The answer... without the scenic route
Select roughly 15 to 30 strong performers and 10 to 20 informative misses, but adjust for channel size. A miss is useful only if it had enough impressions and time to be evaluated. A video with limited reach may reflect narrow topic interest or competition, not a bad title.
Inside this guide 7 parts
Choose winners and misses that teach something
For each video, record:
- Title and topic.
- Publish date and days since publication.
- Views after the same comparison window, when available.
- Impressions click-through rate by meaningful traffic source.
- Average view duration or percentage viewed.
- Primary traffic source and audience segment.
Select roughly 15 to 30 strong performers and 10 to 20 informative misses, but adjust for channel size. A miss is useful only if it had enough impressions and time to be evaluated. A video with limited reach may reflect narrow topic interest or competition, not a bad title.
Add audience and business context
YouTube's Audience tab can show aggregated patterns such as monthly audience, formats viewers watch, when viewers are online, and content they watch outside the channel. Some reports may be unavailable when there is not enough data. See YouTube's audience-report guide (opens in a new tab).
Add only what changes the recommendation:
- Audience promise and core content pillars.
- Key geographies and language.
- Long-form, Shorts, or live focus.
- Typical video length and production budget.
- Topics already overused.
- Accuracy, rights, and advertiser-suitability limits.
Do not treat demographics as a stereotype generator. Age or geography may explain context, but it does not tell you what every person wants.
Use a prompt with an auditable output
An effective model-agnostic prompt can say:
Analyze the comparable winners, informative misses, and audience context below. First identify repeatable patterns and contradictions. Then propose 30 distinct video concepts across core, adjacent, and experimental buckets. For each, provide the viewer need, evidence used, title promise, originality note, production risk, and confidence level. Do not invent facts or copy competitor wording. Mark any idea that requires current-source verification.
Ask for 30 good candidates before 100. Expand only after you review the first set. More output is useful only when the ideas remain meaningfully different.
Don't skip this bit
A winning title may have traveled with a stronger topic, a better thumbnail, unusual search demand, a famous subject, or external traffic. A losing video may have had excellent packaging and weak distribution. Do not ask a model to infer certainty that the data cannot support.
When possible, compare videos of similar age, format, topic scope, and traffic mix. Label every conclusion as observation, hypothesis, or confirmed result.
Where the ledger goes next
Next step: Put the bank to work before closing, but only within the authority you actually have. The next lesson covers safe preparation during a channel handoff.
Continue to Prepare a Channel Growth Plan During the Ownership Wait.
Keep these three things
The short version
- Choose comparable videos, not merely the highest and lowest lifetime view counts.
- Include audience promise, economics, rights, and production constraints in the input.
- Require a rationale and confidence label for every proposed idea.
Sources and further reading
How this was made: Adapted from Roman’s channel operating curriculum, expanded for public education, and reviewed against the ChannelFlips editorial policy. Examples are educational, not promises.
Published · Policy checked
Read the editorial policy