Is writing your own speech still worth doing when AI can draft a polished one in your voice?

The shift

Producing a polished, well-structured speech in a specific person's voice collapses from scarce (hours of the speaker's or a ghostwriter's drafting time) to near-instant and free. What stays scarce is the speaker's genuine conviction about the words, their ability to be held to them off-script, and the thinking the act of writing used to force on the person delivering it.

The axioms

• A polished, voice-matched speech takes scarce drafting effort — either the speaker's hours or a ghostwriter's craft.
• The person on stage is the one who worked through the argument — writing and delivering were bundled, so delivery implied ownership of the thinking.
• Writing the speech is how the speaker figures out what they actually believe — drafting is a reasoning process, not just an output.
• An audience reads polish and fluency as evidence of preparation and care — a good speech signals the speaker took the moment seriously.
• Conviction in delivery comes from having wrestled with the words yourself — you defend live what you built yourself.
• The speaker is accountable for what's said — they answer for it in Q&A, in follow-up, and in whether they live up to it.
• A distinctive voice is scarce and identifies the speaker — sounding like yourself is hard to fake at length.

Invalid axioms

1. A polished, voice-matched speech takes scarce drafting effort. Both the labor of drafting and the craft of matching a specific person's cadence, vocabulary, and rhythm are now near-free — a model produces a competent, on-voice draft in seconds. Habit-trap: speakers and their teams still treat "delivered a well-crafted speech" as proof the speaker did the work, and still budget prestige and prep time around the writing rather than around the harder question of whether the speaker can stand behind and extend the words.
2. A distinctive voice is scarce and identifies the speaker. Sounding like yourself over a full speech used to be hard to imitate — enough recorded material now makes a convincing voice-match cheap. Habit-trap: audiences and speakers still treat "it sounds like them" as evidence it is them and that they meant it, when voice is now the easiest thing to reproduce.
3. An audience reads polish and fluency as evidence of preparation and care. A clean structure and fluent phrasing used to mean the speaker invested real hours. Habit-trap: still using polish as the bar — coaching, judging, and rewarding speakers on how finished the words sound, when finish is now the free part.

Unchanged axioms

1. Writing the speech is how the speaker works out what they actually believe. The drafting was never only producing text — it forced the speaker to confront gaps, decide what they'd stake, and pre-live the argument. Outsourcing the draft removes that, and nothing about a good AI draft supplies the conviction; it hands you words you haven't thought through. The reflective work is still scarce, and it's now optional in a way that quietly costs the speaker on stage.
2. Conviction in delivery comes from having wrestled with the words yourself, and an audience senses when it hasn't happened. A speaker delivering borrowed words they didn't reason through tends to read as hollow — fine on-script, thin the moment they're pushed off it. What an audience responds to isn't the prose; it's the sense that a person means this and has paid for it. That remains hard to manufacture and is not what a drafting tool produces.
3. The speaker is accountable for what's said. They answer in Q&A, in the hallway after, and in whether their conduct matches the words. A model can draft the promise but can't be held to it — accountability for the claim didn't get cheaper because the claim got easier to write. A speech the speaker can't defend or extend live exposes exactly this gap.
4. Ownership of the words is what lets a speaker go off-script. The value of having done the thinking shows up precisely where the script runs out — the unscripted answer, the pushback, the moment that needs a real position rather than a fluent one. That capacity comes from having built the argument, not from having a good draft of it.

New axioms

1. A polished speech the speaker didn't think through is hollow the moment it's questioned. When a finished draft is free, the temptation is to deliver words the speaker never reasoned through — which holds up until Q&A, a follow-up, or any off-script moment, where the absence of real ownership shows. The gap between sounding prepared and being prepared is now easy to fall into and expensive to hide.
2. Disclosure and authenticity become live questions with no settled norm. When a moving, on-voice speech might be entirely AI-drafted, audiences lose a cue they relied on — that the person in front of them built the words and means them. Whether and when to disclose AI help, and what counts as authentic authorship of a spoken message, is unresolved and increasingly load-bearing for whether a speech is trusted.
3. The reflective work writing used to force is now opt-in and easy to skip. Drafting was a forcing function for figuring out your own position; when the draft is handed to you, that thinking doesn't happen unless you deliberately do it anyway. Speakers now have to reconstruct on purpose the clarity that used to be a byproduct of writing — or go on stage without it.
4. Whether delivered conviction can survive borrowed words is an open bet. A speaker might internalize an AI draft deeply enough to own it, or might read fluent words they don't feel — and the same tool produces both. The unsolved problem is telling those apart from the outside, and building the discipline (or the process) that turns a draft into something the speaker has actually made theirs before they deliver it.

Where it breaks

"Delivered a polished, on-voice speech" is still treated as proof the speaker did the work and means it (invalid) at the moment a polished, on-voice speech the speaker never thought through is trivially producible and hollow off-script (new) — the signal audiences and coaches still trust is now the cheapest thing to fake, and the tell only surfaces in the unscripted moment nobody rehearses. Separately, "it sounds like them, so they meant it" (invalid) collides with the unsettled question of disclosure and authentic authorship (new): voice-match is now the easy part, so the thing that used to certify a speech as genuinely the speaker's — that it sounded like them — certifies nothing, and no norm has replaced it.

ai-axioms.vercel.app/axioms/marketing/comms-speechwriting-ai-axioms