Almost everyone sounds worse on camera the first time than they do in conversation, and the gap is not talent. It is that speaking to a lens removes the two things that make conversation easy: a listener reacting, and the freedom to say it badly first and fix it later.

What replaces them is a set of habits, and they are learnable in an afternoon.

Why scripted delivery sounds scripted

Written English and spoken English are different rhythms. Written sentences carry more clauses, more subordinate clauses, more comma-paced pauses. Read aloud, that structure produces a cadence that nobody uses when talking, and listeners hear it instantly as reading.

The fix is not to abandon the script. It is to write in spoken units:

One idea per sentence. If a sentence contains "which means that", split it.
Contractions everywhere. "Do not" becomes "don't". Read a script with no contractions aloud and it sounds like a news bulletin.
Short first sentence, then the next one. Vary length, but keep most sentences short enough to say in one breath.
No written punctuation tricks. Em dashes and semicolons are invisible in speech and turn into awkward pauses.

A useful test: read the line aloud. If you stumble, the line is wrong, not your delivery. Rewrite it until it falls out of your mouth.

The teleprompter, used correctly

A teleprompter is the tool most likely to make you sound like you are reading, because it encourages you to read. Used badly it produces the flat, eyes-scanning cadence everyone recognises.

Used properly it is a memory aid, not a script:

Put bullet phrases on it, not sentences. Three or four words per bullet. Your brain fills in the grammar, which is what makes it sound spoken.
Place it as close to the lens as possible. Eyes tracking text across a wide screen are visibly reading; eyes near the lens are not.
Scroll slower than feels right. If you are waiting for text, you are reading. If the text is waiting for you, you can look away and come back.
Break the rhythm deliberately. Pause, look off-camera for a beat, come back. Reading never does that; talking does.

If you cannot place the prompter near the lens, bullet points on a sticky note beside the camera work nearly as well and cost nothing.

Say it twice, then keep the second

When you have the choice, record the same idea twice. The first take is where you work out what you are saying. The second is where you say it.

This is why the second take so often sounds better even when the first is technically more accurate: accuracy is not what the viewer is judging. Energy is.

So pick takes by these, in order:

Energy and pace. A slightly wrong word delivered with conviction beats a perfect sentence delivered flatly.
Clarity. No mumbling, no trailing off at the end of sentences.
Accuracy. Last. You can fix a small factual slip with a caption or a cutaway; you cannot fix a monotone.

The wind-up you have to cut

Most people begin a take with a verbal run-up: "okay so basically what I wanted to say about this is". It is a nervous habit, and every second of it is a second where a viewer can leave.

Write the first sentence so it can be the first thing heard. If the take you recorded starts with a run-up, cut it — the cut is invisible if it lands on a word boundary and the next word is said with energy.

Hands, eyes and the rest of the body

Delivery is not only voice. On a phone screen the face fills the frame, so small habits read large:

Look at the lens, not at yourself. Watching your own preview is what makes eyes drift.
Use your hands, but keep them in frame. Gestures that leave the frame look like twitches. Gestures visible in the lower third look natural.
Move on transitions. A slight lean forward when you change point gives the edit natural cut points and keeps the frame alive.
Do not sway. Rocking side to side is the most distracting habit in vertical video because the frame is narrow.

Nerves, practically

Not the deep-breathing advice — the mechanical things that actually help:

Record a throwaway first. Ten seconds about nothing. The first take is always the worst, and using it as a warm-up means it does not have to be good.
Say the first line five times before recording. The opening line carries the most pressure; rehearsing only that line removes most of it.
Stand up. Standing gives you more breath support and makes gestures natural. Sitting produces a smaller, flatter voice.
Accept the mess. You are recording many takes in one session; the pressure is off any single one.

Warm up the voice, not the nerves

Ten minutes of humming and reading aloud before the first take is the difference between the first three takes being throwaways and the first three being usable. The voice needs a few minutes of use before it settles into its normal range, and a session recorded cold sounds thin at the start.

What helps, in order of effect:

Hum for a minute, sliding up and down through your range. It warms the resonances that make a voice sound full.
Read a paragraph aloud in your normal speaking voice — not a script, a news article is fine. This is the part that switches your mouth from reading mode to talking mode.
Say the opening line three times at a normal volume, then once louder than feels natural. The take you record will sit somewhere in the middle of that range.

What does not help: silent deep breathing, which calms you without waking the voice, and rehearsing the whole script repeatedly, which makes the delivery stale before the camera is on.

Practise by recording audio only

Once a day, for sixty seconds, record yourself talking about something you know, with no camera and no script. Listen back. You are listening for three things: filler words, sentences that ran out of breath, and whether the ending trailed off.

This is the fastest improvement available because it removes the camera from the equation. You are not practising being on camera; you are practising speaking in complete thoughts, which is the skill that was missing.

What to fix in the edit, and what to re-record

Editable:

Run-ups, false starts, repeated words.
Long pauses and breaths between sentences.
One flubbed word, if a cut or a caption covers it.

Not editable:

A flat, tired delivery across the whole take.
Mumbling, or a voice that drops away at the end of every sentence.
Reading that is visible in the eyes.

Those three are worth another take. Everything else is a cut.

Once the delivery is right, the audio chain in fixing bad audio is what carries it, and the batching method in producing a week of shorts is what stops you recording five separate warm-ups.

How to Sound Natural on Camera