TL;DR
Cambridge Dictionary named human its 2026 Word of the Year, citing changes in how people talk about AI and human activity.
The revealing examples are phrases such as human-written, human-made and human-generated. We’re adding information about authorship where it once often went unstated.
Labels can help people make informed choices. They don’t prove a work is good, original or truthful, and a label itself still needs some way to be trusted.
Cambridge Dictionary’s Word of the Year for 2026 is human.
It’s an old word. Cambridge chose it because its role in everyday language is changing. We’re increasingly adding human to activities and things that once didn’t need the distinction: human-written text, human-made objects and human-generated content.
Before generative AI, you could usually read an article without first wondering whether a person wrote the sentences. Now that question is part of the experience. The work can look perfectly ordinary, and you may have no way to tell from the finished result.
We’re having to explain where the work came from.
An ordinary word is doing a new job
Cambridge announced the selection. Its lexicographers cite language data, online searches and the rise of compounds such as human-made. Their Word of the Year feature asks whether the growth of AI content is changing the value people place on human involvement.
Cambridge goes a step further. Its lexicographer describes a “human premium” and says language data, internet trends and search habits confirm it. That data shows what people are talking and searching about.
We didn’t need to specify human-written every time someone handed us a novel or an email. We might question whether the author was credible or whether the claims were true, but the origin of the sentences was rarely the first issue.
Generative AI has introduced another possibility. The byline can belong to a person while some or all of the sentences came from a model. A human may have researched, outlined, directed, edited, approved or simply copied the result.
Adding human to a description is one way people are trying to make the distinction visible.
I wrote about an earlier, more concrete version of this in Human-Made Is Becoming Metadata. Music platforms are already making distinctions about AI-generated recordings that can affect how tracks are handled. That article was about platform rules and the evidence producers might need. Cambridge’s announcement points to something broader: the same question is entering ordinary language, even when no platform requires a disclosure.
A label is a claim about origin
The Authors Guild has taken that idea beyond vocabulary. Its Human Authored certification allows eligible authors to register books as written by people rather than generated by AI.
The definition is deliberately narrow. The Guild allows only minor AI help, like spelling and grammar fixes. If AI rewrote whole sentences, the book doesn’t qualify. In May 2026, it clarified the terms and dropped references to AI brainstorming and outlining. The current FAQ makes the boundary more explicit.
The certification rests on the author’s own statement, backed by contract terms the Guild says it can enforce. A registration number can confirm the work was registered with the Guild. However, it isn’t a forensic test that independently reconstructs every sentence’s origin. The program covers books for now, not articles.
I wrote earlier in The Detector Is Not the Evidence about the danger of treating an AI text detector’s score as proof of authorship. A declaration of human authorship has a different weakness. It depends on whether the person making the claim is honest and precise about it. A contract gives the claim consequences. It still doesn’t let a reader see how the work was made.
Do people actually prefer human work?
There’s some evidence that what people believe about a work’s origin affects how they value it, but the answer depends on how researchers ask the question.
A study by Oliver Jacobs and colleagues, published online in 2025 and collected in a 2026 journal issue, tested how people judge art labeled as made by a human or by AI. Every painting was actually made by a person. Only the label changed. In one experiment, 250 people rated paintings one at a time, and the label made no measurable difference. In the other, 102 people saw two identical paintings side by side, one labeled human and one labeled AI. They picked the human-labeled one more often as the one they liked and valued.
So in a side-by-side choice, the label alone changed what people preferred. The authors also name the limits. The AI labels were always false. They didn’t check whether people believed them. And the participants were recruited online in the US.
One reason I resist making human into a simple quality grade is that people make plenty of mediocre, mistaken or dishonest work. An AI-assisted result can be helpful, careful and well checked. Knowing the method matters, but the method doesn’t settle the quality.
The more valuable question
I think there are at least two questions a reader deserves to separate.
Who made this? That’s a question about origin. Did a person write the actual sentences? Did a model generate them? Was the AI doing research, editing, drafting or almost all the work? A clear answer helps people decide what they want to consume and what kind of creative contribution they want to support.
Who stands behind it? That’s a question about responsibility. Who checked the sources? Who can explain the judgment calls? Who will correct an error? Human authorship doesn’t answer those questions automatically. Neither does an AI-use disclosure.
For organizations, there’s a practical difference. A policy that requires a “human in the loop” may tell you someone had an opportunity to review the output. It doesn’t establish that the person examined the source material or had authority to reject it. Those are design and accountability decisions.
I’ve written about who keeps decision authority when you work with AI in Working Alongside: The Collaboration Asymmetry. This time the issue starts earlier, with the information readers receive about how the work was produced.
The dictionary picked a word people are using differently. That’s an interesting signal about what readers have started to ask.
I’d like labels that say something specific and can be backed up. If a piece says “human-written,” I want to understand what that excludes. If it says “fact-checked,” I’d want to know what was checked and against which sources.


