holding a microphone, a woman in a red jacket seated on the left. This finds the person reliably within a single recording where they wear the same clothes, and it does not carry across shoots. Treat it as locating a described figure rather than identifying an individual.
How should we store footage for several clients so searches do not mix them up?
Keep each client's footage in its own project or library and run searches inside one at a time. This is mostly an organisational discipline rather than a technical feature, and it is worth enforcing at ingest, since separating material after the fact is tedious. The practical benefit shows up in results quality as much as in separation: a search scoped to one client's material returns fewer plausible-but-wrong matches than one run across everything you have ever shot.
What if the shot turns out not to exist?
Say so early, and say what you did check. The awkward case is that an empty search result does not prove absence, so the defensible answer is to describe the coverage that does exist on that shoot and offer the nearest alternative. On a recording made entirely in wide, no amount of searching produces a close-up. Being able to show the client what was actually covered is usually more useful to them than continuing to look.