Automating Video and Social Posts

Two piles, and the line between them

Video is where automation promises the most and delivers the most unevenly. Some of the work genuinely disappears. Some of it cannot, and the businesses that get burned are the ones who could not tell which was which until after they had posted.

The line is simple once you see it. Automate the handling. Keep the judgement.

Handling means everything that happens to a video after it exists and after you have decided it is worth showing to people: transcribing it, captioning it, cutting it, describing it, sizing it, scheduling it, filing it. These are mechanical, checkable in seconds, and dull enough that nobody enjoys them.

Judgement means deciding whether the thing is worth posting at all, what claim it makes, and whether it should carry your face or voice. No tool can do that on your behalf, and every attempt to hand it over ends with something published that you would not have approved.

In plain English

Transcript:: The full text of what was said, with timings. Everything else in this lesson is built from it.
Cutdown:: A short clip taken from a longer recording, usually one complete point.
Captions:: On screen text of the speech, for people watching without sound or who need them to follow at all.
Synthetic voice:: A generated voice imitating a real person. Powerful, and the fastest way to publish something you never said.

What genuinely automates

Transcription and captions. This is the strongest case in the whole course. Accuracy on clear speech is good, errors are visible the moment you read them, and captions decide whether a large share of people can watch you at all. Automate this and never think about it again, except to read the transcript before it goes out.

Finding the cut points. A model reading a transcript can find the moments where you made a complete point, which is the tedious part of editing. It cannot tell you which of those points is any good.

Descriptions, titles and chapter lists. Mechanical writing derived from something that already exists. Low risk, provided nothing in the description makes a claim the video does not support.

Scheduling and format wrangling. Sizes, aspect ratios, posting windows, filing the master somewhere findable. Pure logistics.

Prompt you can copy: find cutdowns in a transcript

Below is a transcript with timings.
Find every passage that makes one complete point and would
still make sense to somebody who has not seen the rest.
For each one give me:

start and end timings, copied exactly from the transcript
the point it makes, in one sentence
the first eight words spoken, verbatim
CLIPPABLE YES or NO. NO if the passage refers to something
said earlier, names a person not introduced in it, or
ends mid sentence.
Rules:

Do not rewrite or tidy anything I said. Quote verbatim.
Do not suggest a passage shorter than [20] seconds.
If a passage contains a figure, a price or a promise,
add the word CLAIM at the end of its line.
Rank nothing. I am choosing which to use.
TRANSCRIPT:

That last rule matters more than it looks. Ask for the best three clips and you have quietly handed over the judgement you were told to keep. Ask for all complete passages and you have a list to choose from, which takes two minutes and is the part only you can do.

Checkpoint

Automate the handling: transcripts, captions, cut points, descriptions, sizing, scheduling. Keep the judgement about what is worth posting, and never automate your face or voice.

Descriptions that do not overreach

The description is where an automated pipeline most often invents something, because describing is close to selling and generated text drifts towards selling on its own.

Prompt you can copy: write a description from the transcript

Write a description for this video, under [60] words, using
only what is actually said in the transcript below.
It must:

say what the video covers, plainly
name the one thing a viewer will know afterwards
use the speaker's own phrasing where possible
It must not:

state a result, a figure, a percentage or a timescale
that is not spoken in the transcript
describe the video as essential, definitive or the
ultimate guide to anything
claim the speaker is an expert, a leader or award winning
mention anything the video does not actually contain
If the transcript does not support a description, reply
NOT ENOUGH IN THE TRANSCRIPT and stop.
TRANSCRIPT:

❌ Weak prompt

Prompt

Write an engaging description and pick the best clips from this video to post.

Output

In this must watch masterclass, our expert reveals the 3 proven strategies that helped businesses double their results. Clip 1 selected as the strongest hook.

Three claims nobody made, an expert nobody appointed, and an editorial decision taken by something that has no idea which of those moments you would be comfortable defending.

✅ Good prompt

Prompt

Under 60 words, using only what is said in the transcript. No figures, results or timescales that are not spoken. Do not call it essential or definitive. Separately, list every passage that makes a complete point, verbatim timings, and mark any containing a claim.

Output

A walkthrough of how we categorise expenses, and why consistency matters more than getting every category perfect. Runs about nine minutes. Six complete passages listed below, two marked CLAIM.

Nothing asserted that was not said, and the choosing is still yours, with the risky passages flagged so you look at them first.

The line you do not cross

Some things in this space are not a matter of taste.

Never publish a synthetic version of your face or voice saying something you did not say. Not for a testimonial, not for a product claim, not as a shortcut for a message you were going to record anyway. The moment a generated version of you makes a statement, you own that statement without having checked it, and one wrong figure in a convincing voice is a very difficult thing to walk back.

Never generate a person who does not exist and present them as a customer. A fabricated review, a fabricated case study, a fabricated face endorsing your work. This is not a grey area and no framing rescues it.

Never let a video pipeline publish directly. Every other automation in this course ends with a person approving. This one especially, because video is the format people believe most readily and screenshot fastest.

If you do use a synthetic voice or a generated presenter for something entirely benign, say so plainly on the post. Disclosure costs a line and it is the difference between a production choice and a deception. Rules on synthetic media and advertising disclosure also vary by country, so check what applies where you are.

Instruction you can copy: the approval gate before anything goes out

NOTHING PUBLISHES UNTIL A PERSON HAS TICKED ALL FIVE.
1  I have read the captions. Names, figures and product
words are spelt correctly.
2  Every claim in the clip and in the description is one
I would defend to a customer standing in front of me.
3  The clip makes sense alone. It does not refer to
something said earlier that is not in it.
4  Nobody appears, is named, or is quoted without their
agreement. This includes staff and customers.
5  Nothing in this uses a generated face or voice
presented as a real person.
IF ANY TICK IS MISSING, IT DOES NOT GO OUT.
Not with a note to fix later. It does not go out.

Read your captions before publishing, every time. Automatic transcription is reliable on ordinary speech and consistently wrong on the exact words that matter to you: your product names, your customers' names, and anything in an accent the model hears less often.

📝 Quiz

Question 1 of 4

Where does this lesson draw the line between what to automate and what to keep?