Every patient-facing document asks something of the mind. A Clinical Outcome Assessment (COA) demands recall of experiences and mapping onto scales. A consent form requires comprehension of legal and medical clauses. A patient information sheet lists side effects to be monitored. Instructions for Use (IFUs) guide device operation. Wearables and apps push alerts into daily life.

On the surface, these materials appear straightforward. But beneath the surface lies an unspoken assumption: that patients have the cognitive resources to parse, recall, evaluate and act on them. This assumption is often wrong.

Working memory is the mental workspace that supports reasoning and decision-making. It is limited and fragile. In other articles we have detailed how psychologist have debated the size of the working memory since the 1950s with George Miller suggesting a span of seven plus or minus two items, Nelson Cowan revising the limit to four and Alan Baddeley showing that it is not one ‘box’ where working memory lives but multiple. More recently, the rise of digital tools has reshaped how people use their memory, outsourcing not only facts but reasoning itself.

This article traces that arc, from Miller to the digital age, and shows how each stage of theory reveals why patient-facing documents often fail to serve their purpose. Along the way, we connect these insights into our own Nine Circles of Burden.

Miller’s Magic Number Seven

In 1956, George A. Miller published “The Magical Number Seven, Plus or Minus Two”. He noted that in different experiments, from absolute judgement of tones, number spans, word lists, human performance clustered around seven items. He called it a coincidence, not a law and introduced chunking: span depends on meaningful units.

Yet “seven plus or minus two” became received wisdom. It entered textbooks, management workshops and design guidelines as if it were a fixed truth about the human mind.

Implications for patient-facing documents

COAs often expect the patient to chunk well over 7 concepts. COA scales often have more than 7 anchors. Consent forms list risks in dense blocks that far exceed ‘seven, plus or minus two’. The steps of device use in IFUs often describe more than 7 steps. Wearables can deliver multiple alerts in rapid succession. Each assumes patients can reliably handle seven or more chunks. In practice, this sets people up to fail.

Cowan’s Four ± One

Nelson Cowan re-examined Miller’s evidence. His conclusion was that the real capacity of working memory is closer to four ± one chunks. Unlike Miller’s seven, this was not a coincidence across tasks but a robust ceiling across many domains.

The key insight is that four is a maximum under ideal conditions. Under stress, fatigue, illness, distraction or cognitive impairment, effective span shrinks further.

Implications for patient-facing documents:

• COAs: a binary pain item already requires around 4 concepts: timeframe, definition of pain, recall of episodes, mapping to Yes/No. Add frequency or multi-domain judgements and the limit is exceeded.

• Consent forms: Sections that list more than four risks without grouping risk overload.

• Information sheets: 20-40 side effects are cognitively unrealistic; patients just skim or discard.

• IFUs: Procedures longer than 3-4 steps must provide external aids (diagrams/checklists) visible at point-of-use.

• Wearables: more than 3 to 4 concurrent notifications fragment attention; users ignore them.

Cowan’s ceiling shows that many materials exceed human capacity by design.

Baddeley’s Multi-Component Model

Miller and Cowan treated working memory as a single pain, a single ‘box’ where it lives. Alan Baddeley and Graham Hitch (1974) complicated the picture. Their model divided working memory into multiple subsystems:

• Phonological loop: an inner ear/voice that stores and rehearses speech for around 2 seconds.

• Visuo-spatial sketchpad: a workspace for images and spatial layouts.

• Central executive: the attentional system that allocates resources and suppresses distractions.

• Episodic buffer: a limited-capacity integrator that binds information from different sources into coherent episodes.

This model explained dual-task findings: people can do a verbal and visual task concurrently but two verbal tasks clash. It also showed that overloaded is not just about quantity, but which subsystem is engaged.

Implications for patient-facing documents:

• Phonological loop: Long stems in COAs or polysyllabic jargon in consent forms exceed the 2 second rehearsal window; patients lose the sentence before its end.

• Visuo-spatial sketchpad: Grids, tables, diagrams and icons require precise visual tracking; misalignment errors are visual not conceptual.

• Central executive: People must switch between parsing, recalling, acting; stress/anxiety reduce control; consent demands inhibition of irrelevant detail; IFUs require reading-while-doing.

• Episodic buffer: integration across domains, such as QoL question or risk/benefits (consent) overloads the buffer; people default to the most salient dimension.

Baddeley shows multidimensional burden: a document can overload one subsystem, several, or all at once.

Decay, Resource and Interference

Why does working memory fail? Complementary mechanisms explain different failure modes:

• Decay: Information fades unless rehearsed. In consent, a long clause can decay before the verb is reached.

• Resource sharing: a limited attention pool. IFUs that require holding a step-in mind while acting drain the pool.

• Interference: Side-effect lists with overlapping terms create confusion.

Design corollaries:

Think about decay. You have keep sentences short. Keep critical information visible during action and make descriptors distinct.

The Google Effect

In 2011, Sparrow, Liu and Wegner described the Google Effect. When people know information will be available online, they are less likely to remember it. Instead, they remember where it is.

This is cultural shift: memory is increasingly about retrievability rather than internalisation.

Implications:

• Patients may not memorise side effects or steps, but they must be able to find them quickly.

• Consent: make key rights/risks findable (headings, summaries, links), not buried.

• Information sheets: clear headings/hierarchy so patients can re-orient in the moment.

• IFUs: layout/navigation that supports rapid re-finding during urgency.

• Wearables: retain accessible history not just transient alerts.

The Google Effect teaches that retrievability is as important as memorability, if not more important.

LLMs and Cognitive Debt

If Google shifted memory from content to location, ChatGPT shifts it from reasoning to fluency. Search required queries and integration; LLMs provide ready-made answers.

A 2025 MIT Media Lab study reported reduced brain activation for recall/executive control in ChatGPT-assisted users and coined cognitive debt: short-term efficiency at the cost of long-term reasoning practice.

Patients now live in this environment. They are accustomed to summaries, simplifications and external scaffolds. Long documents feel intolerable. Lists feel alien. Instructions are expected to be conversational or visual. Alerts are expected to interpret and not merely report.

Implications:

• Consent: AI summaries may be skimmed; unaided comprehension can be shallow.

• Information sheets: the TL;DR effect: shorter versions may be expected.

• IFUs: video/chatbots guidance preferred; static leaflets bypassed.

• Wearables: users expect interpreted guidance (“You may be dehydrated”), not raw numbers.

LLMs risk making documents performative (for regulatory compliance) rather than fictional (true comprehension), unless we design for active cognition.

Stress, Fatigue and Context

All theories converge: working memory shrinks under real-world conditions.

• Stress hormones disrupt prefrontal networks.

• Pain/fatigue drain attentional resources.

• Ageing and disease reduce capacity and control.

Patient-facing documents are almost always encountered in these conditions. Consent at diagnosis. Package Leaflets while unwell. IFUs during urgency. COAs while fatigued. Wearables during multitasking. The lab span of four becomes two or three. The fragility of memory is compounded by context.

Participant Cognitive Impairment (the often-missing reality)

Many patients live with cognitive impairment. This can be neurological, such as MCI, Parkinson’s, stroke, or psychiatric, such as depression, ADHD or treatment-related, e.g. “chemo brain”. This directly reduces working memory, attention, processing speed and executive control.

Implications:

• Consent forms: weighing risks and benefits with clause-heavy text becomes unrealistic without scaffolds and chunking.

• Package Leaflets: misclassification of symptoms is common when attention/memory are reduced.

• IFUs: sequencing errors increase; reading-while-doing is especially fragile.

• Wearables: prioritising and inhibiting alerts is harder; misses and false dismissals rise.

Ignoring impairment systematically excludes those who most need clarity.

Comparative Profiles

Looking across formats, each engages working memory differently:

• COAs = memory-intensive; recall + integration across domains quickly exceed capacity.

• Consent = comprehension overload via density, jargon, formatting; context and impairment amplify.

• Information sheets = fail through volume and interference; lists swamp span.

• IFUs = fail in action: resource-sharing between holding steps and performing them, especially under stress.

• Apps and wearables = fail by interruption: constant alerts fragment attention and exploit offloading habits.

Each material is a different test of working memory and thus a different configuration of circles.

What We Learn

1.   Designers assume mythical capacities. Working memory is much narrower than is thought and even narrower under stress or impairment.

2.   Different documents fail in different ways. COAs collapse under memory; consent under density; Package Leaflets under volume; IFUs under stress; wearables under distraction. Poor access to the documents (formatting, language) and digital offloading impose further strain.

3.   Digital offloading changes expectations. Patients expect retrieval and interpretation, not memorisation. They anticipate summaries and conversational guidance. Traditional documents feel alien unless redesigned for retrievability and actionability.

4.   Impairment is common, not exception. Materials must accommodate reduced attention, memory and control by design, not as an afterthought.

Conclusion

Working memory theory offers a powerful lens on patient-facing materials. Miller showed us the myth of seven. Cowan showed us the reality of four. Baddeley showed us the fragile subsystems. The Google Effect showed how memory is externalised. Reliance on LLMs showed how reasoning is outsourced.

Together, they reveal why patient-facing documents so often fail. They assume more memory than patients have. They ignore multi-component fragility. They overload lists, jargon and formatting. They collide with impairment and offloading. And they are delivered in contexts where working memory is at its weakest.

The tasks ahead is clear: design documents not for mythical minds but for real patients. These are patients who are tired, anxious, overloaded and sometimes cognitively impaired and, increasingly, accustomed to outsourcing memory and reasoning. That means shorter, plainer, chunked and retrievable. It means instructions visible at the moment action. It means alerts that interpret without fragmenting attention. It means scaffolds that support recall and reasoning and not to punish patients because of their limits dictated by biology and circumstance.

Working memory theory provides the foundation. Patient-facing documents are the test. Bridging the two shows us where we are failing and how to do better.

Matrix: Working Memory Theories Applied to Patient-Facing Materials

 What the Matrix Shows

• COAs stress recall/mapping; consent stresses comprehension density; info sheets stress volume/interference; IFUs stress action under load; apps/wearables stress attention and expectations.

• Impairment and offloading are now core design realities, not edge cases.

Designing for the Nine Circles means respecting capacity, removing avoidable burdens, scaffolding recall/integration, and aligning delivery with real contexts.

Thank you for reading,

Mark Gibson, Madrid, Spain

Nur Ferrante Morales, Ávila, Spain

September 2025