A Stanford-backed analysis of 29.4 million LinkedIn profiles found that nearly one in five workers had retroactively edited the title or description of a previous job after leaving it. The striking part is the growth of AI-related edits: additions of terms such as “AI,” “GPT,” “LLM” and “artificial intelligence” to old roles have increased more than sixfold since ChatGPT launched in late 2022. The median edit happened more than four years after the relevant job ended.
The researchers estimate that a 2026 snapshot of LinkedIn could make 2022 appear about 30% more AI-skilled than it actually was. That does not necessarily mean workers are lying—some may simply be adding details they previously omitted or describing older work using terminology that became common later. Technology and information workers showed the highest rate of retroactive changes, at 31.6%, compared with 19.7% across the overall sample.
This creates a significant data-quality problem for AI-powered hiring and labour-market analysis. LinkedIn profiles are increasingly used as training data and signals for recruitment systems, so if professional histories are continually rewritten, models may learn an inaccurate picture of when particular skills actually became widespread. The timing is especially relevant in Europe, where AI systems used for recruitment and employment decisions entered the EU AI Act's high-risk category on August 2, bringing greater expectations around training-data quality, representativeness and accuracy.
The broader takeaway is that AI is changing not only what workers do, but also how their professional histories are recorded. As AI skills become highly valuable in hiring, people have stronger incentives to describe their past experience through an AI-focused lens. That means future employers and researchers may need to rely less on self-reported profiles and more on verifiable evidence of skills—actual projects, software usage, assessments and work outputs—because a polished digital career history is becoming increasingly difficult to treat as a fixed historical record.
