Vividness of Visual Imagery and Incidental Recall of Verbal Cues, When Phenomenological Availability Reflects Long-Term Memory Accessibility

Close your eyes and picture a lemon — the waxy yellow skin, the weight of it in your hand, the faint smell of citrus. Now ask yourself: is the vividness of that image doing anything useful, or is it just the feeling of remembering? For decades, most researchers assumed vividness was epiphenomenal — a readout, not a cause. A symptom of good memory, not a driver of it. D'Angiulli and colleagues conducted experiments to find out who was right. The field's standard tool for measuring imagery has been the Vividness of Visual Imagery Questionnaire, the VVIQ, which asks people to reflect on their general ability to form mental images. It captures something like trait vividness — a stable individual difference. Past literature found it correlated with memory outcomes at an average of about a correlation coefficient of 0.13. Modest. Barely above noise. But D'Angiulli and colleagues suspected the VVIQ was measuring the wrong thing — that what actually matters isn't how vivid your images are in general, but how vivid any particular image is, right now, in the moment you form it. That's the distinction between trait vividness and state vividness, measured trial by trial. And that distinction turns out to be everything. In Experiment 1, twenty-six university students worked through sixty verbal cues — short noun descriptions of objects, both animate and inanimate — carefully matched on word frequency, imageability, concreteness, and reading time. On each trial, a cue appeared on screen. Participants read it, generated a visual image, and when the image felt complete and at its most vivid, they pressed a button. A seven-point scale then appeared, running from "no image" at one end to "perfectly vivid" at the other. They rated each image individually. No response deadline. No hint that memory was being tested. Then, thirty minutes later, the surprise. Participants were handed a blank sheet and asked to write down as many of the cues as they could remember. They hadn't been told to encode anything. This is what researchers call incidental recall — memory formed without intent, without strategy, without rehearsal. It's a cleaner window into what the mind naturally retains. The result was striking. Cues that had been rated as vivid — scores of five, six, or seven — were recalled at a mean rate of seventy-seven percent. Cues rated as non-vivid — scores of two through four — were recalled at only nineteen percent. That gap, a paired t-test confirmed, was highly significant and explained seventy-four percent of the variance. Three-quarters of the variance in memory, predicted by a single rating made thirty minutes earlier. Individual differences — how vivid a person's images tended to be overall — explained only fourteen percent of the variance in total recall, a significantly weaker effect. So it wasn't just that some people are better imagers. The trial-by-trial signal, that moment-to-moment vividness, was doing most of the work. Now here's where it gets interesting. The researchers also recorded how long participants took to generate each image — the response time, from cue appearance to button press. They found that vivid images were generated faster. Using a statistical tool called ex-Gaussian modeling, which separates the fast, regular part of a response-time distribution from the occasional long deliberations captured in an exponential tail, they showed that less vivid images were shifted more than five hundred milliseconds later on the time axis compared to vivid ones. The model fit was excellent, capturing at least sixty-eight percent of response-time variance at every vividness level. So vividness and speed are linked. But here's the dissociation that matters: speed itself did not predict recall. Mean generation times were fourteen point eight seconds for vivid items and thirteen point three three seconds for non-vivid items — a non-significant difference. Adding response time to a logistic model predicting recall changed nothing; the regression coefficient was essentially zero, with a p-value of 0.587. The logistic model built on vividness and stimulus identity predicted recall with an overall success rate of seventy-two point two percent, correctly classifying eighty-three point seven percent of items that went unrecalled and fifty-four point seven percent of recalled items. Speed contributed nothing on top of that. This is the critical wedge. Vividness and latency are correlated with each other, but they have completely different relationships with memory. Vivid images come faster, yes — but it isn't the speed that helps you remember. This rules out a simple depth-of-elaboration account, the idea that people just tried harder or thought longer on the images they later remembered. If that were the story, longer generation times would predict better recall. They don't. Experiment 2 asked a different question: if trial-by-trial vividness and the VVIQ are both measuring imagery, why does one predict memory and the other barely does? Thirty-nine female undergraduates completed the VVIQ2 — the revised questionnaire — in a screening session, then returned to rate the vividness and perceived speed of images generated from seventeen static and seventeen dynamic scene descriptions. About three percent of trials were excluded for "no image" responses. The behavioral pattern was already revealing. Trial-by-trial vividness was similar across static and dynamic conditions — means of five point three one and five point two nine on a seven-point scale, respectively. But VVIQ2 scores, translated onto the same scale, averaged five point six eight — significantly higher than both. Participants thought their images were more vivid when reflecting globally than when rating them one at a time in the lab. The two measures weren't agreeing with each other. The correlations sharpened the picture. VVIQ2 correlated with static trial vividness at a correlation coefficient of 0.51, but not reliably with dynamic vividness at a correlation coefficient of 0.26. And the questionnaire showed no relationship with perceived imagery speed in either condition. By contrast, trial-by-trial vividness and perceived generation speed were strongly inversely linked in both conditions — faster-feeling images were rated as more vivid. The questionnaire and the trial-level measure were picking up different signals. A small meta-analysis, drawing on a randomly selected sample of sixty-six articles, reinforced this. Across twenty-one entries reporting correlations between VVIQ scores and trial-by-trial vividness ratings, the average correlation was 0.15. The meta-analysis also found that a higher proportion of significant behavioral, cognitive, and neural outcomes were associated with trial-by-trial measures than with the VVIQ. The questionnaire isn't wrong, exactly — it captures something real about a person's global imagery capacity. But it lacks the resolution to capture what's happening on any given trial. To explain why trial-by-trial vividness does predict memory, D'Angiulli and colleagues turn to multi-trace memory theory, or MMT, developed by Moscovitch and colleagues. Under MMT, memories are stored as multiple overlapping sensory traces distributed across the cortex, indexed early on by the hippocampus and progressively integrated into cortical networks through consolidation. The argument the paper makes is that vividness ratings reflect the availability of those long-term sensory traces — the perceived clarity of an image is proportional to how many traces have been consolidated and can be reactivated right now. This is the distinction the paper's title flags, between availability and accessibility. Availability refers to whether the traces exist in long-term memory at all. Accessibility refers to whether they can be retrieved in this moment. The data support the view that phenomenological vividness — the felt intensity of an image — indexes availability and is not merely epiphenomenal. When more traces are accessible, the image feels more vivid, and the odds of later recall go up. When fewer are accessible, the image is dim, and the cue slips away. The practical implication for the field is narrow and clear. For years, researchers tested whether imagery vividness matters for memory, reached for the VVIQ, found modest effects, and moved on. But the VVIQ and the trial-by-trial measure are not interchangeable. They correlate at a correlation coefficient of 0.15 on average. Using one to test claims about the other is like using a thermometer to measure humidity — the constructs are related, but the instruments are not equivalent. The next time a mental image feels especially clear and present — sharp edges, saturated color, a sense that the thing is almost there — that clarity is not decorative. It's informative. It's telling you something about the state of your memory, about which traces are currently active and available. D'Angiulli and colleagues didn't just demonstrate that vivid images are remembered better. They showed why the feeling of vividness deserves to be taken seriously as a cognitive signal in its own right. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.