Hearing a familiar voice produced by AI can feel powerful. A system might imitate someone’s speaking style or answer questions using their recorded stories. That is a digital representation, not evidence that the person’s consciousness has survived.
The distinction matters before you upload family recordings or pay for a service advertised as “digital immortality.” Decide what you want to preserve: an authentic recording, a searchable collection of memories or a system that invents new responses in someone’s style. Those are different projects.
An archive and a simulation do different jobs
An archive keeps material someone actually created: photographs, letters, interviews or recordings. A searchable archive can help you find a passage and return to its source.
A generative simulation can produce words the person never said. Even if the result sounds convincing, it remains a computer-generated response. An imitation of someone’s voice does not make the new message their opinion, their consent or a recovered memory.
NIST’s generative AI risk profile identifies confabulation—plausible but false or erroneous output—as a risk. In a family-history setting, that problem can be especially confusing because an invented detail may arrive in a familiar style.
A fictional example of the difference
Imagine a recorded interview in which a grandfather says he once worked in a bakery. A family member asks an AI replica, “What was the baker’s name?” The recording contains no answer.
A source-based archive should say that the information is missing. A simulation might produce a believable name and a story about an early morning delivery. The story may be moving, but there is no basis for adding it to the family record.
Keep the original audio, a checked transcript and any generated commentary separate. Label uncertainty explicitly. “The recording does not say” is a useful result when your goal is preserving history.
Consent includes the people receiving the experience
A 2024 paper by Cambridge researchers examined speculative scenarios involving AI simulations of deceased people. Its recommendations included transparency, mutual consent from people supplying data and people interacting with the simulation, and sensitive ways to retire a system.
These were ethical recommendations, not a clinical trial showing that a chatbot improves grief. Neither comfort nor distress should be assumed for every user. A family member’s enthusiasm is not permission to surprise someone else with a generated call or message.
Discuss boundaries before creating anything. Ask whether imitation is wanted at all, what material may be used and who may access the result. Leave room for someone to change their mind or decline participation without explanation.
Questions worth answering before paying
- Does the service replay original recordings, generate new material, or mix both?
- Can every factual answer point to a source?
- Can you export the original files and transcripts in usable formats?
- Who can create, share, pause or delete the representation?
- Can uploaded material be used for training, advertising or other purposes?
- What happens to access and files if a subscription ends or the company closes?
Also examine material involving living people. A letter may contain someone else’s private experience, even if you possess the paper. Remove material you do not have a sound basis to share. These are practical privacy questions; ownership and estate questions may require individual legal advice.
A useful legacy can be much simpler
The Library of Congress recommends identifying important photographs, organizing them with useful descriptions and maintaining copies in different places. That kind of preservation does not require creating an artificial personality.
Start with a small collection: a few photographs, the names of the people pictured, approximate dates clearly labeled, and one recorded story made with permission. Check a transcript against the recording. Use AI, if helpful, to suggest an index or draft questions for a future interview; verify its work before adding it to the archive.
A meaningful result might simply let someone hear an authentic story in its original context. Technology can help preserve and find that material. It should not quietly replace what a person actually said with what a model imagines they would say.
Updated September 23, 2026. This AI-assisted article explains the linked sources and includes clearly labeled illustrative examples. See our disclaimer.
Featured photograph: Roman Kraft via Unsplash. Stock photograph for illustration, not the fictional family described.
