Two posts contain the same false claim.
One person shares it because they believe it.
Another knows it is false and releases it through coordinated accounts to make an event look different from what happened.
The visible content may be identical.
The informational act is not.
That difference is why disinformation cannot be classified reliably from falsity alone.
Definition
For this Knowledge Library, disinformation means:
false or materially misleading information deliberately created, manipulated or distributed with an intention to deceive or strategically mislead.
This formulation reflects a common core across influential frameworks.
It should not be mistaken for a universally settled scientific definition.
Different disciplines and institutions disagree about whether harm must be intended, whether strategic manipulation without literal falsity counts, how to classify propaganda, and how much evidence of intention is required.
The Institute therefore treats “disinformation” as a useful but contested category.
The intent problem
Falsity can sometimes be established by comparing a claim with evidence.
Intent is harder.
A misleading artifact rarely carries a reliable label saying why it was produced.
Researchers may infer intent from internal documents, repeated behaviour, coordinated campaigns, known organizational objectives, fabricated identities or patterns that are difficult to explain as accidental.
But no single clue is decisive in every case.
Hayward’s work on the epistemology of disinformation emphasizes precisely this difficulty: intentionality is often central to the concept, yet the evidence available to observers may be incomplete and the concept is used differently across epistemic, behavioural and security framings.
That means careful writing should distinguish:
false content, suspected strategic manipulation, and established deceptive intent.
What disinformation is not
Disinformation is not every false statement.
People make mistakes. Journalists publish corrections. Scientists revise conclusions. AI systems hallucinate. Eyewitnesses misremember.
Without evidence of deliberate deception, labeling these acts disinformation can turn an epistemic category into a moral accusation.
Disinformation is also not identical to propaganda. Propaganda can include true statements, selective framing, emotional appeals and ideological persuasion without relying on factual falsehood.
Nor is it identical to malinformation, a policy term often used for genuine information shared in a harmful or strategically manipulative way.
These boundaries are imperfect, but they matter.
Context makes deception legible
Hameleers’ 2022 work argues for a context-sensitive understanding of disinformation as a deliberate act involving actors, intentions and techniques.
This is useful because a content-only approach misses the communicative system around the artifact.
A misleading claim may become easier to classify when researchers can see coordinated timing, impersonation, repeated targeting, covert sponsorship, fabricated evidence or deliberate alteration.
The unit of analysis can therefore shift from a single post to a network, campaign or organizational process.
This is one reason computational detection of “disinformation content” has limits. The decisive evidence may sit outside the text.
Strategic deception can be partly true
A campaign need not invent everything.
Selective disclosure, decontextualized documents, authentic images with false captions and mixtures of accurate and fabricated details can be more persuasive than complete fiction.
A narrow truth-versus-falsehood classifier may therefore miss strategic misleadingness.
This creates an important conceptual tension.
If disinformation is defined as necessarily false, partly true strategic manipulation sits at the edge of the category. If it is defined more broadly as deliberate misleading information, the construct becomes harder to measure consistently.
Researchers should state which definition they use rather than silently moving between them.
Why attribution matters
To respond proportionately, institutions often need to know more than whether content is wrong.
They may need to know who produced it, whether accounts are coordinated, whether sponsorship is hidden, whether identities are authentic, whether targeting is systematic and whether the campaign crosses platforms or jurisdictions.
These are attribution problems.
Attribution can be technically and politically difficult. False certainty about who is responsible can itself become misinformation.
The epistemic discipline is therefore symmetrical: claims about manipulation require evidence too.
Disinformation research is not one unified measurement tradition
Studies may measure belief in false claims, exposure, sharing, detection accuracy, campaign reach, network coordination, source attribution, trust, behavioural response or intervention effects.
These outcomes answer different questions.
A person may recognize a false message but still share it for identity or entertainment reasons.
A campaign may achieve high reach without changing beliefs.
A correction may improve factual accuracy without changing trust in the actor who distributed the claim.
Using one outcome as a proxy for the entire phenomenon can obscure mechanism.
Media literacy and prebunking address only part of the problem
Individual discernment matters.
So do platform design, provenance systems, journalism, institutional communication, threat analysis and accountability.
It would be unfair to place the burden of organized strategic deception entirely on citizens.
Media-literacy interventions can improve some detection and evaluation outcomes. Inoculation can improve discrimination under studied conditions.
But strategic campaigns adapt.
A society’s response therefore has to operate at both human and institutional levels.
Generative AI changes production economics
Generative systems can lower the cost of producing text, audio, images and video at scale.
They can support translation, personalization and rapid variation.
That capability can be used for legitimate communication or for deception.
The presence of generated content does not establish disinformation.
Nor does an authentic human-created artifact guarantee legitimacy.
The relevant questions remain about truth, context, intention, provenance, coordination and consequence.
AI changes the production frontier. It does not remove the need for classification discipline.
European frameworks emphasize resilience and competence
European digital and media-literacy frameworks increasingly address disinformation as part of broader competence and societal resilience.
DigComp 3.0 integrates critical information evaluation and AI-related competence. The European Commission’s media-literacy work explicitly links media competence to resilience against misleading and harmful content.
These are policy responses, not proof that one educational intervention neutralizes organized manipulation.
Their relevance is institutional: disinformation is treated as a problem requiring both public competence and system-level safeguards.
The danger of turning “disinformation” into a weapon word
A concept built to describe strategic deception can itself be used strategically.
Public actors sometimes label unwelcome reporting, dissent or uncertain claims as disinformation without adequate evidence.
That usage damages the category because it substitutes authority for demonstration.
The remedy is not to abandon the term.
It is to make classification evidence explicit.
What is false or misleading? What evidence supports that judgment? What supports the intent inference? What actor or process is being attributed? What uncertainty remains?
A serious research institution should make those questions visible.
Disinformation is an ecosystem problem
A deceptive campaign can exploit real vulnerabilities without creating them.
Low institutional trust, social polarization, information gaps and weak local journalism can make strategic falsehood easier to circulate. Platform affordances can accelerate distribution. Economic incentives can reward sensational content.
This does not mean those conditions cause every campaign.
It means resilience cannot be built only at the message level.
Institutions need correction capacity, transparent communication, provenance systems and trusted channels that exist before a crisis.
Coordination can be evidentially informative
A single account posting a false claim may tell us little about intent.
Patterns across accounts can tell us more.
Synchronized posting, repeated wording, coordinated amplification, fabricated personas and shared infrastructure can provide evidence that communication is organized rather than spontaneous.
Computational methods can help identify these patterns.
But pattern detection is not self-interpreting. Legitimate campaigns, emergency alerts, news syndication and activist coordination can also produce synchronized behaviour.
The inference from coordination to deception therefore requires contextual evidence.
Attribution has confidence levels
Security analysis often works under uncertainty.
Rather than declaring attribution as binary, analysts can distinguish confidence levels and identify the evidence supporting them.
That discipline should carry into public communication.
A statement such as “this content is consistent with a coordinated campaign” is different from “this actor deliberately produced this falsehood.”
The second claim requires stronger evidence. A responsible assessment should also record what evidence would reduce or reverse its confidence, so attribution remains revisable rather than becoming institutional dogma.
Maintaining that difference protects the credibility of institutions responding to manipulation.
Counter-disinformation measures can have side effects
Interventions are not epistemically neutral simply because their aim is protective.
Content labels can be ignored or misunderstood. Aggressive moderation can generate perceptions of censorship. Public warnings can inadvertently amplify a claim. Poorly justified accusations can damage trust in genuine threat analysis.
This does not imply that institutions should do nothing.
It means countermeasures need evaluation, transparency and proportionality.
The cure should not reproduce the same evidential shortcuts being criticized.
The role of journalism
Professional journalism can reduce the space for disinformation by providing verified reporting, explaining uncertainty and correcting mistakes visibly.
Journalism can also become a target of strategic manipulation when false material is designed to exploit newsroom incentives for novelty and speed.
Verification practices therefore matter upstream.
A newsroom that can trace images, inspect metadata, contact original sources and resist pressure to publish before verification acts as part of society’s epistemic infrastructure.
Distinguishing persuasion from deception
Not all strategic communication is disinformation.
Advertising, advocacy and political communication are persuasive by design. They may use selective framing. Ethical concerns can arise even when statements are true.
The boundary becomes disinformation when the operational definition’s deception criteria are met.
Maintaining this distinction is essential for neutral analysis. Otherwise the term can become a synonym for “persuasion I dislike.”
A research claim should state its unit of analysis
Is the study analyzing individual messages, accounts, networks, campaigns, audience beliefs or institutional responses?
These units produce different evidence.
A content classifier can estimate properties of messages. It cannot by itself establish the strategic intent of a network. A network study can identify coordination. It cannot automatically tell us whether every shared claim is false.
Clear units of analysis reduce category errors and make cross-study synthesis more defensible.
Synthetic identities and credibility laundering
Strategic deception can target not only facts but the apparent identity of the speaker.
Fabricated personas, impersonated organizations and networks of accounts can make a message look socially independent when it is centrally coordinated. AI-generated profile images and text can lower the cost of maintaining such identities.
The epistemic problem is credibility laundering: borrowed signals of ordinary participation are used to disguise the actual source relationship.
Detection requires more than inspecting whether one post is factually false. Analysts may need account history, registration patterns, cross-platform behaviour and infrastructure evidence.
Again, caution is essential. An anonymous or pseudonymous account is not evidence of deception by itself. Privacy, safety and political conditions can give legitimate reasons for anonymity.
Prebunking techniques versus actors
One promising response to disinformation is to teach recurring manipulation techniques such as false dichotomies, impersonation or emotionally manipulative framing.
Technique-based learning has an advantage: it can sometimes generalize beyond one claim or actor.
But techniques are not ownership labels. The presence of an emotionally charged appeal does not identify a campaign as disinformation. Legitimate communication uses emotion too.
The correct inference is modest: a technique may be a cue for closer inspection.
This distinction helps keep prebunking educational rather than accusatory.
Documentation is part of institutional resilience
When organizations investigate suspected disinformation, evidence can disappear quickly.
Posts are deleted. accounts change names. webpages update. metadata is lost.
A defensible process preserves timestamps, source URLs, screenshots or archives, technical indicators, analytic assumptions and confidence levels. It distinguishes raw evidence from analyst interpretation.
This is not only a security practice. It is an epistemic practice.
Future reviewers need to be able to reconstruct why a classification was made.
Without documentation, even a correct attribution can become an authority claim that outsiders cannot evaluate.
Cross-border and multilingual limits
Disinformation research often concentrates on languages and platforms for which researchers have data and tools.
Automated classifiers trained in one language may perform poorly in another. Cultural knowledge affects sarcasm, coded language and references. Platform access differs by jurisdiction.
A campaign that looks minor in an English-language dataset may be important elsewhere.
This creates a recurring generalisability problem.
Northern European institutions should not treat strong English-language tooling as evidence that smaller-language information environments are equally observable. Local journalism, linguistic expertise and regional research infrastructure remain essential.
Correction and accountability after attribution
Even strong attribution can later change.
New evidence may identify a different actor, reveal that accounts were compromised, or show that a supposedly coordinated pattern had a benign explanation.
Institutions therefore need visible correction mechanisms for their own disinformation assessments. A public label should not become irreversible merely because it came from a trusted authority.
This symmetry is central to epistemic legitimacy: organizations that ask citizens to revise beliefs in response to evidence should demonstrate the same capacity themselves.
What the evidence does not show
The evidence does not show that every false claim is disinformation.
It does not show that deceptive intent can be read directly from content.
It does not show that every coordinated campaign changes beliefs or behaviour.
It does not show that AI-generated content is inherently deceptive.
It does not show that individual media literacy alone can neutralize organized strategic manipulation.
And disagreement over a claim is not evidence that one side is conducting disinformation.
What this changes for Inner Technology
Disinformation adds an agency layer to discernment.
The Institute therefore separates three questions that are often collapsed:
What is the epistemic status of the claim? What is the provenance and distribution pattern? What evidence exists about the actor’s intention?
Only after those are separated should they be recombined.
This is Institute synthesis, not a validated attribution protocol.
Its function is to reduce a common cognitive error: moving too quickly from “this appears wrong” to “someone is deliberately deceiving us.”
The two identical posts
The content can be copied perfectly.
Intent cannot.
One sender may be mistaken. Another may be strategic. A third may be a bot following instructions from a system nobody can yet attribute confidently.
Discernment has to tolerate that uncertainty.
The category becomes useful only when it makes our claims more precise than the manipulation we are trying to understand.
Research references
- Wardle C, Derakhshan H. Information Disorder: Toward an interdisciplinary framework for research and policy making. Council of Europe. 2017. https://rm.coe.int/information-disorder-toward-an-interdisciplinary-framework-for-researc/168076277c
- Hameleers M. Disinformation as a Context-Bound Phenomenon: Toward a Conceptual Clarification Integrating Actors, Intentions and Techniques of Creation and Dissemination. Communication Theory. 2023;33(1):1–10. Published online 2022. https://doi.org/10.1093/ct/qtac021
- Hayward M. What We Know About What They Know: An Epistemology of Disinformation. Social Epistemology. 2025. https://doi.org/10.1080/02691728.2024.2346127
- Cosgrove J, Cachia R. DigComp 3.0. European Commission Joint Research Centre. 2025. https://doi.org/10.2760/0001149
- European Commission. Media literacy. Current through 2026. https://digital-strategy.ec.europa.eu/en/policies/media-literacy