---
name: teleprompter_technique
description: Read from a teleprompter or prompter app without sounding like a robot. Use when someone needs prompter setup help, speed calibration to their natural pace, eye-line and delivery technique, or a decision on when to abandon the script.
metadata: { "includeInPrompt": true }
---

# Teleprompter Technique

## Purpose
Get a speaker reading a script on camera while sounding like a person talking, not a machine reading. Covers script prep, hardware and app setup, pace calibration, eye-line discipline, natural delivery, and knowing when to ditch the script. Works for physical prompters (beam-splitter glass rigs) and prompter apps (phone, tablet, desktop, voice-tracking).

## Workflow
1. **Prep the script for the ear, not the eye.** Run `bin/prompter-prep.py` on the script file for a timing estimate and robotic-phrasing flags. Then rewrite per `references/script_prep_guide.md`: read it aloud, convert written phrasing to spoken phrasing, add contractions, mark emphasis (bold or underscore), keep stage directions sparse.
2. **Set up the rig.** Per `references/setup_checklist.md`: glass at 45 degrees with text in mirror mode for hardware rigs; lens centered behind the glass; no strong lights aimed at the glass. For apps: set font size knowing the tradeoff (bigger = fewer words on screen and less lookahead; too small = reading errors), narrow the margins to cut eye movement, park the current line in the upper third of the screen.
3. **Calibrate speed to the speaker's actual pace.** Per `references/pace_calibration.md`: measure the speaker reading 150 words of real material aloud and time it. Set the scroll to match, not to an ideal. Manual control (hand remote, foot pedal, trackpad) stays available even with voice-tracking auto-scroll on; the prompter yields when the speaker overrides it.
4. **Lock eye-line discipline.** Eyes move, head stays still. Memorize the first line so the take starts looking at the lens, not hunting for the text. Read one phrase ahead with peripheral vision so the mouth never catches up to the eyes. Planned glance-aways (a nod, a pause, looking aside to "think") read as spontaneous; random eye darts read as reading.
5. **Deliver like talking, not reciting.** Punch the key words in each sentence, the way you would when telling a friend. Vary pace and pitch; never land every sentence on the same downward note. Short pauses between thoughts beat rushed connectors. Hands and face do part of the work: gesture naturally, let the face match the words.
6. **Decide what stays scripted and what goes off-script.** Per `references/abandoning_the_script.md`: use the script where exact wording buys credibility (numbers, quotes, legal or technical statements). Go off-script for trust beats (personal stories, anecdotes, emotional turns); mark these in the script with a bullet like "My personal story" so the speaker knows the prompter stops there.
7. **Practice and review.** Rehearse with a free prompter (e.g. CuePrompter.com) before the real take. Record a full run, watch it back, and fix the two or three worst moments rather than polishing everything.

## Output Contract
When asked to prepare someone for a prompter read, deliver:
- A marked-up script (emphasis marked, off-script beats marked, long sentences split)
- A timing estimate from `bin/prompter-prep.py` at the speaker's calibrated WPM
- Setup notes: rig type, font/margin/speed settings, lighting check
- Two or three delivery notes specific to this speaker's footage, not generic tips

## Operating Rules
1. **Accuracy rule (non-negotiable).** Every factual claim must be verified via web search or a primary source. Never invent numbers, dates, quotes, or events. Unverifiable = cut it or frame as opinion.
2. Calibrated pace always beats ideal pace. If the speaker's measured WPM is outside 140-160, set the prompter to their number and coach toward the band gradually, not by forcing a speed they cannot hold.
3. Exact-wording material (prices, dosages, quotes, legal claims) never gets improvised. Flag these lines in the script as stay-on-script.
4. Never promise the audience cannot tell. The goal is "sounds natural," not "invisible prompter." Overclaiming kills trust.
5. Style: no em dashes anywhere in skill output. Plain, direct language.
6. If you write or modify Python CLIs, compile with `python3 -m py_compile` before reporting done.