How to use the UploadPack idea generator

How do you use the UploadPack idea generator?

Connect YouTube and add the competitors you track first — the generator reads both, so it is the first step in Publishing OS. Then open step 1, Idea generator, and select Generate ideas. The tool reads your connected data, your published catalogue and the search terms already reaching your channel, ranks candidate subjects in code, then returns a set of filmable titles with the evidence behind each one. Choosing Research this idea carries the title into opportunity research, and an approved idea becomes a complete publishing Pack. Viewing the set you already have is always free; only generating a new set counts against your allowance.

Key takeaways

• The idea generator asks you for nothing — it reads evidence you already have.
• Candidates are ranked in code before any AI is called, so the same evidence gives the same shortlist.
• Every idea names the signal behind it: competitor gap, audience search or format gap.
• Viewing your set is free and unlimited; only generating a new set uses an allowance.

Before your first run

The generator works from evidence, so what you connect determines what it can see — which is also why it begins at the Creator tier rather than on the free plan, where there is one competitor slot and no requirement to connect YouTube. Tracking competitor channels is the single biggest change: with none, the competitor-gap source is silent and the tool is limited to your own catalogue and search terms. Three to five channels in your niche is enough. Connecting YouTube unlocks two further sources — your classified catalogue, which is how gaps are found, and your privacy-thresholded search terms, which is where the strongest suggestions come from. Saving your channel memory (niche, ideal viewer, voice, boundaries) shapes how ideas are phrased and which are excluded as off-brand. If the evidence is thin, the generator says so rather than inventing plausible-sounding topics. That is deliberate: a suggestion with no evidence behind it is worse than no suggestion, because it looks the same as a good one.

Running it

Open Publishing OS. The workflow now begins at Start here — YouTube & competitors, because the generator reads what is connected: running it before anything is tracked produces the thinnest possible set. Once a channel and a few competitors are in place, move to step 1, Idea generator, and select Generate ideas. The tool reads your connected data, ranks candidate subjects in code, then makes a single request to phrase and justify the strongest ones. It usually takes under a minute. What comes back is a set of cards, each one a filmable video title rather than a theme. Your set is then cached, so opening the page again shows the same set and costs nothing: viewing is always free and unlimited. Only generating a genuinely new set counts against your allowance, and the page shows how many you have left. The generator is included from Creator upwards — ten sets a month on Creator, thirty on Studio, one hundred on Agency. Free accounts take the research-and-choose route instead, which starts from a topic you supply. If you run out, you keep the set you already have rather than losing it.

How candidates are chosen, before any AI is involved

This is the part that matters most, and it is the part most tools leave vague. The shortlist is computed deterministically, in code, before a language model sees anything. Three signals are scored separately. A competitor gap is a tracked outlier — a video that beat its own channel’s comparable-age median, not simply a popular video — whose subject does not overlap meaningfully with anything in your catalogue. An audience-search gap is a search term that already reached your channel with no upload answering it. A format gap is a title structure or packaging route that repeats across several tracked channels and is absent from your recent uploads. Only then is a model called, and its job is narrow: turn each selected subject into a title and explain why it fits. It does not choose what appears. Ranking inside a model would make results vary between runs for no gain in quality, and would make cost scale with how much data you have rather than how many ideas you want. Your existing saved ideas are used only to exclude repeats. They are never a source, because a generator seeded from your own backlog will keep proposing variations on what you have already considered.

Reading a card

The page leads with the three best-evidenced ideas rather than the whole set. Candidates are ranked by evidence before the model is called, so the tail of a set is the weakest evidence found — showing all of it would hand you back the decision the tool exists to make easier. The rest are still generated and kept, one click away under “show more ideas with weaker supporting evidence”, because the work is done either way. Every card carries four things. The label at the top names the signal that produced it — Competitor gap, Your audience asked, or Format gap — and that label is not decorative. “A tracked channel reached 240,000 views with this angle” is a different kind of claim from “12 searches for this reached your channel and no upload answers it”. The first says a format is proven somewhere else. The second says demand is already arriving at your door. They justify different levels of confidence and different amounts of production risk. The premise tells you what the video covers and who it serves. Why now explains the fit for your channel at this moment. First step is the single action to take before filming, and it is deliberately concrete — collect these sources, list these references, build this timeline — because the gap between an idea and a script is usually a research task nobody defined.

What it will not tell you

The generator does not predict views, and it says so on every set. It cannot see competitor click-through rate, retention or revenue, because YouTube’s public API does not expose them — so an outlier tells you a subject earned attention on one channel relative to that channel’s own baseline, not that a market gap exists. Search evidence states the actual number of searches rather than implying volume, and terms below a small floor are not proposed at all, because a single search is not demand. An idea scoring highly on evidence is a well-supported hypothesis, not a forecast.

Taking an idea forward

Each card has one action: Research this idea. Selecting it opens step 3, Research and choose an idea, with the title already in the field. From there Opportunity Intelligence runs live research and scores demand, competition, audience fit and — where the topic is genuinely about buying or choosing something — commercial value. A topic that is not about selling anything has commercial value withdrawn rather than scored low, so an explainer is not marked down for having nothing to sell. If the research supports the idea, approving it carries the direction into a full publishing Pack: title routes, thumbnail direction, a paste-ready description, chapters, Shorts briefs and the upload checklist. If the research does not support it, defer it to the archive. That is a good outcome — the research cost you minutes instead of a fortnight of production.

Getting better results

Three habits help. Track competitors who are slightly ahead of you rather than far ahead: outliers from a channel a hundred times your size describe a different game, and the entry-difficulty signal will tell you so. Keep publishing, because your own catalogue is what makes gap detection accurate — the more the tool knows you have covered, the better it gets at finding what you have not. And treat search-term ideas as the highest-priority tier: they are the only signal that is first-party, specific and already demonstrating intent. If a term brought people to your channel and you have no video answering it, that is the closest thing to a queue of viewers waiting for something you have not made.

Official and first-party references

• YouTube Data API v3 reference
• YouTube Analytics API metrics reference

Idea generator questions