Planning from your back catalogue is two jobs. One is knowing what is actually in there, at the level of specific moments rather than titles. The other is deciding what to make next. Claude is strong at the second job the moment you give it the first, and it has no way to do the first on its own, because the archive is video and the assistant reads what you hand it. Most plans built this way fail on the inventory step, not the ideas step.
Why the title list is not an inventory
A spreadsheet of past recordings tells you the subjects you announced. It does not tell you that the best two minutes anyone has ever said about pricing objections happened in the middle of a webinar about something else. Archives get valuable through those moments, and they are exactly what the titles hide. Anyone who has tried to search a back catalogue through a public transcript page and a handful of site-restricted search operators has found where that approach ends.
The routes
Doing the inventory by hand is the baseline. Someone watches, someone takes notes, and the notes are as good as their attention span. For a dozen recordings it is fine. For a few hundred it never gets done, which is why so many archives are theoretically valuable and practically inert.
Pasting transcripts into a conversation works for one show. It falls over at the point where the transcripts do not fit, and it throws away timing, so the plan that comes out cannot point at anything.
A content library with tags gives you an index over whatever someone typed in. It answers well-formed questions about categories and stays silent about everything nobody thought to tag. That work is real, and the workflow around it matters more than the tool.
Pointing an assistant at your storage gets you names and folders. What an assistant can decide from file structure stops firmly at the file boundary: it cannot see or hear anything inside. The last route is a retrieval layer the assistant can query directly, which is what makes the inventory step something you do in the same conversation as the plan.
Ask what you already covered, then plan against it
Vivu can be added to Claude as a custom connector. You upload the recordings you expect to mine into a project once, and because the indexing happens at that point rather than per question, next month's planning conversation searches the same library without setting anything up again. In the conversation you ask for the thing you would actually reuse, like the part where a customer explains why they put off the decision, and you get back openable time ranges with a line of reasoning on each. Claude works from the ranges you keep, so the plan comes out attached to places in the footage instead of a list of topics you hope you covered well.
Ask in the words your audience would use, not your internal ones. The wording gets taken fairly literally, so a query about disagreement can hand back a small procedural correction rather than a real difference of opinion, and naming the subject you care about inside the query is what fixes it.
The part where a person has to sit down
Nothing in this returns a content plan. It returns candidates, some of which are near misses, and the judgment about which of them is worth a piece is yours. An empty result is also ambiguous: it can mean you never covered the topic, or that you asked in a way the archive does not match, and those lead to opposite decisions. Treat the output as a shortlist to watch, and keep the writing step honest by having watched the moments you are planning around.
When your archive does not need this
If you have published a dozen videos, all in the last year, and you made most of them, there is nothing here that beats an hour with the list and a coffee. The same is true if your plan for the quarter is already set and the question is only what to shoot. The archive becomes worth querying when it is bigger than any one person's memory of it, or when the person doing the planning arrived after most of it was made, which in practice is the situation almost every marketing team eventually lands in.
Where this route stops
Each search runs inside one project, so the recordings you want to plan across need to be in the same place. Material has to be uploaded and indexed in the cloud first, and that is a real piece of work for a large back catalogue, not a switch you flip over your existing storage. Searching draws on an allowance. What you get back are ranges you open, not exact frames, so anything heading for an edit still goes through a person. And none of it knows how the original video performed, so if the question is which topic earned the most attention, that answer lives in your analytics and nowhere in the footage.
How to tell whether the archive is worth mining
Pick one thing you are planning to make next quarter and ask whether you have covered it before. If you can answer from memory in a minute, your archive is still small enough to hold in your head and the tooling is premature. If answering it means someone opens six files, and the honest answer is probably, somewhere, then the archive has already outgrown its inventory, and every month of publishing widens the gap. What you decide to do about that is a resourcing question, and it is worth asking before the next big shoot rather than after.
FAQ
Do I have to upload our whole back catalogue first?
No, and it is usually a bad first move. Start with the material you would realistically reuse, which for most teams is long recordings that get mined repeatedly, such as webinars, customer interviews, and event sessions. Uploading and indexing takes time and allowance, so the sensible order is to prove the archive answers one real planning question before committing the rest of it.
Can Claude tell me which of our past videos performed best?
Not from the footage. Views, watch time, and pipeline influence live in your analytics and your CRM, and no amount of searching the video itself surfaces them. What retrieval answers is what exists and where, which is the input to a planning conversation rather than a substitute for the performance data. Teams usually end up bringing both to the same meeting.
How is this different from repurposing tools that cut clips automatically?
Those tools decide for you which parts are worth cutting and then produce the cut. What is described here does the finding and hands the moments back, leaving both the selection and the edit to people. Which one you want depends on whether your bottleneck is producing more short pieces or deciding what to say next, and those are different problems that get solved by different categories of tool.
What do I do with a moment once I have found it?
Open the range, watch it, and pass the timecodes to whoever is writing the brief or building the edit. The useful output of a planning session is a short list of moments somebody has actually verified, with a note on why each one earns its place. That list survives handoff in a way that a topic list never does.