Their Mother Smiled at the Chatbot’s Story. It Wasn’t True.

Mara opened a notebook in August 2024 and wrote down the $24 monthly charge for the chatbot. Under it, she recorded the family materials she and her brother Daniel had entered into the tool.

The notebook was supposed to keep them honest.

Their mother, June, was living with dementia and had stopped recognizing many of the stories her children told. A photograph from a lake trip still held her attention, but the names attached to it often did not. Some days she knew Mara as her daughter. On others, she accepted her as a familiar visitor and seemed relieved that nobody pressed for more.

Mara and Daniel wanted conversation that did not feel like a test. They had watched June withdraw when relatives corrected her or asked her to identify people in photographs, and they hoped a tool built from family records could offer gentler openings.

For two months, they gathered captions from albums and passages from their father’s calendars. Mara typed up part of a letter June had written after a family trip in 1984. Daniel added notes about meals, music, and places their mother had once mentioned often.

The chatbot turned those details into short reminiscences for them to read aloud. It did not claim to be June or their late father, Paul. Still, its calm narration carried authority, particularly when it placed a familiar name beside an event June could almost reach.

At first, that seemed useful. A passage about making sandwiches before a lake trip prompted June to describe how Paul always packed too much food. Mara could not tell whether her mother remembered that particular day, but June stayed with the subject and laughed when Daniel mentioned the uneaten apples that came home.

Mara marked the entry in the notebook as a good visit.

The story that crossed two years

The problem appeared three weeks later.

Mara had prompted the chatbot with records from the 1984 lake trip and a granddaughter’s wedding in 2019. She expected separate passages. Instead, the generated story placed Paul at the wedding, then described him putting its program into the picnic basket from the earlier trip.

Paul had died seven years before the wedding.

Mara noticed the error as she read, but not soon enough to stop the scene from forming. June smiled at the detail about the saved program. She said, indirectly and with confidence, that Paul had always kept pieces of paper that mattered.

For a few minutes, she was settled. Then she began waiting for him.

She picked up her purse and moved toward the front door, telling Mara that Paul would be returning from the wedding. When Mara tried to redirect her toward the kitchen, June became distressed and insisted that she could not leave him outside.

The door chime sounded when June pulled it open. Mara was beside her and closed it without a struggle, but the moment changed the family’s understanding of the tool. A false detail had not remained inside a pleasant conversation. It had shaped what June believed was happening in the house.

In the notebook, Mara joined the two source years with a line. She noted that the story had brought calm first, then searching and distress. The page made the sequence harder to soften later.

Daniel saw the same entry differently. Their mother had spent part of the visit speaking warmly about Paul, he said, and he did not want one difficult ending to erase that. He lived several hours away and carried his own discomfort about the care Mara provided in person, although neither sibling named that feeling during their first argument.

Mara thought the tool had given June a memory she could not check and then left the family to manage its consequences. Daniel thought familiar emotional material had reached her when accurate retellings no longer did.

Both views remained in the room.

They stopped using the chatbot with June for six weeks. During that pause, Mara read through 31 notebook entries covering four months. She marked nine passages that had blended people, dates, or locations, even though each individual detail came from something the family had supplied.

Some errors were small. An uncle’s job appeared in the wrong decade. A birthday meal moved from a restaurant to June’s former home. Others changed who had been present, which mattered more when the absent person was dead or estranged.

The notebook also showed that factual accuracy had never guaranteed recognition. One passage stayed close to June’s own 1984 letter, but she listened to it and said the family described must belong to somebody else. A generated story could be correct and still feel foreign to her. Another could be false and produce a smile.

That contradiction did not disappear when Mara underlined it.

What the family records exposed

The pause brought a second concern into view. The source material was not June’s alone.

Album captions included the names of relatives who had never agreed to have their information entered into a chatbot. The typed letter referred to a family estrangement and an old address. Mara had focused on whether June would have wanted her own memories used this way, while giving less thought to the other people contained inside them.

June had agreed in broad terms when Daniel first described the project. Later, she did not recognize the tool or remember that conversation. Mara could not tell how much the earlier agreement covered, particularly once the family began entering private correspondence rather than familiar anecdotes.

The platform’s privacy page explained how users could remove visible conversations, but Mara could not determine from the language whether submitted text might be retained elsewhere or used to improve the system. She wrote to the support inbox and received general information that did not resolve every concern.

The siblings deleted the uploaded family material they could see. They kept copies of the records outside the platform, including the notebook, which had become more valuable than the generated stories because it showed where the tool had joined facts that the family itself had kept apart.

Their disagreement eased without being settled. Daniel stopped describing the false wedding story as harmless. Mara stopped saying the warm part of the exchange had meant nothing.

When they resumed using the tool, they did not ask it to create reminiscences for June. Daniel used it privately to sort typed notes by decade, then checked the results against the original records. Mara sometimes took a single verified caption into a visit and read only what was on the page.

That narrower use lasted two months. It produced fewer surprises, though it also removed much of what had attracted them to the chatbot: the flowing language that could carry June into a conversation without requiring her to supply the facts.

They mentioned the incident during a routine appointment with a member of June’s care team. The chatbot was not part of her treatment, and the family did not treat the conversation as an endorsement or a set of medical instructions. They described the door incident and showed the notebook entry because those were the details they could report.

This account is not professional medical advice. It describes one composite family’s choices around a tool that generated plausible language from private records, in a situation where recognition, consent, and distress could change from one visit to another.

Mara eventually canceled the $24 subscription. Daniel saved a plain document containing the verified captions, while the generated passages were removed from the folder they shared.

On a later visit, Mara brought the lake photograph without a prepared story. June studied it, touched Paul’s image, and did not identify him. She kept the photograph beside her through lunch.

Mara recorded that in the notebook too.

Questions people ask

Can a memory chatbot combine separate family events?

It did in this family’s experience. The tool drew from genuine records but placed details from 1984 and 2019 into one plausible narrative, including a person who could not have attended the later event. The fluent wording made the blend less obvious until Mara compared it with the notebook.

How did the family respond when the story caused confusion?

They paused the chatbot for six weeks and reviewed Mara’s dated entries, which showed both comforting conversations and nine factual blends. After the door incident, they stopped presenting generated reminiscences to June and used verified captions instead, although the siblings continued to differ over how much the earlier moments of comfort had mattered.

What privacy concerns came from uploading family records?

The records contained information about relatives as well as June, including an estrangement and an old address. The family removed the material visible in the account, but the platform’s general privacy language did not settle every question Mara had about retention. That uncertainty became part of their decision to cancel the subscription.

Was the chatbot part of June’s dementia care?

No. Her children tried it as a family conversation tool and later described the distressing incident to a member of her care team. It was not presented as treatment or medical guidance. Mara brought the notebook to the appointment, open to the page joining the 1984 trip with the 2019 wedding.