What counts as a hallucination?

Incorrectly assigning statements, ideas, or actions to sources or entities.

Incorrectly assigning cause-effect relationships.

Selectively presenting facts or data that support a particular position while ignoring contradictory evidence.

An argument in which the conclusion is assumed in one of the premises; also known as begging the question.

Content designed with misleading or sensationalist headlines to attract attention and generate clicks, often at the expense of accuracy or substance.

Content that violates general, widely accepted knowledge or basic logical reasoning expected of humans.

Expressing greater certainty than the available evidence supports, including presenting unverified information without appropriate uncertainty markers.

Explanatory beliefs about social events as the result of plots by powerful, often hidden, actors, typically lacking verifiable evidence.

Presenting information in a context that alters its meaning or significance, potentially leading to misinterpretation.

Mistakes in linking pronouns or other referring expressions to the correct entities or concepts within the text.

Errors in reasoning that involve incorrectly applying general rules or principles to specific instances.

Incorrectly defining terms, concepts, or phenomena.

Deliberately false or misleading information created and spread with the intention to deceive.

Misidentifying or misclassifying named entities such as persons, organizations, or locations.

Introducing information that is demonstrably false based on external knowledge beyond the source.

Generated output that introduces information not stated or inferable from the source alone; verification may require external evidence.

A use of hallucination centered on whether generated information is correct in the world.

A use of hallucination centered on whether generated output is supported by and accurately represents the provided source or input.

Deliberately fabricated information presented in the style of news reporting to deceive or mislead. A specific form of Disinformation.

Incorrectly linking unrelated ideas, events, or entities as if they are connected or causally related.

Model-generated content judged erroneous under a stated factuality or faithfulness policy, including unsupported or contradicted claims and related reasoning or reference failures.

Content created by impersonating genuine sources, individuals, or organizations to deceive.

Errors in resolving references that depend on the context of utterance, such as 'here', 'now', or 'I'.

Errors in reasoning that involve drawing general conclusions from specific instances improperly.

Content that deviates from or contradicts the specific instructions or guidelines provided in the input prompt.

Output that contradicts the source content or the input provided; this is verifiable by comparing the output directly to the source.

A deductive argument structured as a syllogism that is logically invalid, meaning its conclusion does not necessarily follow from its premises.

A flaw in the structure of a deductive argument or an error in reasoning that renders an argument invalid.

Content that contains internal contradictions or reasoning flaws that violate principles of logic.

Genuine information or imagery that has been altered or doctored to deceive.

Disseminable inaccurate information, which can be either unintentional or intentional.

Information that, while possibly technically correct in parts, is presented in a way that leads to incorrect conclusions or a false impression.

A conclusion or statement that does not logically follow from the previous argument or statement.

Referring to entities, events, or concepts that do not exist or are not present in the given context.

A subtle change in emphasis, scope, or implication that preserves the core proposition while shifting its interpretation. Use this for interpretive drift, not direct contradiction, dropped attribution, or an incorrect reference.

Making something deliberately unclear, confusing, or difficult to understand, often to hide the truth or mislead.

Content that was once accurate but is no longer current.

Making claims that are too narrow or constrained, applying general knowledge in an overly restrictive way or misclassifying broad principles as limited cases.

Placing items, concepts, or cases into overly broad categories or making sweeping claims based on insufficient evidence.

Incorrect representation of data points, statistics, numerical values, or mathematical calculations.

Drawing incorrect conclusions from the information in the source or applying flawed logical steps.

Incorrect citation, attribution of information to inappropriate sources, or errors in how entities or concepts are referred to.

The generated output is off-topic or does not adequately address the user's prompt or question.

Circulating unverified claims or stories, or making inferences without clear evidence or justification.

Unverified claims or stories circulating without clear evidence. Can be part of Misinformation or Disinformation.

Content that uses humor, irony, exaggeration, or ridicule to expose and criticize people's stupidity or vices, which can be misinterpreted as genuine information if context is lost.

Incorrectly defining the breadth of a claim or category, either by overextending generalizations to inappropriate cases or by applying general knowledge too narrowly.

Content that contradicts itself within the same generated output.

Presenting opinions from the source or generated opinions as factual statements without appropriate attribution or qualification.

Incorrect representation of time relationships, sequences, or chronology of events.

Text with severe linguistic issues such as incoherence, unintelligibility, major grammar errors, or major spelling errors that make the text hard to interpret.

Inaccurate information spread without a deliberate intent to deceive, possibly due to misunderstanding or error.

Information that cannot be verified by reliable external sources, making its accuracy indeterminable.

Information that can be verified by reliable external sources, but is not present in the source material.