Copy article

63% of people forget whether an idea was theirs or AI's within a week, new cognitive science study shows

ended 19. May 2026

"The AI Memory Gap: Users Misremember What They Created With AI or Without" a peer-reviewed study presented at CHI 2026 by Zindulka, Goller, Fernandes, Welsch and Buschek, asked 184 people to come up with ideas and write them up, sometimes alone, sometimes with an AI chatbot. A week later, they were asked who contributed what. When people came up with the idea but got AI to write it up, or the other way round, they could only correctly identify the source 37.7% of the time, statistically worse than tossing a coin.

People who used AI for everything or nothing did better. It's the mixing that wrecks recall, stripping away the mental markers humans need to trace their own thinking back. Under 10% of participants had even guessed the study was about memory, so nobody was keeping notes.

Participants overestimated their own accuracy by 12%. In mixed workflows, how confident someone felt had no connection to whether they were right.

This matters for any business expected to declare where AI helped produce a piece of work. If the people doing the work can't reliably remember, the declaration is merely a confident guess not a formal disclosure.

We want your views:

  • Have you ever looked back at something you wrote with AI and genuinely couldn't remember which bits were yours?
  • If you're managing a team using AI tools, does it matter how they use them, not just whether they do?
  • Businesses are increasingly asked to disclose where AI contributed. If the person doing the work genuinely can't remember, what is that disclosure actually worth?
  • If accurate activity logs must be made in real time, how much does that documentation and cognitive overhead eat into the productivity gains AI is supposed to deliver?
     

4 responses from the Newspage community

Copy all

Star Quote
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
If the memory research holds, and people genuinely can't reconstruct what was theirs after a week, then the only reliable alternative is logging as you go. Every prompt saved, every suggestion accepted or rejected, every edit tracked. That's a second job running underneath the first one. The productivity case for AI assumes you save time. Nobody's costing in the time it takes to prove you saved time.
Copy

Editor
Katrina Young
Futurist | Chief Technology Officer CTO/CDO Ethical AI & Digital Transformation Strategist at KYC Digital
The study matters because it challenges a major assumption behind AI disclosure rules: that people can accurately remember how a piece of work was created. In reality, most workflows now involve constant switching between human writing and AI assistance through prompting, editing and rewriting. A week later, many participants could no longer reliably tell which parts came from them and which came from the AI. That creates a problem for businesses increasingly being asked to declare where AI was used. If disclosure depends on human memory rather than system records, it may be far less reliable than organisations assume. The bigger issue is whether AI use can still be properly tracked once human and machine contributions become blurred inside everyday work.
Copy

Editor
Mitali Deypurkaystha
AI Governance Strategist & CMO at AISensa
14% of workers already report symptoms of “AI brain fry”, according to research highlighted by Harvard Business Review and BCG. Yet this new study suggests the deeper problem may be authorship amnesia: people genuinely losing track of which ideas were theirs and which came from a machine. Humans have always absorbed other people’s thinking and mistakenly internalised it as their own, but normally those claims can be challenged through drafts, conversations or evidence trails. AI changes that because the collaboration happens so quickly and invisibly that confidence can survive long after memory accuracy collapses. That creates a serious problem for businesses demanding AI disclosure audits, because a declaration based on reconstructed memory may be little more than a sincere guess. The danger is not deliberate deception, but workplaces drifting into a world where nobody can reliably prove where human judgment ended and machine influence began.
Copy

Editor
Kate Underwood
Founder & Chief People Strategist at Kate Underwood HR and Training
Small businesses need a policy and a simple process for AI use because, in mixed workflows, our brains are basically unreliable narrators. If you wrote the idea and AI polished it, or AI gave you the bones and you made it human, a week later you’ll swear blind which bits were yours and still be wrong. If you don’t log AI use as you go, your ‘disclosure’ later is just a confident guess.
That matters because we’re heading into a world where clients, regulators, insurers, and sometimes lawyers will ask, ‘Where did AI contribute?’ ‘I’m pretty sure’ isn’t a process, it’s a risk.
The answer isn’t a 12-page policy nobody reads. It’s a simple rulebook and a tiny habit: save the prompt, label the doc ‘AI assisted: first draft’ or ‘AI used for editing’, and keep a quick log. It takes seconds, and it stops you getting caught out later trying to reconstruct the truth from memory and vibes.