False memory, repeated narratives and the governance record

Memory is not a recording. It is a reconstruction. Every time a person recalls an event, they rebuild it from fragments. And those fragments can be altered by intervening information, social context and repeated exposure to alternative accounts of what happened. This is not a flaw unique to impaired cognition. It is how human memory works. In project governance environments where AI systems now generate, curate and repeat narratives at scale, this creates a specific and largely unacknowledged delivery risk. 

The evidence on false memory

Decades of research on false memory have established that human recall is profoundly susceptible to post-event suggestion. In one landmark study, participants who watched footage of a car accident and were asked ‘how fast were the cars going when they smashed into each other?’ reported significantly higher speeds than those asked, ‘when they hit each other’, and were also more likely to falsely remember broken glass that was not present. The word choice alone was enough to alter the memory of what was seen. 

Subsequent research has shown the same effect in professional and organisational settings. When people are repeatedly exposed to a particular account of what happened, that account progressively overwrites their actual recollection, not because they are dishonest but because memory does not preserve originals.

What this means for AI-assisted governance records

In project governance, the false memory risk operates through a mundane but consequential mechanism. When AI-generated summaries, retrospective analyses or meeting minutes describe what was decided or what the data showed at a previous point, those descriptions become the reference reality for everyone who reads them. 

If the AI summary is inaccurate, whether through hallucination, selective emphasis or the framing choices embedded in its training, that inaccuracy becomes the group’s shared memory of what happened. Project teams routinely depend on AI-assisted records of decisions, risks and commitments. Each document is an opportunity for small inaccuracies to compound into a collectively held false account of the project’s history, one that becomes progressively harder to challenge as it is repeated and embedded in subsequent papers.

The illusory truth effect in project status

Closely related is the illusory truth effect: repeated exposure to a claim increases the likelihood that people will judge it as true, regardless of its actual accuracy. In project governance, status narratives are repeated with high frequency. The project is on track. The risks are manageable. The sponsor is engaged. 

When these narratives are AI-generated and repeated across board papers, progress reports and stage gate reviews, they acquire a cumulative credibility that may have nothing to do with the underlying project reality. The illusory truth effect is not a sign of cognitive weakness. It is a structural feature of how human judgement responds to repeated information. AI-enabled governance reporting creates this effect at scale.

Protecting the governance record

The practical response is not to distrust AI-generated records but to govern them deliberately. That means dating and versioning AI summaries clearly, preserving original source materials alongside AI-generated syntheses, and requiring human review of governance records before they become the official account of what happened.

It also means building in deliberate challenge. At regular intervals, ask project teams to recall what was actually decided and agreed at previous points, and compare their recollection with the documented record. Gaps between the two are governance intelligence.

Practice checklist

• Date and version AI summaries: Any AI-generated governance record should carry a date, a version number and a note that it was AI-assisted. Treat it as a draft until a competent human has reviewed it. 
• Preserve original source materials: Keep the underlying data and documents alongside AI-generated syntheses. The synthesis may become the group’s memory; the originals are the check on that memory.
• Build recollection checks into governance: Periodically ask teams what they believe was decided at previous review points and compare with the record. Divergence is useful information.
• Name the illusory truth risk: When status narratives are being repeated across multiple documents, ask explicitly: is this still accurate, or has repetition made it feel true?
• Review before it becomes official: No AI-generated governance record should become the agreed account of a meeting, decision or review without human review and sign-off. 

 

You may also be interested in:

• Effective Communication in Projects
• What is artificial intelligence in project management?
• AI-Driven Governance: Intelligent Stakeholder Strategy for Mega-Infrastructure Projects