The format is not your problem. Across 46 studies, reading and listening produce comprehension that is statistically indistinguishable. What audio actually takes from you is narrower and stranger than “you remember less”: you lose the ability to stop and go back, so the conclusions a book implies but never states are the ones that fail to form.
Which means the fix is not another thing to do while you listen. It is one thing to do after:
- Set one question before you press play. Not “what is this chapter about” — something with an edge, like what is the author arguing against here? You are giving the next forty minutes something to attach to.
- At the end of a chapter, say the author’s conclusion out loud, from memory, without scrubbing back. Sixty seconds. The claim, not the anecdote.
- Match the company to the material. Walking and dishes are nearly free. Traffic, email and anything you have to think about are not.
- Come back after a day, then a week, with the same closed-book attempt. No app, no notebook — just say it again and notice what has gone.
Everything below is why this works, which part of the standard audiobook advice is measurably wrong, and where listening genuinely loses to reading.
Do you remember less from audiobooks than from reading?
On the whole, no — and the two best studies on it are unusually clean.
Virginia Clinton-Lisell’s meta-analysis pooled 46 studies covering 4,687 people and found that “the overall difference between reading and listening comprehension was not reliably different (g = 0.07, p = .23)” (Review of Educational Research, 92(4), 543–582, August 2022 — record). An effect size of 0.07 is, for practical purposes, nothing.
The single most direct experiment agrees. Rogowsky, Calhoun and Tallal randomly assigned 91 adults to consume the preface and a chapter of the same non-fiction book as a digital audiobook, as e-text, or as both at once, then gave everyone the same written comprehension test immediately and again two weeks later. Their result: “No statistically significant differences were found for any analyses pertaining to effects of the three different instructional conditions on comprehension at Time 1 or Time 2” (SAGE Open, 6(3), July 2016 — record).
Note what the second study also found: the dual-modality group — listening and reading along — did not beat either one alone. Doing both is not a shortcut.
So when a popular post on this exact subject tells you, without citing anything, that “it’s therefore easier to retain information when you read it as opposed to listen to it” (Chris Slade), that is a hunch, and the evidence does not support it as stated.
Then why does nothing stick?
Because “no overall difference” is an average, and the interesting part of Clinton-Lisell’s analysis is where the average breaks apart. Two moderators mattered:
| What changed | Reading vs. listening |
|---|---|
| Reader controlled their own pace | Reading better (g = 0.13, p = .049) |
| Pace controlled by the experimenter | No reliable difference (g = −0.32, p = .16) |
| Test measured literal comprehension | No difference at all (g = −0.01, p = .93) |
| Test measured inferential comprehension | Reading better (g = 0.36, p = .02) |
Read those last two rows together, because they are the whole story. Audio delivers the facts. Ask what the book said and listening does exactly as well as reading. Ask what follows from it — the thing the author set up over three paragraphs and never spelled out — and a gap opens.
That gap has an obvious mechanical cause, and it is the same one as the pacing row: you cannot go back. When a sentence in print does not resolve, your eye returns to the top of the paragraph in half a second, at no cost, often without you noticing you did it. In audio the equivalent move costs you a fumble for the 30-second button, a lost place, and usually the decision that it wasn’t important enough. So you keep moving, and the inference never gets built.
The second cost is the one everybody half-knows: you are almost never only listening. In one lab study, participants listened to short story segments either on their own or while doing a concurrent visual task. Mean comprehension accuracy was 88.96% listening alone, 84.58% under a light concurrent task, and 65.63% under a demanding one (Xie, Brodbeck & Chandrasekaran, Neurobiology of Language, 4(2), 318–343, 2023 — full text).
Take that as direction, not as a number to quote: it was 16 young adults, 60-second segments, and a lab task rather than an actual commute. But the shape is right, and it is more useful than the advice it corrects. The popular guidance says to multitask only on “autopilot” activities and then lists cleaning, laundry, driving and walking together as though they cost the same. They do not. Folding laundry is close to free. Merging onto a motorway is the demanding condition.
Why the standard audiobook advice doesn’t fix it
Pull the recommendations off the pages currently ranking for this question and you get a remarkably consistent list: take notes, use the app’s clip-and-bookmark feature, change the playback speed, pause every 20–30 minutes to summarise, build a mind map afterwards, re-listen to the good ones, and tell a friend about it.
Two problems.
First, the three most-recommended moves are the three weakest ones we know of. In their review of ten common learning techniques, Dunlosky and colleagues rated summarisation, highlighting and rereading low utility, while practice testing and distributed practice came out high (Psychological Science in the Public Interest, 14(1), 4–58, 2013 — abstract). Note-taking, summarising and re-listening are the audio versions of exactly those low-utility three. We took that argument apart in detail for print in how to remember what you read without taking notes; it applies unchanged here, with the extra handicap that you are on a bicycle and cannot write anything down anyway.
Second, and worse: not one of those pages asks you to recall anything closed-book. Every “reflect”, “summarise” and “mind map” step happens with the material still ringing in your ears or the app still open. That is recognition — the comfortable feeling of yes, I know this — and it is a famously bad guide to what you will still have next week. One of the most-linked guides on the query lists eight tips and never mentions testing yourself at all (AudiobooksGeek).
While we are here, one claim you will meet on half these pages deserves to die. A popular guide states that “studies show that we maintain about 90% of what we teach… whereas if you just read, the numbers are more like 10%” (EntreResource), and cites nothing, because there is nothing to cite. Those numbers come from the “learning pyramid”, and when Letrud and Hernes traced its history through digital libraries they found versions circulating for over 160 years and concluded flatly that “the models did not originate from empirical research” (Cogent Education, 5(1), 2018 — record). Explaining an idea to someone is still a good move. It is a good move because retrieval works, not because of a percentage somebody made up.
What to do while you listen
All four of these are hands-free by design, because the reader most likely to be listening is the one who cannot stop to write.
Set the question first. Thirty seconds before you press play, decide what you are listening for. “What is the author arguing against?” “Why doesn’t the obvious answer work?” A chapter heard with a live question in mind is processed differently from a chapter heard as ambient audio, and it costs nothing.
Say the conclusion at the chapter break, not the summary. This is the single highest-value sixty seconds available to you. Out loud, from memory, before scrubbing back: what was he actually claiming, and what was the argument? Notice that this attacks the exact deficit the research identified — you are forcing the inferential step that audio’s missing look-back suppressed. If nothing comes out, that is not failure, it is the most useful signal you will get; make a mental note of the chapter and move on. Its limit is worth knowing: explaining out loud catches where you stall, but an empty car cannot tell you when you are fluently wrong — the same problem the Feynman technique has when you use it alone.
Choose the company deliberately. Walking, washing up, folding, a familiar route on foot: cheap. Driving in traffic, anything with a screen, anything you would describe as “getting through”: expensive enough that you should either accept you are listening for pleasure, or save the chapter.
Drop the speed for the dense parts, don’t raise it. The standard advice is to speed up to stay engaged, and for narrative or familiar material it is fine. But the deficit audio imposes is on the inferences, and inferences need the seconds between sentences. If a chapter is doing real argumentative work, 0.9× buys you thinking time that 1.5× spends.
What to do after you finish
The chapter-break check is encoding. It is not retention, and on its own it will fade like everything else.
The move that actually keeps a book is boring: come back to it, closed-book, on a widening gap. A day later, then about a week, then longer, each time saying what the chapter argued before you look at anything. Each gap wants to be long enough that recall takes visible effort — the effort is the mechanism, not a side effect of it.
We are not going to publish another review calendar here, because the honest answer is that the right interval depends on how long you need to hold the material and how well the last attempt went; that argument, with the evidence, is in how to beat the forgetting curve as an adult learner. If you want a concrete chapter-by-chapter schedule you can run by hand, it is in spaced repetition for books, without making flashcards. And if you find you cannot produce anything at all a week later, the problem is upstream of retention — that diagnosis is here.
When you should read instead of listening
Being honest about the limits is more useful than defending the format:
- Anything with notation. Equations, code, proofs, tables, chemical structures. Audio cannot render them and narrators paper over them.
- Anything you will need to quote or find again. Retrieval practice makes you know a book; it does not let you locate page 214. If you are writing something of your own, read.
- Anything where the argument is doing heavy lifting per sentence. Philosophy is the clean example: the inferential penalty is at its worst exactly where every clause counts.
- First contact with a genuinely hard book. Listen to the second pass, not the first.
If you own a title in both formats, following the text while the narration plays does restore the look-back — and Audible now ships that natively: Immersion Reading, announced 18 February 2026, syncs highlighted text to the narration for customers who have both the audiobook and the Kindle eBook. It is a genuinely good feature and better at read-along than anything we do. Note two things: it needs you to own both editions, and Rogowsky’s dual-modality group did not outperform either single format — so treat it as a way to stop losing your place, not as a retention multiplier.
How Study Junkie closes the gap between listening and testing
Everything above is doable by hand. Two parts reliably rot: nobody keeps writing good questions about their own listening, and nobody keeps a schedule they invented on a walk.
Study Junkie is built to run those two parts for a book you chose. You upload the source — a PDF, DOCX or text file — and it becomes a course with chapters generated from the full text, preserved rather than summarised. Then:
- You listen to the chapters as AI narration, at 0.75× to 2×, which is where the slow-down-for-dense-material advice above stops being a nuisance and becomes a slider.
- Assignments come back with feedback on what you wrote, so the chapter-break check produces a verdict instead of a private feeling. This is the move that targets the inferential gap directly: you are asked to draw the conclusion, and something reads the answer.
- Scheduled recall runs the calendar, with questions generated from the chapter itself and a reminder when a session is due — so the day-later and week-later attempts happen without you administering them.
- The tutor is grounded in your uploaded source, not general internet knowledge, so “wait, what was he actually claiming in chapter 4” gets answered from the book.
The honest boundaries. Study Junkie narrates the document you upload; it is not an audiobook player and will not open your Audible library. Headway will quiz you on audio today, but over its own fifteen-minute summaries rather than the whole book you picked — a real trade, and if a summary is what you want, it is the better tool. A knowledgeable human who has read the same book beats all of this. And if you want to see what feedback on a written answer looks like before signing up, the Kant walkthrough shows one on a hard text.
You can start with 1,500 free Study Credits — no card required, no subscription, and credits that do not expire. If the four moves above are enough on their own, use those; they are free and they work. The how it works walkthrough shows where the listening and the testing meet.
Frequently asked questions
Do you remember less from audiobooks than from reading? Not overall. Across 46 studies (N = 4,687) the difference was not reliable (g = 0.07, p = .23), and a randomised study of 91 adults found no difference at two weeks. The exception is inferential comprehension — conclusions the text implies rather than states — where reading held a real advantage (g = 0.36, p = .02).
What playback speed should I listen at? Faster is fine for narrative and familiar material. Slow down — 0.9× or below — when the chapter is making an argument, because the inferential step is the one audio costs you, and it needs the gaps between sentences.
Is it bad to listen to audiobooks while driving? For learning, yes, in proportion to how demanding the driving is. In a lab dual-task study, comprehension fell from about 89% listening alone to about 66% under a demanding concurrent task. Keep the difficult chapters for a walk and let the car take the easy ones.
Should I read along with the text while listening? It helps you keep your place and it restores the ability to go back, which is a real advantage. But it is not a retention multiplier: the dual-modality group in Rogowsky’s study did not beat listening or reading alone.
How do I remember a non-fiction audiobook I finished months ago? Start with a closed-book attempt before re-listening to anything: say what the book argued and where it went. What you cannot produce tells you which two or three chapters to revisit — and this time, do the chapter-break check and come back to it on a widening gap.
Do audiobooks count as reading? For comprehension, the evidence says the question barely matters. What matters is whether anything ever asks you to produce the material again without it in front of you.