Schema Theory: What It Is, Bartlett's Study and Examples

Go from storyboard to an interactive course in minutes

Start Free

Picture a service technician with fifteen years on petrol and diesel cars, doing a first annual service on the dealership's new electric hatchback. They have watched the "what's new about EVs" video and can recite the new safety rules. Then they lean into a car that makes no sound, assume it is switched off and start work. The car is in Ready mode, and it can move. Nothing in the video was wrong. Fifteen years of knowledge about what a quiet car means simply got there first.

I'm an instructional designer at Coursebox, where we make AI tools for employee training that turn a brief into a staff course, so I'll mention our product where I tested it. Schema theory explains what happened to that technician. In this guide I set out the theory in Frederic Bartlett's own words, his War of the Ghosts study and the experiments that followed it, how schemas change and what prior knowledge does and does not predict. Then I plan training for experienced staff whose old habits will mislead them, with three real cases, a product test and two free templates.

What is schema theory?

Schema theory, first applied to memory by Frederic Bartlett in 1932, holds that people understand and remember new information through schemas: organised knowledge from past experience that shapes what they notice, how they fill the gaps and what they later recall.

View the full-size diagram

Bartlett, a Cambridge psychologist, set the idea out in his 1932 book Remembering: A Study in Experimental and Social Psychology. His definition is short: "'Schema' refers to an active organisation of past reactions, or of past experiences, which must always be supposed to be operating in any well-adapted organic response" (p. 201). The important word is active. A schema is not a filing cabinet that new facts drop into. It does work: it decides what a person expects, what they treat as important and what they supply from habit when the detail runs out.

The cognitive scientist David Rumelhart gave the modern version in a 1980 chapter, Schemata: The building blocks of cognition: "A schema theory is basically a theory about knowledge. It is a theory about how knowledge is represented and about how that representation facilitates the use of the knowledge in particular ways." In his account a schema is a general pattern with slots. For a mechanic, a "routine service" schema has slots for oil, brakes and tyres. A particular job fills the slots, and when a slot is empty the schema supplies a default.

Three points set the scope:

• It covers memory, comprehension and expertise. The classic studies are about remembering stories, understanding texts and how experts see their field. Education took it up from the late 1970s, mostly through research on reading.
• It is a theory of how existing knowledge works, not a teaching method. Any design steps built on it, including the workflow in this guide, are an interpretation.
• It changes how you see experienced learners. Adults at work bring rich schemas for their job. Most of the time that helps them learn. Sometimes it is exactly what gets in the way.

Here are the main ideas at a glance, with the question each one asks of training.

Medicine is one field that builds schemas on purpose in training. Clinicians call them illness scripts, and this study shows what that looks like.

Case study: Family medicine residents at four US hospitals

This was an uncontrolled before-and-after study. Family medicine residents at four community hospitals in California, Oregon, Washington and Ohio spend only short rotations on children's wards, so they may never see some childhood illnesses in a given season. An illness script is the clinical form of a schema: a pattern for each disease that a doctor compares a new patient against.

• What they did: Four 1-hour sessions, one per presenting complaint, such as respiratory distress or a limp with fever. Each session laid four to eight look-alike diagnoses side by side in the same template (who gets it, history, examination, tests, management), with case vignettes and a compare-and-contrast worksheet, and a 10-minute test before and after.
• Evidence and result: In Madduri and colleagues' 2024 paper in MedEdPORTAL, paired tests from 42 to 44 residents per session showed scores rising from 72% to 92% for respiratory distress, 82% to 96% for abdominal pain, 68% to 81% for acute febrile limp and 58% to 73% for the well-appearing febrile infant. Residents rated the sessions 4.8 out of 5 for helping them organise knowledge, though only about half answered the survey.
• Limitation: There was no control group and no delayed test, the test was multiple choice and taken straight after the session, and the authors designed the curriculum. It shows short-term knowledge gains, not better diagnosis with patients.
• Lesson: Putting look-alike problems side by side in one fixed template gives learners a ready-made schema to compare new cases against. Before claiming it beats ordinary teaching, ask for a delayed test and a comparison group.

Where the word "schema" came from

Schema theory did not start with Jean Piaget. The word has a longer history, and each user meant something slightly different by it.

• Immanuel Kant, 1781. In the Critique of Pure Reason, Kant used "schema" for the rule that links a concept to what we perceive. His example is a dog: the concept "indicates a rule, according to which my imagination can delineate the figure of a four-footed animal in general", in J. M. D. Meiklejohn's translation.
• Henry Head and Gordon Holmes, 1911. The two neurologists used "schema" for the body's running model of its own posture, in a long paper in the journal Brain on patients with brain lesions.
• Frederic Bartlett, 1932. Bartlett took the word from Head for memory. His preface says, "My debt to Sir Henry Head will be obvious to everyone who reads this book."
• Jean Piaget. Piaget used his own term, schème, usually translated as "scheme", for the action patterns a developing child builds and revises. It grew up separately from Bartlett's work.
• Rumelhart, Ortony and Anderson, 1977 to 1980. David Rumelhart and Andrew Ortony set out the features of schemata in 1977, Richard C. Anderson and colleagues applied them to reading in 1977 and 1978, and Rumelhart's 1980 chapter gave the version most education writing still uses.

Bartlett was not fond of the word he borrowed. "I strongly dislike the term 'schema'. It is at once too definite and too sketchy," he wrote (pp. 200 to 201), and he said "organised setting" was closer to what he meant. He kept "schema" anyway, and so did everyone after him.

Piaget's schemas and Bartlett's schemas

Piaget's idea is about how children's thinking develops, and Bartlett's is about how adults remember. In a 1964 lecture published in the Journal of Research in Science Teaching, Piaget defined assimilation as "the integration of any sort of reality into a structure" and wrote that "Learning is possible only when there is active assimilation." Accommodation is the other half: the structure changes when experience will not fit it. Our guide to constructivism quotes Piaget on both.

The two ideas meet in a useful place. Assimilation describes the technician who fits a silent electric car into the petrol schema "quiet means off". Accommodation is what the training has to produce. But Piaget's stages describe children, so for adult staff the useful variable is what they already know about the task, not their developmental stage.

Bartlett's War of the Ghosts study

Bartlett's best-known experiment used a story chosen because it would not fit his readers' schemas. He tested it in two ways, and several details of the study are easy to get wrong.

The story. "The War of the Ghosts" is a folk tale from the Pacific Northwest, told in the Kathlamet (Chinookan) language by Charles Cultee in 1891 and translated by the anthropologist Franz Boas in 1901. Bartlett used his own adaptation of Boas's translation (p. 64). He picked it because it "belonged to a level of culture and a social environment exceedingly different from those of my subjects" (p. 64).

The participants. The readers were Bartlett's own subjects in Cambridge, not the people the story came from. "Twenty subjects were given The War of the Ghosts in this experiment, seven being women and the rest men" (p. 79). The serial reproduction chains used "English undergraduate students at the University of Cambridge" (p. 138).

The method. In repeated reproduction, each person read the story twice at their normal pace, wrote it down about 15 minutes later and again "at intervals of increasing length" (p. 63), up to months or years. Bartlett let intervals vary from person to person and told subjects only that it was a test of accurate recall (pp. 65 to 66). In serial reproduction, one person's version was passed to the next person to retell, like a chain.

What changed. The stories did not simply fade. They changed towards what the readers knew. One subject's first version, written 20 hours later, was "considerably shortened, mainly by omissions", and "The more familiar 'boat' once replaces 'canoe'; hunting seals becomes merely 'fishing' ... The main point about the ghosts is entirely misunderstood" (pp. 66 to 67). Across the experiments Bartlett describes omission, rationalisation of the strange parts, transformation of details into familiar ones, changes of order and a drift towards conventional forms. In the chains, "names and numbers rarely survive intact for more than a few reproductions" (p. 175).

Two findings from the study bear directly on training:

• The first version sticks. "The most general characteristic of the whole of this group of experiments was the persistence, for any given subject, of the 'form' of his first reproduction" (p. 83). Whatever a learner builds first tends to shape what they remember later, which is an argument for getting the first explanation right.
• Remembering is rebuilding. Bartlett's summary: "Remembering is not the re-excitation of innumerable fixed, lifeless and fragmentary traces. It is an imaginative reconstruction, or construction", and "It is thus hardly ever really exact" (p. 213). A learner who has "got the gist" may have rebuilt the detail from an old schema.

The study's methods were informal by later standards, and by 1999 nobody had managed to repeat the repeated-reproduction results. In 1999, Bergman and Henry Roediger ran it again with fixed intervals of 15 minutes, a week and six months, and reported: "We did replicate Bartlett's results, in that (1) subjects forgot the story over delays but (2) introduced rationalization and distortion into their accounts of the story".

Schema theory examples: three experiments that made it usable

Bartlett showed that schemas reshape memory. Three later experiments showed when and how, with numbers a designer can use.

Bransford and Johnson: the frame has to come first

In 1972, John Bransford and Marcia Johnson read students a passage that opens, "The procedure is actually quite simple. First you arrange things into different groups." It goes on in the same vague way. Without the topic it is hard to follow. With the topic, washing clothes, every sentence makes sense.

In their second experiment, 52 students at the State University of New York at Stony Brook heard the passage with no topic, with the topic afterwards or with the topic beforehand, then rated how well they understood it and wrote down what they could recall.

The published paper puts the finding in one line: the same information given after the passage "produced much lower comprehension ratings and recall scores." Comprehension ratings followed the same pattern, 4.50 out of 7 with the topic first against 2.12 with it afterwards. Telling people afterwards did almost nothing, because the schema has to be active while they take the information in. It is a short lab passage with students, but the design rule is plain: give the frame before the detail, not as a summary at the end.

Anderson and Pichert: the schema you read with decides what you notice

Richard C. Anderson and J. W. Pichert had college students read a story about two boys skipping school at one boy's house, either as a burglar or as someone thinking of buying the house. A leaky roof is important to a buyer; a coin collection, to a burglar. After one recall, some students were asked to switch perspective and recall the story again.

In the first experiment, with 39 students, those who switched "recalled an additional 7.1% of the now important information". The figure comes from the authors' 1977 report; the study was published as a 1978 journal article. The new perspective worked as a retrieval plan: it brought back details the students had read but not reported. For training, the lesson is practical. The schema a learner brings to a page decides what they pick up from it, so tell them what role to read it in, and give them a reason to look at it again from a second angle.

Chase and Simon: expertise is stored patterns that only fit some situations

In a 1973 study of chess, William Chase and Herbert Simon showed three players, one master, one strong club player and one beginner, a position from a real game for 5 seconds. "M was able to place an average of about 16 pieces correctly on the first trial, while A and B placed about eight and four, respectively" (p. 59). With pieces placed at random, "there was no relation at all between memory of the position and playing strength" (p. 60).

That is one person at each level, so treat it as a demonstration. A later study by Fernand Gobet and Simon found that strong players keep a small advantage on random boards, "although the relative difference ... is much smaller". The point survives: an expert's edge is a store of familiar patterns, and it shrinks when the situation stops looking like the ones they know. An experienced technician on a new kind of car is partly on a random board.

Bartlett, Rumelhart and the rest: which to read

• Bartlett, Remembering (1932; Cambridge reissue 1995). Read this one first, at least the theory chapter (pp. 199 to 214) and the War of the Ghosts chapters. It is readable, and it is the best check on claims made in Bartlett's name.
• Rumelhart, "Schemata: The building blocks of cognition" (1980). Read it for the cognitive science version, with slots, defaults and how schemas are tested against what we see.
• Simonsmeier and colleagues, "Domain-specific prior knowledge and learning: A meta-analysis" (2022). Read it before you make any claim about how much prior knowledge predicts learning.

How schemas change: accretion, tuning and restructuring

Schemas are not fixed in childhood. Rumelhart and Donald Norman described three ways they change in a 1978 chapter, Accretion, tuning, and restructuring: Three modes of learning:

• Accretion adds facts to a schema that already fits. A technician learns a new car's tyre pressures; nothing about how they think of a service has to change.
• Tuning adjusts an existing schema so it fits better, such as changing the default for when brakes need attention.
• Restructuring builds a new schema when the old one cannot be stretched to fit. New relations or contrasts may be needed. Seeing where an old explanation fails is one option, not a required stage of every schema change.

There is no agreed list of "three types of schema theory"; the nearest thing is these three modes of learning.

For a course designer, the useful question is which mode each part of the content needs. Most new product information is accretion, and a short page will do. A changed rule for something people do every day is tuning at least. A different idea of how the system works is restructuring, and a contrasting case can expose where the old schema stops fitting. This is a design option, not a required mechanism for every new schema.

How schemas affect learning: what prior knowledge predicts

David Ausubel put the classroom case for schema theory in one line, as the epigraph to his 1968 textbook: "The most important single factor influencing learning is what the learner already knows. Ascertain this and teach him accordingly" (p. vi). It is a principle, not a finding, and it does not show that prior knowledge is the biggest predictor of learning.

The best current evidence is more interesting than that. Bianca Simonsmeier and colleagues pooled 8,776 effect sizes in a 2022 meta-analysis in Educational Psychologist.

People who know more before a course tend to know more after it. But the average association with normalized gains was close to zero, and in some studies it helped while in others it hurt. The authors say the spread "falsifies general statements such as 'knowledge is power' as well as 'the effect of prior knowledge is negligible.'" Most of the studies are from schools and universities, so treat the figures as a guide, not a rule for every workplace.

The design consequence is specific. Sorting staff by years of experience tells you little about how a course will land. What you need to know is which pieces of their knowledge the new task can build on, and which old habit will interfere. To find out, ask experienced staff open questions about what they do now; and use task-focused questions or observed practice to check the relevant knowledge. The response format alone does not decide what an assessment measures.

When old schemas get in the way

Schemas usually help people learn. They get in the way when the new situation looks like the old one but works differently. Educational researchers call the result a misconception, and the process of replacing it conceptual change.

The best-known account is from George Posner and colleagues, in a 1982 paper in Science Education. They proposed four conditions for people to give up an existing idea:

• they must be dissatisfied with the idea they hold;
• the new idea must make sense to them;
• it must seem plausible from the start;
• it must prove useful, opening new ways to solve problems.

In the EV example below, the briefing explains a plausible new rule but does not test the old habit. A video could include that contrast; the medium alone does not tell you which conditions it addresses. Christine Tippett's 2010 review of two decades of research on refutation texts, which state a misconception and then explain why it is wrong, found that reading them "rather than traditional expository text is more likely to result in conceptual change". A study with teachers in Spain shows both what that achieves and how quickly it can wear off.

Case study: Teachers' myths about learning in Spanish schools

These were two experiments in which each teacher acted as their own control. Practising teachers from kindergarten to vocational schools in Spain rated how far they believed statements about learning, including myths such as "We only use 10% of our brain".

• What they did: 45 days after the first ratings, each teacher read six short refutation texts of about 150 to 181 words. Each stated a myth, refuted it and gave the correct idea. Three target myths got no text, as the comparison. A second experiment added a survey of whether teachers intended to use myth-based practices, and asked again 30 days later.
• Evidence and result: In Ferrero and colleagues' 2020 paper in the Journal of Experimental Psychology: Applied, belief in the refuted myths fell clearly in the first experiment (45 teachers from 12 schools), and strongly held myths moved the most. In the second (35 teachers from 3 schools), the effect "vanished after a 30-day period". Intention to use myth-based practices did not improve, and in fact "a small but significant backfire effect was found".
• Limitation: The samples were small, the experiments were not preregistered, there was no ordinary-text comparison, and the study measured stated belief and intention, not what teachers did in class.
• Lesson: Naming and refuting a wrong belief moves it quickly, but one reading does not hold, and people can say the right thing while intending the old practice. Come back to the corrected idea later, and check what people plan to do, not only what they now say is true.

A 2025 trial with nurses, doctors and final-year medical students in the UK and Finland found a similar gap: people who read worked examples of handling a vaccine-hesitant patient picked up the easy, visible step, an empathetic opening line, but were no more likely to correct the misconception itself. Reading examples was not enough; changing a schema, your own or someone else's, probably needs practice with feedback as well.

A worked example: experienced technicians moving to electric vehicles

Here is the hypothetical example I'll plan with from here on. A franchised car dealership's service workshop is taking on the brand's first two battery-electric models, called here the E1 hatchback and the E2 small SUV. The learners are six service technicians with 5 to 20 years on petrol and diesel cars. All six have done the manufacturer's separate high-voltage safety awareness course, which this training does not replace. Two also hold the brand's high-voltage technician qualification and four do not. Years on petrol cars do not show that anyone knows electric ones. The service manager signs off each technician's first EV service.

The problem is the old approach. The technicians watched a 20-minute "what's new about EVs" video and read the service checklist. On their first EV jobs, three old habits showed up: one technician assumed a silent car was switched off, one passed the brakes on a low-kilometre car on pad thickness alone, and one went looking for a starter motor fault on a car that would not "start". Each of them could recite the new rules when asked. All three would have passed a quiz on EV facts.

The invented workshop bulletin supplies these training facts. They are not a manufacturer's EV safety or isolation procedure. Real servicing must follow the approved procedure, task-specific training and authorisation; switching off or moving the key does not establish high-voltage isolation.

1. "Quiet does not mean off." An EV in Ready mode is silent and can move. Before any work, confirm the Ready light is off and the car is in Park, put the key fob in the key box at least 5 metres away, and hang the "HV: do not switch on" tag on the steering wheel.
2. Service bulletin SB-07. Regenerative braking means the friction brakes are used less, so discs corrode and caliper slide pins seize even with plenty of pad left. At every annual service, inspect both faces of each disc for corrosion and check the slide pins move freely, whatever the pad thickness or kilometres.
3. Two electrical systems. The car has a 12-volt battery and a separate high-voltage system with orange cables. A flat 12-volt battery can stop the car going into Ready mode, so it looks like a car that will not start. Only high-voltage qualified technicians disconnect the high-voltage system or touch orange-cabled parts. Under this hypothetical workshop's approved procedure, technicians trained for that task can test and charge the 12-volt battery.

Much of the job carries over unchanged: tyres and rotation, wheel alignment, suspension and steering checks, wipers and washers, the cabin air filter, lights, the brake fluid test and the customer report.

So that decisions can be judged, the course uses the workshop's own documents: the E1 annual service checklist, bulletin SB-07, the high-voltage rules card and four job cards.

• Job A: E1 annual service, 14 months old, 9,000 km, no complaints.
• Job B: E2 at 12,000 km; the customer reports "a grinding noise when braking slowly in the car park".
• Job C: E1 towed in, "won't start, dashboard dark".
• Job D, kept back as the final task: E2 annual service at 6,000 km; the customer's note says "the car crept forward in the car park when I thought it was off".

The intended outcome is that, by the end of the course and their first supervised EV service, a technician can carry out a routine annual service and brake inspection on an E1 or E2 under the workshop's rules, can say which petrol habit they are overriding at each point, and knows which jobs need a high-voltage qualified colleague. It does not give anyone high-voltage qualification or authority to work on high-voltage parts.

The example suits schema theory because the learners are experts. Their schema for a routine service is rich, fast and mostly right, which is exactly why three parts of it mislead on an EV. Most of the course is accretion. The brake check is tuning: the same inspection with a new default. "Quiet means off" and the single electrical system need restructuring, because the technician's picture of how the car works no longer holds.

How to use schema theory to design training

Bartlett and Rumelhart wrote no design method, so this workflow is my own. It draws on Bransford and Johnson, Landman's study of airline pilots, the conceptual change research and the cases, and it is carried through the EV workshop course.

1. Find out what learners do now. Before any new content, ask open questions about current practice and keep the answers. For the technicians: how do you know a car is safe to work on, how do you decide brakes pass at an annual service, and what do you check first when a car won't start? Open explanations can reveal reasoning. A recall quiz on isolated EV facts cannot show the old habit in action; a well-designed scenario or observed task can assess it.
2. Give the frame before the detail. Tell learners what the course covers and what carries over before they meet anything new, as the topic-first group in Bransford and Johnson's study were. In the course, a short "Before you start" page sets the frame and a "What carries over" page links the E1 and E2 service to the petrol service they already do. Ausubel's advance organisers are the classroom version of this move, and our guide to elaboration theory compares them with other ways of sequencing a course.
3. Let the old habit meet a case where it breaks, then give the rule. Show the situation, ask what the learner would normally do, show what happens on an EV if they do it, then give the bulletin's rule. This is Posner's first condition: dissatisfaction before the new idea. A scenario-based page can set this up; a list of differences alone may not.
4. Put old and new side by side. One comparison table, as in the illness-script case: task, petrol habit, E1 and E2 rule, who can do it.
5. Practise on varied cases, not one demo. Jobs A, B and C each test the rules in a different situation. Varying the cases builds a schema for the principle rather than a script for one job, which is the idea behind task classes in the 4C/ID model.
6. Test on an unseen job. Job D is held back until the end, answered in open text and read by the service manager.
7. Come back to it. Revisit the step 1 answers at a week-4 check-in and at the first supervised service. The teachers' case shows why: a corrected belief can drift back within a month.

Step 5 has the clearest evidence behind it, from a simulator study with airline pilots.

Case study: Airline pilots training for surprise failures in the Netherlands

This was a two-group experiment in a research simulator at Delft University of Technology, with pilots flying a light twin-engine aircraft model most had not flown. Standard checks rely on predictable scenarios, and the researchers argued that skills trained that way can be "'brittle' instead of adaptive".

• What they did: Twenty serving airline pilots, about 17 years in the job on average, practised the same 14 runs with engine and rudder failures. One group was told only that a malfunction would occur, with its timing, side and the weather varied. The other was told the failure and the response in advance, and every run was a repeat. Then everyone flew a surprise test at a different airport with a combined, partly new failure.
• Evidence and result: In Landman and colleagues' 2018 paper in Human Factors, the run ended in a successful landing for 9 of 10 pilots in the varied group and 2 of 10 in the repeated group, and four of the repeated group lost control in flight. The varied group was slower to call out failures during practice, which the authors read as time spent making sense of events. They link the result to "frame or schema construction".
• Limitation: Twenty pilots, assignment that was mostly but not fully random, a very brief practice session, an unfamiliar aircraft model and a test in the same session, so it says nothing about long-term retention.
• Lesson: Practise one announced case and people learn the script. Vary the cases and hold back the details, and practice is slower and harder, and the varied group handled this surprise simulator test better. The study does not establish performance in a real flight or isolate variation from the other training changes.

The schema plan for the EV workshop course

Here is the plan for the three habits that break, using the bulletin's rules and the four jobs.

Job D is the transfer task. It mixes rule 1 and rule 2 in a situation none of the practice jobs showed, and the service manager reads each technician's plan before their first supervised EV service.

What Coursebox course requests say about prior knowledge

Schema theory says the first thing to know is what learners already know. To see how often a course brief says it, I looked at course requests typed into Coursebox in the 90 days to 7 October 2026: one per person, from English-language browsers, with no uploaded file. Of 266 prompts, 104 were course requests. Two independent AI readers each classified all 104 against written definitions, and every positive call was then re-read in full. Counts are classification estimates from the union of the two readers. Positive-call checks do not establish an upper bound because missed cases remain possible.

The classification identified 7 requests asking the course to deal with mistakes or wrong beliefs learners bring, only 1 of them for staff training, and none asked for a refresher of something learners had done before. Those 7 were mostly hobby, side-business and self-study courses, such as beginner makeup mistakes or the belief that starting a business needs a large amount of money. The one staff course was workplace harassment training that described harassment being "dismissed as 'just a joke'". Only one request in the whole sample was for people moving from an old way of working to a new one, a course on the move from peer to supervisor.

The sample is small, it covers what people typed rather than finished courses, and organisations may know their learners well and simply not say so in a prompt. Still, a clear brief should state the prior knowledge to use or challenge, and the designer should verify it with learners. If you want a course to start from what your staff already do and deal with the habit that will mislead them, you have to write both into the brief yourself.

Schema theory vs cognitive load theory and related ideas

Schema theory is the parent of several ideas that do narrower jobs. The table is my synthesis, not a ranking.

The closest neighbour is cognitive load theory. It uses schemas as the unit of long-term memory: in his 1988 paper, John Sweller described "domain specific knowledge in the form of schemas" as "the primary factor distinguishing experts from novices in problem-solving skill", and the theory is about how to build those schemas without overloading working memory. Schema theory asks a different question: what does the knowledge learners already have do to the new material? Use cognitive load theory to size the task, and schema theory to decide what to activate, what to confront and in what order.

Ausubel's advance organisers turn "give the frame first" into a lesson device, and reviews of the research, such as Richard Mayer's in 1979, say they help "if used in appropriate situations and measured properly". Merrill's principles of instruction include activation as one phase, and Gagné's nine events include stimulating recall of prior learning. Both treat prior knowledge as something to call up; schema theory adds that some of it has to be challenged. For where schema theory sits among the big traditions, see our guide to behaviourism, cognitivism and constructivism. For the technicians, I would use schema theory to plan the habit cases and cognitive load theory to decide how much of the bulletin each page carries.

Criticisms of schema theory

Schema theory is widely used and loosely defined, and its critics have pointed at both.

• It is vague, by its originator's own account. Bartlett said the word was "already widely used ... to refer generally to any rather vaguely outlined theory" (p. 201). "Schema" can mean a story structure, a routine, a concept or a skill, so the theory can explain almost any result after the fact and predicts only roughly.
• Memory is not only schematic. Joseph Alba and Lynn Hasher's 1983 review, Is memory schematic?, tested a typical schema theory against how accurate, incomplete and distorted memory actually is, and argued that people often keep more specific detail than the theory allows.
• The classic studies are small or short. Bransford and Johnson and Anderson and Pichert used short passages with students and immediate recall. Chase and Simon tested one player at each level. Bartlett's own procedures were informal.
• The supporting evidence is short-term. Prior knowledge predicts post-training knowledge; its association with normalized gains varies widely, and corrected beliefs can drift back within a month.
• The brain research is early. Studies such as Tse and colleagues' 2007 work with rats suggest new information that fits an existing schema can be consolidated quickly, and a 2012 review by Marlieke van Kesteren and colleagues found that information congruent with a schema is "usually better remembered", while novel information is "also sometimes better remembered". None of it yet tells a designer what to do differently, and claims that schemas live in one part of the brain go beyond the evidence.

The workplace evidence is thinnest where designers need it most. The three cases here show short-term changes in knowledge, belief and simulator performance. I found no verified workplace case that measured on-the-job performance after a schema-based course with a delayed follow-up.

Using schema theory with an AI course builder

On 9 October 2026 I briefed Coursebox to build the course for the EV workshop. Coursebox is our product, so weigh my account with that in mind. The brief described the six technicians, the three rules word for word, what carries over and the four job cards. I asked for about 45 minutes, no multiple choice and nine sections in this order:

1. A short "Before you start".
2. "What do you do now?", with three open-text questions about current petrol practice.
3. "What carries over".
4. "Where your old habits break", one page per rule, each asking for the learner's usual practice before showing the consequence and the rule.
5. Petrol and EV service side by side.
6. Jobs A to C as open-text scenarios.
7. Job D on its own, with no hints.
8. Questions for the service manager.
9. A wrap-up as the very last section.

The builder stopped three times before producing an outline: once to report that it could not apply sections such as "Petrol and EV service side by side", "Three jobs" and "Job D", then with questions about page sizing and wrapper sections. I picked the recommended answer each time, including "Exact requested outline: Use 9 sections and 13 pages".

The first outline got the opening right. "What Do You Do Now?" came first, before any EV content, and each of the three old habits had its own page. The rest drifted from the brief. It had 4 sections and 9 pages, about 29 minutes, not the 9 sections it had confirmed. The brake rule page was a 3-question quiz, so it tested the rule instead of asking what the technician would normally do. The three jobs were a 3-question quiz rather than open-text scenarios. And Job D and the service manager questions were merged into one closing page with the wrap-up. A quiz on the rule is the kind of check experienced staff can pass while still holding the old habit, which is the problem the course exists to fix.

One follow-up message fixed the structure. I asked for the nine sections in order, rule pages that ask what the learner would normally do in open text before the consequence and the rule, Jobs A to C as open-text scenarios, Job D as its own section and no quizzes. After reloading, I checked the outline: 9 sections and 13 pages, no quizzes and 8 open-text assignments, with each rule on its own open-text page, Job D as a separate transfer plan and the wrap-up last.

I then read the generated "Rule 2: Brake wear can mislead" page, and it needed one more fix. It asked for "your understanding of the specific risks associated with regenerative braking", the rule itself before it was taught, described the case as "excess pad wear" when the case has plenty of pad left, and referred to brake information in section 2 that does not exist. I rewrote the task in the page editor: the situation (an E2 at its annual service with plenty of pad, low kilometres and no complaints), then "Before you read anything about electric cars, answer in your own words: on a petrol or diesel car in this state, what would you normally check on the brakes, and what would make you pass them?", with the consequence and the SB-07 rule word for word in the model answer. I reloaded the page and both fields had saved.

I checked the outline and the Rule 2 page, not every page; the Rule 1 and Rule 3 pages still need the same read. The course came to about 1 hour 33 minutes against the 45 I asked for, so it needs trimming. It does not replace the manufacturer's high-voltage course or the supervised first service, and an open-text box is no proof that anyone reads the answers: that is the service manager's job. This outline and page check did not observe a technician's habits or establish competence for high-voltage work. With the fixes in place, the outline asks for current practice before any EV content, and the brake page now asks what the technician would do on a petrol car before it shows where that habit breaks on an E2.

Putting schema theory to work

Schema theory answers a question training plans often leave out: what will learners do with this new information, given what they already know? Bartlett's answer, which a 1999 replication supported, is that they will fit it to their existing schemas, fill the gaps from habit and remember the version they built first. For new content that is mostly a help, and a frame given before the detail lets them use it. For experienced staff facing a change, the same process means a clear explanation of the new rule can leave the old habit in charge.

So use the theory to decide what to find out before you teach and which habits need a case where they break. Do not treat it as a promise that activation alone will do the work.

Your next step: take one course you are building for experienced staff and write down the one habit most likely to mislead them, with the real case where it would go wrong. Build that case into the course before the rule.