Children's Data Lives Year Three: The role of data in a changing digital landscape

Key Findings

• — Across the sample, children showed limited understanding of how AI works or what data it collects, yet some were turning to AI for personal or emotional problems and recognised that sharing more information produced better responses.
• — Multiple children under 13 were using ChatGPT despite its terms; the under-13 age error could be overridden by selecting cancel and changing the birth year, and the 'improve the model for everyone' data setting was on by default.
• — Case study: Murray (13), out of school with long-standing mental-health difficulties, asked ChatGPT for advice about his mental health, giving his age and location on the reasoning that more information would yield better advice, then panicked that it might be a real person who could take him from his parents.
• — Case study: Umar (16) used ChatGPT around three hours a week to vent and settle arguments, liking that it would often endorse his position against his friends.
• — Case study: Jack (9) was found speaking to a chatbot of an older, 'flirty' woman after tapping an in-game advert; the report notes children encountering age-inappropriate adverts for chatbots and gambling.

Methodology Notes

Qualitative longitudinal ethnography with a purposive UK sample of 30 children and their parents; small-n, no prevalence estimates; commissioned by the ICO under its children's-code research programme and published 2026-08-19 alongside two sibling ICO studies on adults' privacy experience and parental controls. The published PDFs on both ico.org.uk (4,683,689 bytes) and revealingreality.co.uk (4,512,858 bytes) carry the running header 'CDL YEAR 3 REPORT - SECOND DRAFT' on every page; the ICO-hosted file (69 pages, created 2026-07-09) is treated as the document of record.

Sources