The learning pyramid is made up.

Somewhere today, a slide is telling a room of professionals what people supposedly remember. 10% of what they read; 20% of what they hear; 30% of what they see; 50% of what they see and hear; 70% of what they say; 90% of what they do. The percentages descend a colorful pyramid. A citation may gesture at “research” or “NTL Institute” or “Edgar Dale.” The room nods. The numbers feel roughly right, they flatter hands-on training, and they will be repeated in someone else’s deck within the month.

The slide has been in circulation, in recognizable form, since before the people presenting it were born. It has survived the move from overhead projector to PowerPoint to Canva. It has acquired and shed attributions along the way, and been translated into every language corporate training speaks. By staying power and reach, it may be the most successful piece of content the learning profession has ever produced.

Here is everything wrong with it, stated plainly: no study produced those numbers. Not an old study, not a flawed study, not a small study — no study. The percentages are a fabrication with a traceable non-history. The scholars who did the tracing have published the void in detail. This article tells that story and explains why the fabrication is so durable. And — because the pyramid’s direction gestures at something real — it lays out what the honest version of its claim looks like.

What Edgar Dale actually made

The pyramid’s supposed ancestor is real and blameless. In 1946, the audiovisual-education pioneer Edgar Dale published the “Cone of Experience,” a map of teaching media running from the concrete (direct purposeful experience, dramatization) to the abstract (visual symbols, verbal symbols). The cone contained no percentages, no retention claims, and no ranking of effectiveness. Dale presented it as a way of thinking about how concrete each medium is.

He explicitly warned readers not to treat the bands as a ladder of value. Abstract and concrete experiences, he noted, weave together in all real learning (Dale, 1946). The field’s historian describes how the cone was misread almost at once — reproduced, simplified, and moralized into the very ranking its author disclaimed (Molenda, 2004).

Dale’s actual contribution deserves rescue from its afterlife. The cone was a genuinely useful idea. It mapped how mediated an experience is, built for an era deciding when film and radio belonged in classrooms. Its bands described abstraction, not effectiveness. Reading is more abstract than a field trip — a fact about representation, not a ranking of value. And Dale revised the work across three editions without ever adding a number to it (Dale, 1946), (Molenda, 2004).

The percentages arrived later, by grafting. At some point in the 1960s, numbers began appearing attached to cone-like diagrams in training materials. They were commonly credited to the National Training Laboratories, or to a 1967 Mobil Oil publication by Treichler that asserted the figures without any study, method, or source (Subramony, Molenda, Betrus & Thalheimer, 2014).

When researchers later asked NTL for the underlying research, the institute reportedly could not produce it. The numbers were “believed to be accurate,” but the origin was lost. That sentence deserves a moment. The most-cited statistics in corporate training are sourced to an organization’s belief about a study nobody can find.

The citation archaeology

What lifts this from anecdote to scholarship is that people did the tedious work. They pulled every version of the chart they could find, followed every citation to its end, and published the map of where the trails go. The debunking literature reads like detective fiction with footnotes. Systematic attempts to trace the pyramid’s supporting references find a closed loop. Publication A cites B, B cites C, and C cites A — or a source that contains no data. The chains circle without ever touching a measurement (Letrud & Hernes, 2018).

The pattern has a name in the myth-studies literature: citogenesis, facts manufactured by citation itself. The pyramid is its textbook case. Each repetition adds apparent authority. The diagram’s visual confidence stands in for evidence. And after enough cycles, the burden of proof silently flips — doubting the numbers starts to feel like the extraordinary claim.

The direct refutations add the obvious final blow: the numbers are implausible on their face. Real retention varies hugely with material, delay, learner knowledge, and — as this entire library documents — the design of the encounter. Could “reading” as a category really yield 10% retention regardless of what is read, by whom, over what interval? That idea contradicts a century of memory research, whose central finding is that conditions, not media, drive retention (Letrud, 2012). And the suspiciously round steps — 10, 20, 30, 50, 70, 90 — are the fingerprints of invention, not measurement. Real data is never that tidy, a point the myth-busting handbooks make with some relish (De Bruyckere, Kirschner & Hulshof, 2015).

Reading the numbers as numbers

It is worth pausing on what the percentages would even mean if they were real, because the muddle is instructive. Retention of what, measured when? A retention percentage without a delay interval is meaningless. Our forgetting-curve review shows recall of the same material ranging from near-perfect to near-zero depending on when you ask.

Retention for whom? The knowledge-effects literature shows the same text yielding wildly different retention across readers with different prior knowledge. Retention of what grain — verbatim wording, gist, application? The three come apart in every serious study. The pyramid answers none of these questions, because it was never an answer to a question. It is a rhetorical object shaped like data.

The tidy arithmetic gives the game away twice more. Real learning studies produce effect sizes with decimals, confidence intervals, and moderators. They do not produce six categories that climb in round steps and top out at a satisfying 90. And the categories themselves shift across versions of the chart — “teach others” appears and disappears, “demonstration” swaps tiers, audiovisual splits and merges. That is how folklore mutates: each teller adjusts the story. Measurement works the other way — the numbers would constrain the telling (Subramony et al., 2014).

Why this fabrication is so hard to kill

The pyramid survives for the same reasons 70:20:10 survives, plus one of its own. Like the ratio, it flatters its audience: trainers who deliver hands-on learning hear that their format is nine times better than reading. It also supplies numerical precision where practitioners crave it. Its extra advantage is directional truth. Active, generative engagement genuinely does beat passive reception in study after honest study — so the pyramid keeps looking roughly vindicated by real research it never descended from. A myth that points roughly the right way is far harder to dislodge than one that points wrong, because correcting it feels like attacking its conclusion rather than its evidence.

The transmission network matters too. The pyramid lives heavily in train-the-trainer materials, teaching certificates, and instructional-design coursework — the very channels that credential learning professionals. Prevalence studies of educational misconceptions show how well such channels spread them. Majorities of educators endorse well-refuted claims, at rates barely dented by general education or even neuroscience exposure (Dekker, Lee, Howard-Jones & Jolles, 2012).

People do not adopt the pyramid because they evaluated it. They adopt it because the person who certified them presented it — in a diagram, with percentages, in a course about how people learn. Fluency is doing the work evidence should do. The myth literature notes that irony without ever quite exhausting it.

What the honest version says

Debunking has a hollow feel when it ends at the rubble. So the constructive half matters more here than the demolition. The pyramid owed its career to answering a genuine question its audience still needs answered: which learning experiences stick? Strip the fake numbers, and a real, defensible claim remains. It just has to be built from actual literatures, each with its own effect sizes and boundary conditions.

Retrieval beats re-reading — with backing from meta-analyses, the studies that pool many experiments into one estimate, and with known moderators (Dunlosky, Rawson, Marsh, Nathan & Willingham, 2013). Our testing-effect review covers that terrain. Generation beats reception: producing an answer, an explanation, or a solution attempt outperforms consuming the same content, per the generation, self-explanation, and productive-failure literatures this library reviews one by one. Practice-with-feedback beats exposure, per the deliberate-practice and simulation evidence. And the engagement taxonomy that organizes all of it — interactive beats constructive beats active beats passive — carries systematic-review support precisely because it ranks cognitive activities, not media (Chi & Wylie, 2014).

Notice what the honest version refuses to do: assign one number per format. “Doing” ranges from brilliant (deliberate practice with feedback) to worthless (activity theater with neither). “Reading” ranges from passive skimming to self-explaining study of a well-built text — and the second can outteach a badly run workshop. The medium never carried the retention. The cognitive work does. That is why the pyramid’s ranking sometimes matches reality, and why its logic never will.

A field guide to the genre

The pyramid is the type specimen of a genre every learning buyer should recognize on sight, because the genre is still producing. It has four markers. A universal number attached to something deeply context-dependent — retention, attention span, the share of communication that is “body language.” A diagram whose visual authority stands in for a method section. An origin story that names an institution rather than a study. And citations that resolve to other citations.

The eight-second attention span, the 93%-nonverbal-communication figure, and the pyramid share every marker. Each has its own debunking literature, and each outsells its debunkers every year (De Bruyckere et al., 2015).

The genre’s defense mechanism is the same in every case: when challenged, retreat to the moral. “Well, the exact numbers don’t matter — the point is that doing beats reading.” But the exact numbers were the entire sales pitch. Without them, the claim collapses into the real literatures, which say something more useful and less quotable. A profession that wants to be evidence-based has to refuse the retreat: either the number has a study, or it leaves the slide. The moral can then stand trial on the honest evidence — which, for active learning, it comfortably survives.

The cost of teaching with fabrications

Does it matter, if the pyramid points roughly the right way? The question is the myth’s last line of defense, and it deserves a serious answer rather than a shrug. “Harmless simplification” is exactly what every fabrication calls itself. It matters three times over.

First, credibility. A profession that opens its decks with invented statistics forfeits standing when it later argues from real ones. The executive who discovers the 90% was made up learns to discount the next number L&D shows them — including the true ones. Second, design error. The pyramid ranks media, so it licenses swapping an excellent reading-plus-retrieval design for a mediocre “hands-on” activity — a downgrade the real literatures would have flagged.

Third, opportunity cost. Every slide spent on fake percentages is a slide not spent teaching the actual retention levers — spacing, testing, generation, feedback. Their honest effect sizes are large enough to need no embellishment (Dunlosky et al., 2013).

What the evidence doesn’t show

• It doesn’t show passive formats are fine. The debunk targets the fake numbers, not the real superiority of generative engagement — which honest literatures support with task-dependent effect sizes (Chi & Wylie, 2014).
• It doesn’t implicate Edgar Dale. The cone was a media taxonomy with explicit anti-ranking caveats; its author is the myth’s first victim, not its source (Dale, 1946), (Molenda, 2004).
• Absence of the pyramid’s evidence is not absence of retention evidence. Retention is among the most-studied phenomena in psychology; the scandal is that the pyramid ignored that literature, not that no literature exists.
• Debunking alone doesn’t change behavior. The neuromyth prevalence studies show corrections fade without replacement; the durable fix is installing the true model — conditions over media — where the pyramid used to sit (Dekker et al., 2012).

Where the evidence stops

1. 1It doesn’t show passive formats are fine
2. 2It doesn’t implicate Edgar Dale
3. 3Absence of the pyramid’s evidence is not absence of retention evidence
4. 4Debunking alone doesn’t change behavior

© 2026 FUTURE PROOF™

What this means for practice

For an individual practitioner, the fix takes an afternoon. The notes below assume you are also responsible for what your organization believes. Delete the slide — and replace it rather than merely retracting it, because the vacuum is what invites the myth back and the replacement is where the actual persuasion happens. The replacement fits in the same space: retention follows the cognitive work, not the medium. Show three honest exhibits — a testing-effect result, a spacing curve, a generation comparison — each with a real citation that survives being checked. Audiences respect “the truth is messier and here it is” more than trainers fear they will.

Then install the institutional habit the pyramid’s career warns you to have: provenance-check the statistics in your own materials, starting with the ones that feel most obviously true. The pyramid, 70:20:10, and the learning-styles matching claim share a signature — round numbers, missing methods, citations that cite each other. A one-hour audit of your core decks against that signature is among the cheapest credibility investments an L&D function can make. A field whose subject is learning should be the hardest place for a fabrication to survive forty years. That it was among the easiest is the uncomfortable finding. Fixing it locally is a decision, not a research program.