Define the audience question

Most weak short videos start from a topic: "the cliff walk," "budget skincare," "my study routine." A topic gives you a lot to talk about. It doesn't tell you when you've finished, what to leave out, or whether anyone got what they came for.

A script works better with a purpose, meaning one audience question it actually answers. An audience question is the specific thing a viewer needs to understand, decide or do, put in their words.

Here is the case for this course. You posted a video about a coastal cliff walk. Under it, a parent comments: Can our kids, six and eight, manage this walk next month? And how do we get back to the car?

• Topic: "the cliff walk."
• Audience question: "Can children of six and eight finish the cliff walk, and how does the family get back to the car?"

The second version tells you what goes in the script (difficulty for young children, the return trip) and what stays out (the café, the history of the coastline). When an AI tool drafts the script later, the question becomes its brief.

Your viewers can now get a general answer without you. In a Pew Research Center survey of 5,119 US adults (February 2026), 49% said they use AI chatbots, up from about a third in 2024. Looking things up was the most common reason, at 42%.

So a parent could ask a chatbot whether the cliff walk suits young children and get a reasonable-sounding general answer. What the chatbot doesn't have is your walk. You did it in May. It was gentle until the last mile, which has 300 steps, and your nephew sat down on the 200th.

The hosts of the creator-economy show Colin and Samir said something similar about the flood of generated video. Any video idea chosen only for strategy, they argued, will be "made by someone or something" anyway. The question worth asking is what only you can add.

Instagram head Adam Mosseri makes the same point from the platform side: the tools change who can make things and how fast, but creators keep their taste, judgment and commentary.

A good audience question is one where your checked facts and your experience make the answer better than a general one.

The words in a comment are often not the whole question. Under "Can they manage it?" the parent is really asking two things:

1. Will the steps be too much for them? (a worry about the hardest part)
2. Will we make it back? (a worry about getting stranded)

To find the real question, ask what the viewer needs to understand, decide or do:

• Understand: where the hard part is and how hard it is.
• Decide: whether to go with these children, and when to start.
• Do: how to get back to the car, and by what time.

Check whether others are asking the same thing. In this case, forty people are wondering about it. That shows the need is shared and the question isn't one family's quirk.

One privacy habit to keep: take the need from the comment, not the family's details. Your script can say "children around six to eight." It doesn't need the commenter's name or anything about their children beyond what the answer requires, and those details shouldn't go into a cloud AI tool.

Before you prompt an AI tool, sort your source packet, the small set of material the script will draw on. It usually mixes three kinds of information, and the script must keep them apart:

• Verified fact: confirmed in a primary source you can name and date. Example: the park's timetable page shows the last shuttle leaving the far end at 5:30.
• Third-party or expert claim: something another person or site states. Useful, but still needs checking. Example: a hiking blog says "most children over five manage this route."
• Personal experience: what you saw or did, marked as yours. Example: your nephew sat down on the 200th of the 300 steps.

Video journalist Cleo Abram holds AI to the standard she'd hold "a research assistant" or colleague to: anything that must be exactly right needs its source shown. The more a viewer will act on a fact (a shuttle time they plan their day around), the stronger the source has to be.

Personal experience is valuable. It just isn't proof. "My nephew struggled" doesn't mean every eight-year-old will, and the script should say it as your experience.

Trend lists, idea queues and "what's performing" dashboards tend to reward provocative wording. Headlines like "The walk that nearly stranded us" or "Never take kids here" grab attention. They also make promises the video can't honestly keep. You weren't stranded. Plenty of families do the walk.

When a trend list hands you an idea, put it through three tests before it becomes your audience question:

1. Is someone actually asking this? Look for comments and repeated questions, not just search volume.
2. Is it true to what happened and what the sources say?
3. Can the viewer use the answer? Will they understand, decide or do something?

You can still note a trend. In this course you'll write a short trend note: what is circulating, and how your question relates to it. Write it down as context, not as the reason you make the piece.

Rosanna Pansino, a creator and prominent critic of AI that copies creators, warns that floods of bot-made content make it hard for authentic work to find a place. Chasing the same provocative angle as everyone else makes you easier to lose in that flood.

A short video has room for very little. A fifty-second answer on TikTok, with an intro and a sign-off, leaves you perhaps forty seconds of real content. Every extra point you add takes time from the others.

That's why a useful question is narrow. For the cliff walk, write down what's out of scope: the café, walking with dogs, the winter route. Each is a fair question and each could be its own piece. Putting them in this one would squeeze the steps and the shuttle time into a few rushed seconds.

A practical test: if you can't answer the question in your available time while giving each point enough room to be accurate, with any qualifier like "in my experience" or "check the timetable before you go" left in, the question is too broad.

Try the slider. A spoken point usually needs several seconds to state it, give its source or say it's your experience, and add any caveat. Watch how fast your time per point shrinks as you add more topics.

Now put it together. Use an AI chat tool (ChatGPT, Claude or similar) for the legwork, and keep the choice for yourself.

A prompt you can adapt:

Here is a viewer comment and a source packet. Each item is labeled verified fact, third-party claim or my experience. List the needs underneath the comment as things the viewer must understand, decide or do. Suggest three one-sentence audience questions a 50-second video could answer honestly. Out of scope: café, dogs, winter route. Do not add facts that aren't in the packet.

Then choose and edit. A strong result for the case:

"Can children of six and eight manage the cliff walk's 300-step final mile, and what time do you need to finish to catch the last shuttle back to the car?"

Creator Ali Abdaal's advice fits here: rather than studying AI in the abstract, apply it to one real task on your own channel. Your output for this lesson is an audience question document, with the question, the need behind it, the labeled packet, what's out of scope and your trend note.

Next lesson, you'll write honest hooks for this question.