Definition
An AI-assisted interaction, labeled Human + AI in Proof of Actor, is an interaction that a person directs while using AI assistance. The person decides what to do and remains in control; an AI system helps draft text, summarize information, suggest next steps or fill in parts of a task.
It sits between two other classes. Unlike a purely human interaction, AI contributed to what the application receives. Unlike an autonomous agent, a person is still choosing and confirming the actions. For how this class sits next to Human and Agent, see Human, AI-assisted or agent: a practical taxonomy for application traffic.
Why it matters
Many applications have workflows where AI assistance changes what the right response is:
- A support platform may want to label tickets drafted with AI so agents can route and review them appropriately.
- A content or review platform may welcome AI help but require disclosure.
- An assessment or certification workflow may restrict AI assistance for specific steps.
- A productivity app may want to measure how often its own built-in assistant is used.
In each case, the application benefits from knowing that assistance was involved, while still giving the person a normal, respectful experience.
How AI assistance can be recognised
The key word is observable. An application can only recognise AI assistance it can actually see or that is declared to it.
Declared assistance
The person or the application states it. Examples: a "drafted with AI" checkbox, a disclosure field in a form, or an application flag set when the user invokes the app's own assistant. In Proof of Actor terms, this is a declared evidence basis.
Directly observed assistance
A supported integration sees the assistance happen. The clearest example is an application's built-in AI feature: the app knows when it generated content for the user.
What usually cannot be seen
When a person asks a chatbot in another browser tab, or on their phone, and types or pastes the result into a form, ordinary web traffic typically shows nothing about it. The session looks human-directed because a human directed it. Reliable classification requires evidence, and in that case the evidence is not present.
Common misconceptions
"AI-written text can always be detected." Content-based guesses are unreliable and are not a basis for treating someone differently. Human + AI should rest on declared or observable evidence.
"Human + AI is a lesser category." It is not a risk label. In many products it is the expected, encouraged way to work. The class exists so the application can respond appropriately, not to penalize anyone.
"If assistance is not detected, none was used." A Human result means the observable signals indicate human direction. It does not prove that no AI was used anywhere.
Designing for AI-assisted users
Good practice is to make disclosure easy and the consequences clear. If a workflow restricts assistance, say so in the interface and ask for a declaration at the right moment. If assistance is welcome, use the declaration to improve routing, analytics or review, not to add friction.
How Proof of Actor treats it
Proof of Actor reports Human + AI only when assistance is declared or observed by a supported integration, and always states the basis. Your application decides what the class means for each workflow. Read more in Detection & Limitations or apply for early access.