We Asked ChatGPT to Write Our Dad's Memoir. Here's What We Learned.

We have been asked, many times, some version of this question: Can’t I just use ChatGPT to write my dad’s life story?

It’s a fair question. ChatGPT is remarkably good at writing. It can produce clean prose, organize a narrative, adjust tone, and turn bullet points into flowing paragraphs faster than any human editor. If the goal is words-on-page about a human life, it can do that.

So we ran an experiment. We had two adult siblings work on preserving their father’s story. One used ChatGPT and a questionnaire the father answered in writing. One used a phone-recorded conversation following up in real time. We then showed both outputs to the father, to his wife, and to his three grandchildren, and asked them to respond.

Here is what we found.

What ChatGPT produced

The AI was given a questionnaire the father had filled out — two pages of handwritten answers to 12 questions about his childhood, career, marriage, and values. It was given his birthdate, where he grew up, and the names of his siblings.

The output was clean. Structured. Readable. The sentences were well-constructed and the narrative arc was clear. It read like a competent magazine profile of a man the writer had never met.

That was the problem.

His wife read it and said it was “like a very nice eulogy written by someone who didn’t know him.” His eldest grandchild, who is 14, read it and said it “sounded like a school project.” His son, who had submitted the questionnaire in the first place, said the prose was good but that he felt strange reading it — like looking at a photograph of someone wearing his father’s face but not quite right.

What it got right: the facts. Dates, places, jobs held, milestones reached. The factual skeleton of a life.

What it missed: everything that makes a life a life. The way he pauses before he answers anything important. The four-letter word he uses only when he’s actually impressed by something. The story about his father that he always starts twice — once briefly, then again from the beginning because he realizes he left something out. The laugh that appears in his voice before the punchline of any story he’s told more than five times.

What the recorded conversation produced

The recording session ran 97 minutes. The daughter did not prepare formal questions — she had three prompts on a notepad and followed the conversation wherever it went. Her phone was propped against a glass of water on the kitchen table.

She got the same facts the questionnaire captured. She also got a seventeen-minute detour about a man her father had worked with in the 1980s who had died young and whom he thought about “more than you’d think.” She got her father’s voice breaking slightly when he described the first time he held his eldest child. She got a two-minute impression of the foreman from his first construction job that made her laugh so hard she had to pause the recording.

She got silence. Several of them. The silences before his most considered answers, which were different from the silences when he was searching for a word.

When the 14-year-old grandson listened to the recording, he asked to hear it again. He asked who the man in the 1980s was. He asked his grandfather about it directly the next time they saw each other — a conversation that lasted 40 minutes and that his grandfather described afterward as “one of the best I’ve had in years.”

What this experiment actually tested

This wasn’t really a test of ChatGPT. ChatGPT performed exactly as designed — it synthesized clean prose from the information it was given.

What the experiment tested was the questionnaire versus the conversation. The written answer versus the spoken one. The prepared response versus the unguarded one.

A questionnaire produces what someone is willing to write down when they have time to edit themselves. A conversation produces what someone actually thinks, including the things they didn’t mean to say, the tangents that turn out to matter, the emotional texture that doesn’t survive the process of being written.

AI can be useful at the output end of this process — organizing transcripts, cleaning up filler words, structuring a narrative from raw interview material. That’s a legitimate use. What it cannot do is replace the input. It cannot substitute for the act of actually talking to the person.

The real question underneath the question

When people ask “can I just use ChatGPT?,” they are usually asking something slightly different: Can I skip the part that feels hard?

The hard part is sitting down with your parent and asking questions that matter and having a conversation that might get emotional and running out of time and knowing that the window is closing. That’s the hard part. ChatGPT can produce prose, but it cannot sit down with your father and have that conversation for you.

No tool can.

“What you’re really asking when you ask me to tell you about my father’s life is: did anyone bother to ask him? AI can write the story. It can’t be the person who cared enough to show up and listen.”

Use every tool available to you — including AI — to make what you capture more organized, more accessible, more useful to the generations coming after you. But the thing those generations will actually want, more than a well-structured memoir, is your father’s voice telling his own stories in his own words.

There’s only one way to get that.