Editorial
Corrections
Last updated .
When I get something wrong, I fix it, and I say so. A changed fact, number, name, date, attribution or conclusion always gets a note on the page and an entry here. A fixed typo does not.
If you think there is an error, write to [email protected] with the page and what you believe is wrong. I would rather hear it from you first, and I do not edit a mistake out quietly.
Corrections log
- AI feedback on student writing, November 13, 2026: The confidence section said the two experimental school-age studies of older tools found no significant overall advantage for the tool. That blended two comparisons: Ware (2014) found no evidence of improvement against online teacher or peer feedback, and the Grade 7 and 8 trial found no significant overall effects against students who did not use the tool. The sentence now states each comparison and notes that the positive high school trial tested generative feedback. In the evidence table, the Ware row now gives the review's wording, and the label on the Mah and colleagues row no longer says model feedback matches teacher feedback, which that study did not find. The case for optimism said the Grade 4 study 'is large', which could be read as describing its sample of 60 pupils; it now says the study reports a large effect. The funding statement said I had no financial relationship with any AI feedback vendor and, in the same paragraph, that I pay Anthropic for the model used here; it now names that payment as the one exception. The opening now says the high school trial is known only from another paper's description, and the Method now lists Steiss and colleagues among the sources used only as described by other authors.
How I decide what to publish is on the standards page.