A reader highlights a passage. The system recognises what kind of content it is, generates an appropriate visual, and renders it inline — without the reader leaving the document.
Every interaction follows the same path — from the moment a reader selects text to the moment a contextual visual appears beneath it.
The reader highlights any descriptive passage of five words or more within the reading interface. This uses the standard text selection event already present in every web and native reading application — no new interaction model is required. The selected passage is captured and passed to the next step.
A trained language model analyses the selected passage and identifies which of the visual categories it belongs to — meteorological, anatomical, architectural, chemical, geographical, and so on. A cardiac cycle passage is recognised as anatomical. A storm system passage is recognised as meteorological.
The recognised category determines what kind of output is appropriate for that passage. The system selects between visual outputs and contextual text outputs depending on what the content can support. The reader never sees this decision — they see only the result.
The selected output type is generated from the content of the passage. A diagram, an animation, an illustration, or a contextual observation — whichever the recognition step has determined is most appropriate. The output is constructed specifically from what the passage describes.
The output is rendered inline below the highlighted passage, within the reading interface, with the reader's position preserved. The surrounding text remains in place. When the reader dismisses the output, the document reflows and they return to exactly where they were reading. No navigation occurred. No tab was opened. The reader never left the document.
The method supports a range of generation approaches, selected automatically based on the nature of the passage. The reader experiences a single coherent feature; the system handles the variation behind it.
Process-based content — cardiac cycles, molecular pathways, chemical reactions — is rendered as structured diagrams whose elements derive from the passage itself.
Scene-based content — landscapes, historical settings, architectural views — produces illustrative visuals constructed from the passage's descriptive language.
For passages that don't lend themselves to visualisation, the system produces a brief, useful observation that adds comprehension value. Never an error state.
"A method, not a model — independent of any specific AI provider used to implement it."
The method does not depend on any particular AI provider or model. An implementer may use whichever model best suits each step, and may change models over time as the technology evolves.
The method works across EPUB, PDF, and web-based reading interfaces. It is implementable at device level, application level, or platform level.
The augmentation layer sits within the reading application. The underlying content file is untouched. A publisher's existing catalogue is immediately compatible — no editorial production cycle, no re-authoring, no new content investment.
The method is designed so that the reader always receives a useful response. Different passages produce different kinds of output, but the experience never breaks down into an error state.
Nothing appears without the reader's active choice. The feature is completely invisible until wanted — unlike enhanced ebooks, which were always-on and disruptive.
The visual is rendered without taking the reader away from where they were. When dismissed, the document reflows and the reader returns to the exact word they were reading.