Machine Readership: The Reading Crisis Publishers Cannot See
The reading crisis is real.
But reference publishers face a stranger problem.
Their content may be read more than ever.
Just not by humans.
At the 2026 London Book Fair, Pan Macmillan CEO Joanna Prior warned that the decline of reading may be a greater threat to publishing than AI.
She was talking about human reading.
Children reading less for pleasure. Adults abandoning books. Students struggling with sustained attention. A culture increasingly shaped by scrolling, summaries and distraction.
For trade publishing, that warning is direct. A novel needs a human reader. A book that is bought but not read creates a weaker cultural and economic signal.
But for reference publishers, the story is different.
STM, legal, medical, scholarly and professional catalogues may not be read less.
They may be read more than ever.
The reader has changed.
In short
Machine readership refers to the use of publisher content by AI systems rather than by human readers.
For reference publishers, this can include AI systems drawing on medical catalogues, legal databases, scholarly archives, educational materials or professional reference works to ground answers, support workflows, summarize information or reason over trusted sources.
That changes the measurement problem.
Human reading creates familiar signals: clicks, pageviews, sessions, downloads, loans, citations, dwell time and renewals. Machine readership may create value without producing those signals.
A catalogue can support thousands of AI-mediated answers without creating a single conventional analytics event.
At Citations Logic, we use the term usage evidence to describe rights-aware records that make AI content use visible, attributable and commercially actionable.
The reading crisis asks whether people are reading less.
Machine readership asks a harder question for reference publishers:
Can you prove when your catalogue is being used, even when the reader is no longer human?
If the answer is no, the issue is not only cultural decline.
It is a measurement failure.
The reading crisis and AI are not separate stories
It is tempting to treat the decline of reading and the rise of AI as two different problems.
Humans are reading fewer books.
AI systems are consuming more content.
But for publishers, these are two sides of the same structural shift.
For most of publishing history, usage and human attention were closely linked.
A page read meant a person had read it.
A click meant a person had arrived.
A download meant a person had accessed the work.
Dwell time, circulation, loans, citations and renewals all assumed a human reader somewhere in the chain.
That assumption is now only partly true.
Human attention is weakening in many contexts. At the same time, machine consumption of content is increasing.
The result is a dangerous misreading.
A publisher whose only measurement system is human attention may see declining engagement and conclude that content value is falling.
But the value may not have disappeared.
It may have moved into a form of usage the publisher cannot see.
This is why AI usage evidence for publishers is becoming essential.
The crisis is not only that reading is declining.
It is that the reading publishers still measure may be the kind that is shrinking, while the reading that is growing may leave no trace.
Why reference publishers feel this first
Reference publishers are especially exposed to this shift.
Unlike trade books, reference content was never primarily about linear reading.
It was about consultation, reliance and use.
Nobody reads a drug monograph for pleasure.
Nobody reads a legal database from beginning to end.
Few people read a full scientific archive as a continuous experience.
They consult.
They search.
They extract.
They rely.
That is exactly how AI systems use reference content.
This is why reference publishers may be the first to experience machine readership at scale.
The catalogue is still valuable.
The content is still being used.
The reliance may even be increasing.
But the old signals do not capture it.
There may be no click.
No session.
No human dwell time.
No referral.
No visible reader.
For reference publishers, the problem is not that the catalogue has stopped mattering.
The problem is that the catalogue may matter more while appearing to matter less.
That is where AI content attribution for publishers becomes more than a visibility issue.
Attribution must connect machine use back to source identity, rights and value.
The old meter can point the wrong way
This is the most dangerous part.
Traditional engagement metrics do not merely become incomplete.
They can become misleading.
A reference publisher looking only at human-facing analytics may see lower visits, fewer sessions, weaker direct engagement or declining visible demand.
That can lead to the wrong conclusion:
The catalogue is losing relevance.
But inside AI systems, the opposite may be true.
The catalogue may be used to ground answers, support workflows, improve reliability, reduce uncertainty or strengthen the quality of generated responses.
It may be consumed more frequently than before, but through systems that do not generate the old evidence of reading.
That changes the meaning of decline.
A fall in human engagement does not automatically mean a fall in content value.
It may mean the publisher is measuring the wrong reader.
Human reading and machine readership are different events
Publishers need sharper language.
Human reading and machine readership are not the same kind of use.
Question | Human reading | Machine readership |
|---|---|---|
Who consumes the content? | A person | An AI system, agent, model, retrieval layer or workflow |
What signals appear? | Clicks, sessions, downloads, loans, dwell time | Retrieval, grounding, summarization, extraction, citation, output support |
What does the publisher usually see? | Analytics and engagement data | Often little or nothing unless usage is instrumented |
What is the risk? | Declining attention | Invisible use |
What evidence is needed? | Audience metrics | Rights-aware usage records |
This distinction matters because reference publishing does not sell only attention.
It sells trusted use.
A medical reference is valuable because a user or system can rely on it.
A legal commentary is valuable because it can support interpretation.
A scholarly archive is valuable because it can ground research.
If machines are now part of that use chain, publishers need records of machine readership.
Not just audience metrics.
Not just citation screenshots.
Not just platform summaries.
Records.
Content, usage, meaning: the link that breaks
The deeper issue is the chain that makes publishing economically sustainable.
Content creates value when it is used.
Usage creates value when it produces meaning.
Meaning creates value when it supports learning, research, decisions, trust or professional work.
The chain is simple:
content → usage → meaning
The reading crisis weakens the human side of that chain.
Less deep reading means less sustained engagement and fewer meaningful encounters with content.
AI changes the machine side of the chain.
Content may be used without a visible reader, without attribution and without a record the publisher can hold.
For reference publishers, the chain often breaks at one specific point.
Not between content and usage.
The usage is happening.
It breaks between usage and accountability.
A model may rely on a catalogue.
An answer may be improved by it.
A workflow may depend on it.
But if nothing records that use, the publisher cannot connect the content back to value.
This is why retrieved is not cited matters.
A source can support an AI answer without becoming visible as a source.
You cannot restore value to a use you cannot see.
Why machine readership needs its own record
Machine readership cannot be measured with tools built for human attention.
A pageview is not enough.
A click is not enough.
A citation is not always enough.
A platform dashboard may not be enough.
What publishers need is a record of machine use.
Which asset was used?
Under which right?
In which AI workflow?
At what moment?
With what attribution?
For what kind of output, answer or professional use?
This is not about monitoring human readers.
It is about making machine-mediated usage visible enough to govern, attribute and value.
The distinction matters.
A publisher does not need to know everything a person reads.
But when an AI system accesses a licensed catalogue, that interaction should not disappear into a black box.
A machine read should become a record.
That is also the foundation for AI licensing renewals.
If a catalogue is being used by AI systems, future licensing value depends on whether that usage can be shown.
That is why proof of AI content usage is not a legal luxury.
It is the commercial foundation of machine readership.
The role of provenance
Provenance matters, but it is not enough by itself.
A publisher needs to know where content comes from, who owns it, which version is authoritative, which rights apply and how it should be represented.
That is the foundation.
But machine readership adds the next question:
What happened to the content after it became available to an AI system?
Was it retrieved?
Was it cited?
Was it summarized?
Was it used for grounding?
Was it displayed?
Was it used silently to support an answer?
Provenance answers where content comes from.
Usage evidence answers where content goes.
That is why machine-readable provenance for reference publishers must connect to usage records.
Otherwise, publishers can prove origin without proving use.
The objection: machine readership is too hard to measure
There is a fair objection.
Machine readership is harder to measure than human reading.
A person opens a page. A system may retrieve a passage, compare sources, generate an answer and expose only part of the chain.
The event is less visible.
The infrastructure is more complex.
The counterparty often controls the logs.
That is true.
But difficulty is not a reason to ignore the shift.
It is the reason the category matters.
Publishers do not need a perfect map of every internal model operation to improve their position.
They need reliable usage records at the points they can govern: licensed access, retrieval events, API calls, rights checks, attribution states, content passages, answer support and audit trails.
The goal is not surveillance.
The goal is accountability.
Machine readership does not need to be mystified.
It needs to be instrumented.
Where Citations Logic fits
This is the gap Citations Logic is built to close.
Not the human reading crisis itself.
That is a cultural, educational and societal problem.
Citations Logic addresses the measurement crisis that appears when trusted content is consumed by AI systems without leaving a usable record.
The approach is simple.
Every machine use of a trusted catalogue should become an event.
Which asset was drawn on.
Under which right.
In which AI workflow.
With what attribution.
With what evidentiary value.
When the reader becomes a machine, the read needs to become a record.
That record reconnects the broken chain between content, usage and value.
It turns invisible machine readership into something a publisher can measure, attribute and price.
It changes the conversation from:
“Our engagement is falling.”
to:
“Here is how our catalogue is actually being used.”
That is the difference between losing sight of value and being able to govern it.
The question for publishers
Joanna Prior is right: the decline of human reading is an existential question for the book trade.
But for reference publishers, there is another question hidden inside it.
If your catalogue is being read more than ever by machines, can you see any of it?
If the answer is no, the reading crisis has already reached you.
It just arrived in a form your dashboards were never designed to register.
The old question was:
Can people find and read our content?
The new question is:
Can we prove when our content is being used, even when the reader is no longer human?
Because a catalogue does not lose its value when humans read it less.
It loses value when no one can show it is being read at all.
Frequently asked questions
What is machine readership?
Machine readership refers to the use of publisher content by AI systems rather than by human readers. For reference publishers, this can include AI systems drawing on catalogues to ground answers, support workflows, summarize information or reason over trusted sources.
How is machine readership different from human reading?
Human reading creates signals such as clicks, pageviews, sessions, downloads, loans or time spent. Machine readership may create value without producing those signals. The content is used, but traditional analytics may not record the interaction.
Why does this matter for reference publishers?
Reference publishers create content that is often consulted rather than read end-to-end. This makes their catalogues especially useful for AI systems. But if machine use is not recorded, the publisher may underestimate the value of its own catalogue.
Is the reading crisis caused by AI?
Not entirely. The decline in reading, especially reading for pleasure and sustained attention, predates the current AI wave. AI may accelerate the consequences by making summaries, answers and mediated access more common.
Can machine readership be measured?
Yes, but not with traditional human-engagement analytics. It requires capturing usage at the point of access: which asset was used, under which right, in which AI workflow, and with what attribution.
Continue the evidence chain
AI Usage Evidence for Publishers
AI Content Attribution for Publishers
Machine-Readable Provenance for Reference Publishers
Book an AI usage evidence assessment
Sources
The Bookseller — “Reading crisis bigger threat to book trade than AI, Joanna Prior warns”
https://www.thebookseller.com/news/reading-crisis-bigger-threat-to-book-trade-than-ai-joanna-prior-warns
Publishers Weekly — “LBF 2026: Pan Mac CEO Says Books Need to Be ‘as Urgent as Notifications’”
https://www.publishersweekly.com/pw/by-topic/international/london-book-fair/article/99904-lbf-2026-pan-mac-ceo-says-books-need-to-be-as-urgent-as-notifications.html
National Literacy Trust — “Children and young people’s reading in 2026”
https://literacytrust.org.uk/research-services/research-reports/children-and-young-peoples-reading-in-2026/
National Endowment for the Arts — “Federal Data on Reading for Pleasure: All Signs Show a Slump”
https://www.arts.gov/stories/blog/2024/federal-data-reading-pleasure-all-signs-show-slump
Times Higher Education — “One book in three weeks? At some universities that would be a triumph”
https://www.timeshighereducation.com/opinion/one-book-three-weeks-some-universities-would-be-triumph
University of Florida News — “Reading for pleasure in free fall: New study finds 40% drop over two decades”
https://news.ufl.edu/2025/08/reading-for-pleasure-study/