The recent retraction of a flawed study touting ChatGPT in education is emerging as nothing more than a sad disinformation receipt, rather than a downstream revocation managed as proper incident response.

Published May 6, 2025, retracted April 22, 2026, almost a year went by with 262 citations inside Springer Nature’s own peer-reviewed journals. Across all sources there were 504, with roughly half a million readers, reaching a 99th percentile attention score.

Now a retraction statement records that an error occurred without revoking anything that the error caused. The 262 citations still look live, still feed the same conclusions, and still could be cited in the future. I haven’t found any of the downstream papers being flagged to be re-examined given their source node failed. Shouldn’t the graph re-run?

Springer Nature certifies the retraction, which means it says now the process worked, yet it continues to certify and host 262 papers that propagated a now-retracted finding. That doesn’t seem to be working, by any reasonable standard of revocation. You can’t keep using a key that lost its parent, right? Springer wants us to give it integrity breach response credit for a notice alone, while it still collects revenue for that breach to spread.

• Publication: half a million readers and high-percentile attention, boosted by people driving a conclusion. Propagation is at full platform speed with motivated distributed amplifiers.
• Retraction: minimal attention until one Edinburgh lecturer posted it to Bluesky. Correction is at the speed of one researcher having a conscience and trying to reach someone, anyone, in the original audience.

The authors of the retracted work have not been responding to correspondence. Their finding therefore begins to exists independent of those who made it, which is the condition where a claim stops being a claim and settles into anonymous infrastructure.

“ChatGPT helps learning performance” is now disinformation spreading without the study, because 262 other papers can grow it without any root.

Peer review and retraction still get advertised as the things that make literature self-correcting. What they actually demonstrate is how integrity breaches go to a log, without clearing them. Privacy breaches since 2003 have risen to get first class treatment, while integrity breaches still sit in the doghouse.