The two words get used interchangeably, and the substitution is expensive. Information is a possession — a noun, something a person or a database holds, countable and transferable. Context is a relation. It exists between an utterance and whoever is receiving it, and it does not survive being extracted from that relationship.
The distinction sounds academic until it shows up in a product decision. Build for information and the obvious answer is a single source of truth: put everything in one place, grant everyone access, and the problem is solved by definition. Firms have been building that answer for thirty years. It reliably produces a repository nobody reads.
Why the same words land differently
Charles Fillmore’s frame semantics gives the reason. Word meanings are not understood in isolation; comprehension requires access to a frame — a structured body of background knowledge about a recurring situation. His canonical example is commerce: buy, sell, pay, charge and cost all evoke one scene with the same roles, each verb foregrounding a different participant and backgrounding the rest.
The consequence is the part that matters. Understanding depends on the listener’s stored knowledge, and misunderstanding arises precisely when two people activate different frames for the same word. Frame availability is a function of what each person has previously encountered — which means it is a function of training and experience, not of access.
Charles Goodwin demonstrated this about as concretely as it can be demonstrated. In Professional Vision, he compared archaeologists and lawyers examining the same physical scene and showed that trained coding schemes lead professionals to see genuinely different objects of knowledge inside identical material — color variation in soil read as a post hole by one observer and as nothing at all by another.
Put a lawyer and a finance professional in the same meeting and they do not receive the same meeting. One hears an indemnity structure; the other hears a working-capital assumption. Neither reading is wrong, and neither is complete. The transcript is identical. The context is not.
What this means for a shared record
If context is frame-dependent, then a flat, canonical record of what was said is not shared context — it is raw material that each reader still has to convert, at their own cost, using their own frame. Which is exactly what a document repository asks people to do, and exactly why they don’t.
There is a tempting analogy from physics here, and it is worth following until it breaks. Two observers on a moving train see a dropped ball the same way; an observer on the platform sees something different. Same event, different frames, and no privileged one — a useful corrective, because it removes hierarchy from the question of whose reading is correct.
Where the analogy fails is instructive. Physical frames of reference relate to one another through a fixed, computable, lossless transform — one continuous parameter, fully reversible, nothing discarded. Professional frames have nothing of the kind. There is no formula that converts a lawyer’s reading of a conversation into a CFO’s. They are closer to translation between languages with no exact equivalents, and they are a discrete set of different kinds rather than points along a single axis.
Which rules out the easy architecture. One canonical summary cannot be mechanically reprojected per reader the way a coordinate can. Each frame needs its own reading of the source material.
The practical form of the distinction
Information can be stored and retrieved. Context has to be produced, for a particular person, at a particular moment, against a particular question. That is a different kind of engineering problem, and it comes with a different test of success: not whether everything was captured, but whether the right slice reached the person who needed it, in a form their frame could actually use.
A test that separates them
There is a quick diagnostic for any system claiming to hold context. Ask what it returns when two people with different roles open the same record. If the answer is “the same thing,” it holds information. That may be exactly what it should do — a signed agreement is a document and should read identically to everyone. But it is not context, and it will not do context’s job.
The second diagnostic is temporal. Ask what the record looked like a month ago. If the answer is “the same,” and the underlying relationship has moved since, the record is information that has quietly stopped being true.
The systems that treat context as information end up complete and unread. The distinction is not philosophy. It is the specification.
Sources
Fillmore, C. J. (1982). Frame semantics. In Linguistics in the Morning Calm.
Goodwin, C. (1994). Professional vision. American Anthropologist, 96(3), 606–633.
On the limits of the physics analogy, cf. Kuhn on paradigm incommensurability.