The family wrote the obituary with ChatGPT — now what?

One funeral-software vendor, Passare, told the Washington Post its AI tool has written tens of thousands of obituaries nationwide in the past few years — one vendor, one tool. The same reporting found funeral directors now ask families at the arrangement table whether they’d like AI to write the obituary, and some funeral homes use ChatGPT for the job without mentioning it to anyone. However your funeral home feels about that, the authorship of the town’s most permanent document has already moved. The question left on the table is who checks it.

Can ChatGPT write an obituary? It already does

Yes — fluently, in seconds, for free, and often through the funeral home’s own website. Legacy.com launched ObitWriter in June 2023: free for families and directors, built to generate a finished obituary from a name, an age, a hometown, and service details, with one-click submission to newspapers. Funeral-management platforms ship the same feature inside the arrangement workflow — tone selectable from a dropdown. And this isn’t a 2025 novelty: an Ontario funeral home told CBC the practice was already commonplace among its client families in 2023.

For a grieving family, the appeal is honest and humane. One man who used ChatGPT to memorialize his mother told the Post he simply wasn’t in a mental place to write it himself — and the result read well to the people who knew her. Nobody should be scolded for that. The drafting problem is solved. What’s not solved is everything the draft gets wrong.

The finished look is the trap

Generative tools produce fluency unconditionally — the prose reads equally confident whether the facts underneath are right or invented. On factual question benchmarks, a peer-reviewed 2025 survey measured leading models fabricating between roughly 14% and 35% of answers — one answer in five, wrong, in a voice that never wavers. This is general model behavior, not an obituary measurement, but obituaries are where it lands hardest, because an obituary’s errors are intimate: the Post’s reporting documented AI-drafted obituaries in which the deceased’s family members were mixed up — and the survivor list is the paragraph a family rereads forever.

The inventions aren’t limited to facts. When a Deseret News columnist asked ChatGPT to draft her own obituary, it fabricated a saying it claimed she used regularly — “We are here to love, to lift and to lead — with faith as our guide.” She’d never said it. In an obituary, an invented favorite saying reads as the most intimate line on the page, and no reader can challenge it. The Post’s own testing of these tools found them padding drafts with unearned biography — “legendary,” “inner fortitude” — airbrushing a specific person into a template.

If it helps to see the failure mode outside deathcare: in 2023, a lawyer filed a federal brief ChatGPT had researched for him. It cited six precedents that did not exist — the judge found “bogus judicial decisions with bogus quotes and bogus internal citations” — and when the lawyer asked ChatGPT whether the cases were real, it assured him they were. The tool’s confidence is identical when it’s right and when it’s inventing. A funeral home publishing an unchecked draft is trusting exactly that confidence, in the one document that can’t be quietly corrected later.

The mistakes that publish an error

“The family wrote it, so accuracy is on them.” The obituary publishes under your funeral home’s name, on your website, into the town’s permanent record. Whoever drafted it, the last professional hands it passes through are yours — and families, readers, and search engines will all attribute what it says to you.

“It reads well, so it’s probably right.” Reading well is the one thing these tools guarantee. The Avianca brief read well too — that’s what got it filed. Fluency is evidence of nothing except fluency.

“AI drafts are that fake-obituary problem I’ve read about.” Different things. A family using ChatGPT on their own loved one’s obituary is a legitimate act that needs editing; the scraper sites and spam farms fabricating obituaries for ad clicks are a separate harm — and their spread is precisely why the verified obituary on your funeral home’s site keeps gaining value as the copy of record.

Saying nothing about how it was written. The Post found some funeral homes use ChatGPT without telling their clients. There’s no law against that, and after a real editorial pass the result is honestly your work-product — but silence is the version with reputational downside if a family pieces it together later. A warm sentence costs nothing: “We used a tool to get a first draft started, then went through every line together.”

The editor-of-record pass, line by line

Passare’s own help documentation gives you the operating assumption, in the vendor’s own words: “Some facts and information might be altered by the AI. Please carefully review the obituary for any inaccuracies.” When the company selling the feature says its output alters facts, believe it. Treat every AI draft as unverified until a person or a document confirms it — start from suspicion, not trust.

The pass itself is concrete. Verify every load-bearing fact against the family and the paperwork, not against how right it sounds: full legal name and spelling, maiden name, both dates, place of birth — and the survivor list above all, because spouse, children, grandchildren, their order, and who’s living versus predeceased is where the documented failures cluster. Hunt down anything specific the family didn’t actually say — a quote, an employer, a hometown, an adjective doing too much work — and delete it. Then do the reverse: put back the true specifics AI sands off. The machine writes “a beloved veteran”; the family knows he was a Marine corporal who drove a ’72 Chevy he refused to sell. The second version is the person.

Two parts of the pass no tool performs. Ontario directors interviewed by CBC described the family terrain a machine can’t see — estrangements, second marriages, which grandchildren get named, which faith gets foregrounded — the exact judgment arrangers are trained for, and as one director put it, AI simply doesn’t know any of it. And before anything publishes, read the finished obituary aloud with the family. Reading aloud is where someone catches the wrong middle name and remembers the story that belongs in the second paragraph.

None of this changes the publishing craft afterward — the permanent URL, the real HTML text, the page built so machines can quote it accurately. AI changed who types the first draft. It changed nothing about what a durable record requires.

What the edit is worth

There’s no professional standard for any of this yet — no NFDA position paper on AI-drafted obituaries, no rule to point at. The norm is being written at arrangement tables right now, which means a funeral home that writes it well gets to be known for it. The honest version of the service sounds like this: the draft is free now, from us or from anyone — what we provide is the guarantee that the published record is true.

That’s not a diminished role. “We’ll write it for you” stopped differentiating the day the biggest obituary platform started giving the writing away. What a family cannot get from ChatGPT, at any price, is a professional who verifies the middle name against the birth certificate, catches the estranged son before print, restores the ’72 Chevy, and reads the final version aloud in a quiet room. The typing was never the service. The judgment was — and judgment just became the whole job.