The ten year old test: make any explainer video easy to follow (3 Claude prompts, free)

If you want your explainer video to be easy to understand, run your script through the ten year old test before you film it. Your viewer is scrolling, half watching, with the sound turned down. Three Claude prompts get your script ready for that viewer.

What you'll have when you're done

• A script where every line is one short idea a ten year old could follow
• A list of every hard word in it, each one explained in five words or cut
• A shot list that says what's on screen for every sentence, so the picture matches the words
• One before/after example to check your own work against
• [SCREENSHOT: the finished two-column script, line on the left, on screen on the right]

Before you start

• A Claude account. The free plan is fine. (The prompts also work in ChatGPT.)
• One script you want to turn into a short video. A rough draft is fine. So is a blog post you want to cut down.
• 15 minutes.

The test, in one line

Picture a ten year old watching your video on a phone, half paying attention. If they would get lost at any line, that line fails. The three steps below fix the words, the hard terms, and the picture, in that order.

Our own retention playbook writes this as a rule for every educational reel: use 5th-grade language. A ten year old is a fifth grader. The test is that rule with a face on it.

Step 1: Rewrite every line for a ten year old

Open a new chat in Claude and paste this, with your script at the bottom:

Rewrite my video script so a ten year old could follow it on the first
listen.

Rules:
- One idea per sentence. If a sentence has "and", "because" or "which"
  joining two ideas, split it into two sentences.
- Short, everyday words. Say "buy" not "purchase", "help" not
  "assist", "start" not "commence".
- Keep my meaning and my order. Don't add new points or cut my ending.
- Keep it sounding like a person talking, not a textbook. Contractions
  are fine.
- Put each sentence on its own line, numbered.

After the script, list any line where you had to drop detail to make it
simple, so I can decide if that detail matters.

My script:
[paste your script here]

Read the result out loud. Each numbered line should take one breath. If a line still makes you pause halfway, ask Claude to split that line again.

The list at the end matters. Sometimes a dropped detail is the whole point of the video. Put it back as its own short line instead of leaving it out.

Step 2: Flag every word your viewer might not know

Simple sentences can still hide hard words. Tool names, acronyms and industry terms slip through because you use them every day. Send this in the same chat:

Now go through the rewritten script and flag every word or phrase a
ten year old might not know. Include:
- jargon and industry terms
- tool, app and product names
- acronyms and abbreviations
- numbers or units without context

Show me a table with four columns:
1. LINE: the line number
2. WORD: the word you flagged
3. FIVE-WORD MEANING: what it means, in five words or fewer
4. KEEP OR CUT: say KEEP if the video needs this word, CUT if the
   line works without it

Then give me the script again with every CUT word removed and every
KEEP word explained the first time it shows up, using the five-word
meaning. Only explain each word once.

The rule is short: explain it in five words, or cut it. If you can't explain a word in five words, the viewer won't get it from a passing mention either. Cut it, or give it its own line.

Step 3: Write down what's on screen for every sentence

This is the step people skip. The words are simple now, but a viewer with the sound down only gets what they see. Every sentence needs a picture that says the same thing.

Two rules cover most of it:

• If you say a thing, show the thing. Say Claude, show the Claude logo. Say Instagram, show the Instagram app.
• If you say a number, show the number. Big, on its own, with its unit.

The rule we edit our own reels by: on a quick cut, viewers see, they don't read. No screenshots of paragraphs. No highlighted sentences. Keep the words on screen to a short headline, a number or a button.

Now turn the final script into a two-column shot list.

Column 1: LINE, the exact sentence I will say.
Column 2: ON SCREEN, what the viewer sees while I say it.

Rules for ON SCREEN:
- If the line names a tool, app, brand or object, show that thing: its
  logo, its real screen, or a photo of it.
- If the line says a number, show that number big, with its unit.
- If the line describes a process, show it moving: a cursor clicking,
  a list filling in, an arrow from A to B. Describe the motion.
- No paragraphs, no screenshots of text, no highlighted sentences.
  At most 8 words of text on screen per line.
- If a line has nothing to show, say "FACE" and flag it. Too many FACE
  lines in a row means the line needs a picture or needs to go.

Be specific. "A screenshot of the app" is not a shot. "The Claude chat
box with my script pasted in, cursor on Send" is.

Go down the right column and ask one question per line: if the sound were off, would this picture still tell the viewer what I'm saying? Fix any line where the answer is no.

Before and after: one script through all three steps

Here's a short explainer about how AI chat apps answer questions about your own files. It's written the way most first drafts come out.

Before:

Most people don't realize that when you upload docs to an LLM, it's not
actually training on them, it's using RAG, which retrieves relevant
chunks from a vector database and injects them into the context window
so the model can ground its answer in your data instead of
hallucinating, and that's why the quality of your source docs matters
so much more than the prompt.

That's one sentence with about six ideas in it, plus RAG, LLM, vector database, chunks, context window and hallucinating. A ten year old is gone by the second comma. So is most of your audience.

After step 1 and step 2 (one idea per line, hard words explained or cut):

1. When you give an AI app your files, it doesn't learn them.
2. It looks things up instead.
3. You ask a question.
4. It finds the parts of your files that match.
5. It reads those parts right before it answers.
6. So it answers from your files, not from a guess.
7. This is why your files matter more than your question.
8. Messy files, messy answers.

LLM, RAG, vector database, chunks and context window all got cut. None of them were needed to get the idea across. "Hallucinating" became "a guess".

After step 3 (what's on screen):

Every line has one idea and no hard words. Every picture tells the same story with the sound off.

Step 4 (optional): Save the three prompts as a Claude Project

If you make videos every week, don't paste three prompts every time. In Claude, click Projects → New Project, name it Ten Year Old Test, and paste all three prompts into the project instructions with one extra line at the top:

When I paste a script, run these three passes in order: the rewrite,
the jargon flag, then the on-screen shot list. Show me each pass
before moving to the next one.

Now every chat inside that project already knows the job. Paste a script, get all three passes.