Workers Boost AI Skills on LinkedIn: 2022 Data Shows 30% Rise
▼ Summary
– A study of 29.4 million LinkedIn profiles reveals that nearly one-fifth of users retroactively edited past job descriptions after leaving those roles.
– Since the launch of ChatGPT, mentions of AI-related terms in these retrospective edits have increased more than sixfold.
– The researchers term this behavior ‘time travel,’ noting that median edits occurred over four years after the relevant employment ended.
– This practice leads to a significant distortion in historical data, with a 2026 snapshot potentially overstating 2022 AI skills by approximately 30%.
– These findings raise compliance concerns for AI recruitment systems under the EU AI Act, which requires training data to be accurate and representative.
A significant portion of professional histories on LinkedIn are being rewritten after the fact, creating a distortion in how labor market skills are tracked. Research analyzing 29.4 million US profiles reveals that nearly 20% of users have modified job titles or descriptions for positions they had already left. This trend has accelerated sharply with the advent of generative AI, leading experts to estimate that current data overstates the prevalence of artificial intelligence skills from 2022 by approximately 30%.
The phenomenon, described by researchers as “time travel,” involves updating past employment records to reflect current technological trends. Since the launch of ChatGPT in late 2022, the inclusion of keywords such as “AI,” “GPT,” “LLM,” and “artificial intelligence” in previous roles has increased more than sixfold. These retroactive edits typically occur long after the relevant employment period has concluded, with the median adjustment landing more than four years after the job ended.
This behavior is not uniform across all industries. The practice is most prevalent among technology and information professionals, where 31.6% of profiles show such modifications. This rate surpasses those in arts and entertainment (25%) and professional and scientific services (24.1%), both of which exceed the overall study average of 19.7%. Conversely, other trending topics are seeing different patterns. By late 2025, the addition of remote-work terminology balanced out with its removal, while mentions of diversity and inclusion terms dropped significantly in early 2025.
While these edits might seem like mere embellishments, the authors caution against assuming immediate dishonesty. Workers may simply be adding details previously omitted or reframing older tasks using contemporary vocabulary. However, the aggregate effect creates a substantial measurement error. For labor-market research and algorithmic hiring systems that rely on this data, the discrepancy is critical. In an environment where automated tools increasingly screen candidates, inaccurate historical data can skew results and misrepresent workforce capabilities.
The implications extend beyond data accuracy into legal compliance, particularly in Europe. As of August 2, AI systems used in recruitment and employment decisions were classified as high-risk under the EU AI Act. Article 10 of the legislation mandates that training data for these systems must be relevant, representative, and free of errors to the greatest extent possible. Although regulators did not specifically anticipate profile editing when drafting these rules, the new findings provide concrete evidence of potential data contamination. Companies developing hiring models now face a quantifiable challenge: their training sets include a known percentage of career history that was constructed retrospectively, potentially violating the integrity standards required by law.
(Source: The Next Web)
