Skip to content

Together

A Foster Parent Used AI but Kept a Child’s Memory Gaps Visible

A chatbot helped organize a child’s fragmented memories, but it also supplied details that sounded real. His foster parent built the life-story notebook around what remained uncertain.

Nadia RelfNarrator, Together

August 9, 2026 · 7 min read

An open notebook with memory fragments, question marks, and a closed laptop on a kitchen table.
An open notebook with memory fragments, question marks, and a closed laptop on a kitchen table.

The notebook had been on Lena’s kitchen table for six weeks before she considered asking a chatbot for help.

Her foster son, Micah, had lived with her for eleven months. He was ten. The notebook was part of his life-story work, a place where he could gather what he remembered alongside information that adults had recorded about where he had lived and who had cared for him.

Some pages came easily. Micah remembered a dog sleeping under a kitchen table. He remembered a laundromat where the vending machine did not work. He knew which elementary school had a garden beside the playground, although he could not place that school before or after the apartment with the outside stairs.

Other pages held one line and a question mark.

Lena did not expect a child to produce a clean timeline. Still, she found herself stalled by the loose pieces, especially when two memories seemed as if they might belong together but neither of them carried enough context to say so. The notebook was meant to give Micah some continuity. She worried that her efforts to create continuity could quietly replace what he knew with what an adult found likely.

A chatbot appeared to offer a modest kind of help. Lena had used one to reorganize notes for household tasks, and she wondered whether it could sort short, de-identified memory fragments without changing them. She did not want it to write Micah’s story. She wanted help seeing a possible order.

That distinction felt clear until she tried it.

The detail the chatbot supplied

Lena copied a few altered fragments into the chatbot. She removed names, locations, school references, and the details of Micah’s case. The fragments included outside steps, an adult arriving to pick up a child, and uncertainty about whether this happened near lunchtime.

She asked the tool to arrange the fragments without adding facts.

The response was smooth. It connected the steps to a school day, placed the pickup after school, and supplied a car color. None of those details had appeared in Lena’s prompt. They were ordinary additions, which made them harder to catch than an obvious mistake would have been.

Lena wrote the result in the notebook’s margin, then stopped. Beside the car detail, she printed four words: AI ADDED THIS. NOT MEMORY.

The line changed the page.

When Micah later glanced at the draft, Lena mentioned the car without thinking much about it. He paused and said it might have been that color. A moment later, he offered another color. He did not seem upset, and Lena did not press him, but the exchange stayed with her because the chatbot’s invention had already entered the space where he was trying to decide what belonged to him.

She crossed out the whole generated sentence. The original fragments remained underneath, less readable as a story and more faithful to what they were.

At the next meeting with Micah’s counselor and caseworker, Lena brought the notebook open to that page. She was embarrassed by the margin note, though neither adult treated it as a confession. They looked at the crossed-out sentence and talked about the difference between organizing language and generating narrative, a boundary the tool had crossed even after Lena had asked it not to.

The problem was not that the invented car color carried great significance. It was that a plausible sentence could gain authority from its fluency, while Micah’s real recollections often arrived with pauses, revisions, or no sequence at all.

His uncertainty deserved a place on the page.

A notebook that did not resolve every gap

The adults agreed that the notebook would distinguish between Micah’s memories and information from adult sources. Lena drew a line down the middle of the next open page. One side held what Micah remembered in his own terms. The other held context from records or caregivers, with its source identified in general language.

A question mark did not mean the work was unfinished. It meant the answer was not known.

They also kept chatbot output out of the notebook unless each phrase could be traced back to words already provided by Micah or an adult source. In practice, that meant very little generated prose survived. The tool was more useful when it grouped altered fragments by broad setting or pointed out that two entries might refer to the same period, which gave the adults something to check rather than something to tell Micah.

Lena did not turn this into a lesson about technology for him. The counselor explained that the computer could arrange words but had no memory of his life. Micah was allowed to reject a connection without defending his choice, and on one page he separated two apartments that the chatbot had grouped together because, to him, the kitchens felt different.

There was no record that settled it. The pages stayed separate.

That decision mattered to Lena more than the possible chronology. For four months, she had watched Micah look to adults for confirmation when he was uncertain, sometimes changing a small detail after hearing another person’s version. She did not interpret that for him. She became more restrained about filling silence.

The notebook changed slowly. A memory about eating cereal on the floor remained by itself. An adult record placed Micah in a particular home during that period, but Lena wrote that information on the source side of the page rather than joining the two accounts. The line down the center kept them near each other without claiming they were the same event.

Micah sometimes chose not to work on it. The notebook returned to a drawer until he brought it out again.

What they kept away from the tool

Privacy had worried Lena before the invented sentence, though she had first understood it as a question of removing names. The discussion with the other adults made the concern wider. A child’s routines, family relationships, and unusual combinations of places could be identifying even when obvious details were gone.

Lena had once considered photographing notebook pages and asking the chatbot to turn them into a timeline. She abandoned that idea. The notebook never went into the camera frame, and no court information or raw case material was entered into the platform.

For the limited experiments that followed, she retyped short fragments and altered incidental details that were not needed for grouping. She kept the chatbot’s output in a separate caregiver file long enough to review it with the other adults, then removed the working copy. They did not treat deletion as proof that the platform had never retained data. Their main protection was reducing what entered it in the first place.

This made the tool less efficient. It also made its role smaller.

Lena noticed that she had initially been drawn to the chatbot because the blank spaces in the notebook made her uneasy. Once the adults stopped treating those spaces as problems to solve, the work felt different. AI could occasionally help them notice a possible connection, but it could not decide whether the connection belonged in Micah’s account.

Near the back of the notebook, he added another memory about the dog under the kitchen table. He still did not know which home it belonged to. Lena wrote the sentence where he wanted it and left the rest of the page open.

Questions people ask

Can a chatbot organize a foster child’s fragmented memories?

In Lena’s experience, it could group fragments or suggest a possible sequence, but it also added details that had never been supplied. The adults treated its output as a prompt for review, not as part of the child’s history, and most generated prose never entered the notebook.

How did they keep

AI additions from becoming part of the story?

They kept generated text separate and traced any surviving phrase to Micah’s words or an identified adult source. The notebook’s divided pages showed which account came from him and which came from elsewhere, while question marks remained visible where neither side could establish what happened.

Did the chatbot replace the child’s counselor or caregiver?

No. The tool had no direct role in deciding what Micah remembered or how the work should proceed. His counselor, caseworker, and foster parent reviewed its limits together, while Micah retained control over whether two fragments belonged together and whether he wanted to continue.

What happened to the child’s private information?

Lena did not upload notebook images, records, names, or raw case details. She used altered fragments for limited tests and kept the output apart from Micah’s story, without assuming that deleting a chat erased every copy. The notebook itself went back into the locked file box.

ShareFacebook
relationship trustchild safetyprivacyfoster careartificial intelligencecaregivingchild privacyrelationships

One story a day

The story of the day, in your inbox

One real story about AI each morning — no hype, no alarm, just company for the road.

Read next