Watch a co-hosted video die and the cause of death is usually on the page, not on the set. One person wrote the script, in their own voice, and then dealt the sentences out like cards — you take this paragraph, I take the next. The result records exactly as written: two mouths, one voice. Both people explain in the same register, qualify in the same places, land jokes with the same rhythm, and the viewer, who cannot see the page, still senses that this conversation is a lecture wearing two costumes. Interviews collapse the same way from the other side: questions written by the answerer sound like an answerer’s questions, feeding lines with no curiosity in them. Writing for two voices is a different craft from writing for one, and the first step is admitting that a script divided is not the same thing as a dialogue written.
The craft begins with giving each voice a job, not a share. Two people should not be splitting the explaining; one should drive and one should resist — asking the question the viewer is shouting at the screen, doubting the convenient number, wanting the example. Expert and proxy, enthusiast and sceptic, the one who knows and the one who checks. Once the jobs exist, the language can follow: the driver gets the longer, structured sentences; the challenger gets the short ones, the interruptions, the plain words. Vocabulary should split too — if one voice says leverage, the other says so what does that mean. The test is mechanical and unforgiving: print the page, cover the names, and read. If you cannot tell who is speaking from the words alone, the viewer will not be able to either, and no amount of on-camera warmth will disguise it.
The second half of the craft is the seams. Hand-overs are where two-voice scripts die on camera: both people speak at once, or neither does, and the take dissolves into the apologetic little pause that no edit fully removes. A hand-over survives when it is written as one — when a turn ends in something the partner can catch. A direct question does it. A name does it. A list left deliberately unfinished for the other person to complete does it. What kills the catch is sameness: two consecutive turns that end on the same falling shape, so nobody can hear where one thought stopped. Vary the turn lengths while you are at it — a script that alternates neat forty-word paragraphs sounds like a relay race, while real conversation volleys, with one voice getting a run and the other breaking it. The seam is a written object. If it only works when both people improvise goodwill, it does not work.
The page format for all of this is the one you would invent in a notebook anyway: the speaker’s name, a colon, the line. In Talk2Camera that format is the feature. Tag lines with a name and a colon, and the prompter colours each speaker, so a glance tells you whose text is coming; it announces every hand-over, so nobody is hunting the page for where their part begins; and voice follow stays exact because the tags are never fed to the recognizer — the name is stage direction, not dialogue, and nobody has to say it aloud for the scroll to keep pace. The machinery reads the script the way the two of you do: as parts, with joins.
A workable way to get there is to accept that the first draft will be a monologue and to budget for two more passes. Write the argument straight through in one voice if that is how it comes; the material does not care. The second pass is the differentiation pass: assign the jobs, move every question to the challenger, every claim to the driver, and split the vocabulary until the cover-the-names test passes. The third pass is for the seams alone — walk the joins one by one and ask of each turn what, exactly, the other person is supposed to catch. While you are in there, shorten everything. A sentence that reads as brisk on the page runs long in the mouth, and dialogue amplifies the effect, because every over-long turn is a hand-over delayed. Punctuation is breathing: a full stop is a place to look up, a comma is not, and a turn with no full stops is a turn nobody can take over from.
Talk2Camera’s AI writer will also draft in this format — a toggle, and the draft arrives as tagged dialogue rather than a monologue for you to chop up. Use it for what a draft is: the skeleton of alternation, the shape of the argument, a first pass at who carries what. What it cannot know is your two actual voices — the words your co-host would never say, the question only she would think to ask — and that layer is a rewrite you do together, out loud. Which is the last rule, and the cheapest: before you record, read the script aloud once as a pair. The seams you stumble on in rehearsal are not the ones to practise harder; they are the ones to rewrite. Dialogue that two people can actually read is written like a game of catch — every line thrown so the other person can take it cleanly, and the whole page organised around the throws.