A short passage can feel settled by the fourth reading. The next sentence is almost visible. The person would bet on themselves. Ask for the same passage on a blank page once the week has turned, and a large part of that certainty has left. The feeling of knowing it was real. It was also the wrong receipt.
Henry L. Roediger III and Jeffrey D. Karpicke, both at Washington University in St. Louis, reported the comparison in Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention, Psychological Science, 2006. The materials were two prose passages from a test-preparation book, one on the sun and one on sea otters, 256 and 275 words, each scored as 30 idea units. The tests were free recall. People wrote what they could remember, wording and order did not have to match, and nobody was shown the passage again afterwards. There was no feedback.
Four readings against one
In the second experiment, 180 undergraduates took part, 30 in each of six groups. Everyone learned one passage across four periods. One group studied it four times and took no test in that stretch. One group studied it three times and then recalled it once. One group studied it once and then recalled it three times. Half of each group was tested again after five minutes. Half came back after a week. During the study periods they kept a tally of how many times they got through the passage. The four-reading group went through it 14.2 times on average. The three-reading group went through it 10.3 times. The group that studied once and then recalled went through it 3.4 times.
Five minutes later, recall followed the readings. The four-reading group remembered 83 per cent of the idea units. The three-reading group remembered 78 per cent. The group that had studied once and been tested three times remembered 71 per cent. On that clock, rereading won. It is the result a form on the way out will catch, because the form goes out while the passage is still warm.
What the week kept
A week later the order reversed. The group that had studied once and recalled three times remembered 61 per cent. The group that had studied three times and recalled once remembered 56 per cent. The group that had only reread remembered 40 per cent. Roediger and Karpicke also report the drop as a proportion of what people recalled at five minutes: 14 per cent forgotten in the repeated-test group, 28 per cent in the single-test group and 52 per cent in the rereading group. The people who had seen the passage the most times kept the least of it.
They had also predicted the better week. At the end of the learning session everyone rated how well they thought they would remember the passage in a week, on a seven-point scale where a higher number meant more confidence. The four-reading group sat at 4.8. The three-reading group sat at 4.2. The repeated-recall group sat at 4.0. The paper reports that the rereaders were significantly more confident than either of the other two, who did not differ from each other. Confidence tracked the condition that lost. The first experiment, with 120 undergraduates and the same two passages, showed the same crossover. Extra study beat a test at five minutes, 81 per cent to 75. A test beat extra study after two days, 68 per cent to 54, and again after a week, 56 per cent to 42. No feedback in either experiment.
What a demo measures
An AI training day that walks the same prompt four times, with the answer still on the slide, is the four-reading condition. People will feel ready. They have just seen the passage 14 times. The check you wanted them to own, what to do with a fluent wrong answer, what not to paste into the customer's note, when to stop and look at the source, is an idea unit. If they never had to produce it onto a blank page, you measured the five-minute test and called it capability.
The blank page is the point of the design, and it is also the uncomfortable part. Recall without feedback still beat rereading a week later. The repeated-test group was not being shown the model answer between attempts. They were being made to find the route to it. A day that reveals the answer the moment someone hesitates has removed the attempt. You will get a smoother room and the 40 per cent week.
What the day should make someone produce
Pick one workflow the team already runs, and one check that has to happen inside it. Teach the check once, on a real draft, with the slide up. Then shut the slide. The person writes the check in their own words and runs it on a second draft they have not seen. They do that again before they leave, still without the slide. You are not quizzing them on sea otters. You are giving the Monday task the same shape as the condition that remembered 61 per cent, rather than the one that remembered 40.
Do not take the confidence rating as the result. The rereaders would have given you the better number, and they were wrong about their week. Take a sample a week later, on a draft that was not in the room, and see whether the check is still there when nobody is watching. If the only record is the form on the way out, you bought the five-minute test.
Collective Campus has done capability building for teams since 2015. The AI training workshop is for a team that can bring one live workflow and will let people produce the check with the slide shut. What that day should leave behind is in what a corporate AI training day should produce. The belief people walk in with, and how it rewrites the lesson, is in the belief people bring to the day. Moving the work off a completion certificate is in from compliance AI training to judgment.
What to bring
Bring one workflow, one check inside it and a draft the team did not write the slide from. If the answer stays on the screen for the whole session, you have booked the rereading.
