Wedding Speech AI

The method

A blank chat box writes what you told it. We find out what you forgot to say.

You can get a wedding speech out of any free chatbot in about ten seconds. The problem is not that it writes badly. The problem is that it writes from the one paragraph you thought to type, and the best thing about your friend was never in that paragraph.

The same starting material, two different outcomes

Here is what somebody actually types when they sit down to write a best man speech. Vague, rushed, adjective-heavy. Everybody starts here.

What you typed

“Best mate from uni, about 11 years. Met at an open mic night, he heckled me. We got lost on the way to his stag do. He’s really loyal.”

A general chatbot

Writes the speech immediately. It has no way to know that the heckle had actual words in it, that the stag do detour ended in a farmer’s driveway, or that “really loyal” means he sat in a hospital corridor until four in the morning without being asked.

So it fills the gaps with the only thing available, which is the average of every wedding speech ever written.

What we ask next

Real output from the product, generated from the answers above:

  • “You said Jordan heckled you first at the open mic night. What did he actually shout, and what were you doing on stage when he said it?”
  • “On the stag do detour, how lost did you actually get, whose fault was it, and what happened when you finally arrived?”
  • “You said he shows up for people without being asked. What is the specific time he did that for you?”

Nobody types the hospital corridor unprompted. They type “really loyal”, because that is what the question asked for. The whole method is asking a better second question.

What happens between your answers and your speech

  1. Seven questions

    The same opening questions for everybody: how you know them, how you met, the funniest memory, what you admire, the inside joke, your hope for the marriage. Ordinary so far.

  2. We read them, then ask again

    This is the part that does not exist anywhere else. Your answers go back through a second pass that works out what is missing, then asks four to six follow-ups built from your own words. Not a longer form: different questions for every person, because they are derived from what you actually said.

  3. Write against constraints

    The draft is written under a fixed set of rules: a banned phrase list, role-specific calibration for a best man versus a father of the bride, a structure with one engineered peak, and a punctuation rule enforced in code after the model has finished.

  4. Built to be spoken

    Two drafts at different tones so you can hear which one sounds like you, then revision passes, pause and laugh markers, and a teleprompter. A speech is a performance, not a document.

Six findings, turned into questions

The follow-up questions are not improvised. Each one is chasing a specific thing that research on storytelling, memory and humour says a room responds to. None of these studies were run on wedding speeches. What we did was turn them into the questions we ask you.

Narrative transportation Green and Brock

A listener is moved by a story only once they can picture it. So we ask where you were, who else was in the room, and what somebody said out loud.

The peak-end rule Kahneman and Fredrickson

A room remembers the most intense moment and the ending, and forgets the length. So the speech builds one peak instead of spreading evenly, and the last line does more work than any other.

The dramatic arc Zak, on narrative and empathy

Tension that resolves creates empathy. Flat praise creates nothing. Most wedding speeches contain no conflict at all, which is the main reason most are forgettable, so we ask what went wrong.

Concreteness in memory dual coding research

Concrete detail is recalled far better than abstraction. So every adjective you write is treated as a question we have not asked yet. Generous becomes name the last time, and what it cost her.

The beautiful mess effect Bruk and colleagues

People consistently overestimate how badly their own vulnerability lands. Listeners hear courage where the speaker fears weakness. So we ask for the thing you are slightly embarrassed to say, and tell you why it works.

Benign violation McGraw and Warren

Something is funny when it breaks how things ought to be while staying safe. Wedding humour is a narrow band, so we chase both halves: what went wrong, and what made it okay to laugh about.

13 phrases this product will never write

Not a style preference. This list is enforced on every draft, and it is the same list the writer works from, published here in full. A general chatbot will hand you several of these without blinking, because they are the most statistically likely words in a wedding speech, which is exactly what makes them dead on arrival.

  • “without further ado”
  • “for those who don't know me”
  • “Webster's dictionary defines”
  • “match made in heaven”
  • “tie the knot”
  • “better half”
  • “ball and chain”
  • “journey”as a metaphor for the relationship
  • “soulmate”unless quoting the speaker
  • “they complete each other”
  • “I've never seen [name] so happy”
  • “raise your glasses”as an opener
  • “a definition or a quote from the internet”as an opener

There is one more rule that runs after the writing is finished: no em dashes or en dashes, anywhere. A dash is a writing-desk mark, this text gets spoken out loud, and it is also the single loudest tell that a machine wrote something. That one is enforced in code rather than asked for, because asking is not reliable.

What we are not claiming

Every page like this one is tempted to overstate. Here is the honest boundary, so you can weigh the rest of the page knowing where it ends.

The questions are free to answer

Work through the interview and read the opening of two drafts before deciding anything. No card, no account.

Start the interview