Word of the Year 2026 · AI Output Specs

Human, for Now

Cambridge Dictionary chose “human” as its 2026 Word of the Year. We used to label what machines could do. Now we’re labeling what humans still do, and the label is only worth something if it says which claim it is making.

November 2026 / 9 min read
Poster titled Human, for Now. Left side: then, we labeled the machine output, with tags AI-generated, AI-written, AI-video, AI-image, AI-slop. Right side: now, we are labeling the human role, with tags human-created, human-edited, human-reviewed, human-curated. Quote: when the artificial becomes ordinary, the human becomes the distinction.
When the artificial becomes ordinary, the human becomes the distinction.
In Plain English

What Cambridge actually said

On 17 November 2026 Cambridge Dictionary named human its Word of the Year. The definition it cites is old and plain: having the qualities, faults, and feelings that people have, “as opposed to gods, animals, or machines.” That last word is doing the work this year.

Lexicographer Colin McIntosh put the reason in one line: “AI is changing how we talk about being human.” The announcement collects the evidence. The Authors Guild has launched a “Human Authored” mark for books. Bill Gates writes about “human reserved jobs” such as caregiving. Andy Weir says he might be among the last generation of “fully human writers.” Artists’ groups talk about “human-generated content” versus “machine-generated replicas.”

Read those phrases again and look at where the adjective sits. In every one of them, “human” is a qualifier. It is attached to something that, five years ago, nobody would have thought to qualify.

Defaults don’t get labels

For centuries, human creation needed no disclaimer. It was simply creation.

Linguists have a name for this: markedness. The ordinary case is unmarked. It carries no label because everyone assumes it. The unusual case is marked, and the label is how you flag it.

Everyday language is full of it. A letter is just a letter, but a handwritten letter is worth mentioning. A shirt is just a shirt, but a handmade one gets its own tag. A tomato was just a tomato until industrial farming made “organic” the thing worth saying. In each case the label moved to the exception the moment the default changed.

The label rarely means “better.” It means “this was made in a way you shouldn’t assume.” Sometimes that difference carries value. Sometimes it only carries information.

The adjective is changing sides

Then AI arrived, and we labeled the exception. We started with the artificial: AI-generated, AI-written, AI-edited, AI art, AI slop. Content policies, platform toggles and disclosure rules all marked the machine side, because the human side was the assumed baseline.

Now the exception is becoming ordinary, and the ordinary is becoming exceptional. So we’re beginning to label the human: human-written, human-created, human-edited, human-reviewed.

That baseline is moving. In my own work, a first draft of a design note, a test file or a summary now usually starts with a model. The interesting fact about a piece of work is no longer “a model touched this.” It is “a person did this specific part.” Cambridge’s list is the language catching up.

Phase Unmarked (assumed) Marked (labeled)
Before writing tools Handwritten Printed, typed
After typewriters, PCs Typed handwritten
Early generative AI Made by a person AI-generated
Now, in more places Drafted with a model Human Authored, human-reviewed

I am not claiming the last row is universal yet. Plenty of fields still label the machine side, and should. My claim is narrower: the human side now needs a label in enough places that the dictionary noticed.

Where the handwriting analogy breaks

The handwritten note is a clean picture of the language pattern. It is a bad picture of the authorship question, and I want to be precise about why.

A typewriter changes the method of writing. The person still chooses every word. A language model can change the author. The words may come from the model, with the person choosing among drafts, editing, or barely touching the result. “Typed” never raised the question of who wrote something. “AI-generated” always does.

So the analogy holds for how labels move. It does not hold for what the label means. That gap is exactly why “human” is harder to use well than “handwritten” ever was.

The short form

Handwriting tells you about the pen. “Human” is being asked to tell you about the author, the reviewer and the person who answers for it. One word, several jobs.

“Human” is four different claims

Every time I see “human” used as a qualifier around AI, it is making one of four claims. They are not interchangeable, and conflating them is how a label turns into decoration.

Label The claim What would back it What it doesn’t promise
Human-authored A person wrote the substance A defined threshold, like the Guild’s de minimis rule That it is good
Human-reviewed A person checked it Who reviewed, against what, and when That the person wrote any of it
Human-in-the-loop A person has a step in the process Which step, and whether they can block it That the person had real authority
Human-accountable A named person answers for the result A name, not a team or a role That the person did the work

The last column matters most. Each label is silent on something people tend to assume it covers. “Human-reviewed” code can be entirely model-written. “Human-in-the-loop” can describe someone clicking Approve on 400 items an hour. A label that leaves its last column to the reader’s imagination is borrowing trust it hasn’t earned.

The sticker problem

The Authors Guild mark is a useful counterexample because it is not just a word. It is a certification mark. An author or publisher has to register the work and agree to a license. It covers books whose text was “fully authored by one or more human beings and not generated by AI,” with a de minimis exception for things like spelling and grammar checkers.

That is what a label looks like when it has a definition behind it: a threshold, a registration step, and a named exception. You can disagree with where the Guild drew the line. You can’t say the line is vague.

Compare that to “human-crafted” on a landing page, or “reviewed by our experts” at the bottom of a generated article. Same adjective, no process. I expect a lot more of the second kind over the next year, because the word is starting to sell.

Where I’d put it instead

This journal is about specs for AI output, so here is the practical version. I don’t want “human” as an adjective in marketing copy. I want it as structured fields in the artifact’s metadata, one per claim, so nobody has to guess which one is meant.

Software already has a habit for this. Git commit trailers like Signed-off-by: and Reviewed-by: have long separated who wrote a change from who checked it. AI tools now add Co-Authored-By: lines naming the model. The human side deserves the same precision.

output-spec.yaml · provenance block
provenance:
  drafted_by: model            # which model, which version
  model: claude-opus-5-5
  authored_by_human: partial     # none | partial | full
  human_edits: "restructured sections 2-4, rewrote conclusion"
  reviewed_by: anuj
  review_scope: "facts and sources checked; tone not reviewed"
  accountable: anuj             # a person, never a team
  reviewed_at: 2026-11-21
same idea as commit trailers
Rewrite provenance section of output spec

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reviewed-by: Anuj Sadani
Human-Edits: rewrote error handling, kept tests   # not a git convention, yet

Two things I like about this shape. First, authored_by_human: partial is honest in a way the adjective can never be, because it admits the middle ground most real work lives in. Second, review_scope forces the reviewer to say what they did not check, which is the column every label leaves blank.

Human, for now

Will the label last? I see two paths, and I think both happen, in different places.

In routine work, “human” fades the way “typed” did. Nobody will label an internal status update or a generated test fixture. The question stops being interesting once the answer is almost always “a model, and someone glanced at it.”

Where provenance carries value or risk, the label hardens. Books, journalism, legal documents, medical notes and code that ships to production will need to say who wrote what and who answers for it. My bet is that it hardens into fields and certifications rather than staying an adjective, because an adjective can’t say partial and can’t name a reviewer.

That is why I read Cambridge’s choice as a word for a transition. “Human” is the adjective we need while the default is moving. Once it settles, the useful version of the word won’t be on the cover. It will be in the metadata.

Perhaps the biggest sign of AI becoming normal isn’t that we stop calling it artificial. It’s that we start calling ourselves human.

The frame to leave on

Disclosure

Fitting for the topic: I drafted this with AI assistance. The argument, the edits and the responsibility for every claim are mine. Sources: Cambridge Dictionary Word of the Year 2026 and the Authors Guild Human Authored usage guidelines.