When AI Sits Between Two People
The Hidden Repetition in AI-Assisted Communication
Generative AI is increasingly becoming part of how people communicate with one another. Someone may use it to organise a complex thought before sending an email, make feedback clearer, structure a client requirement, understand a technical response, refine a proposal, or find better words for something difficult to express. In these situations, AI is not the person we are ultimately trying to communicate with. Another human is. AI helps shape what passes between the two people.
Consider a conversation in which Person A uses AI to help draft a message and sends it to Person B. Person B reads it and responds. Person A then gives that response to AI, perhaps asking it to explain the answer, identify concerns, suggest alternatives, or help prepare a reply. A new, polished message goes back to Person B, who responds again. That response may once more become input for AI, producing another variation of the conversation.
Every individual step can be reasonable. Person A may genuinely be trying to understand the issue better, Person B may be responding carefully, and AI may be doing exactly what it was asked to do: clarify, analyse, expand, or reformulate. Yet after several exchanges, the underlying conversation may begin returning to substantially the same questions, concerns, suggestions, or requests. In some cases, the material being discussed may also begin changing in ways that neither person explicitly intended.
The communication challenge begins when AI becomes part of the pathway through which two people repeatedly interpret and respond to each other.
AI Can Help Us Say What We Mean
There are good reasons people use AI before communicating with someone else. Knowing what we want to say and knowing how to say it are not always the same thing. A client may understand a business need but struggle to describe it clearly. An employee may know that something is not working but find it difficult to structure constructive feedback. Someone receiving a specialist explanation may use AI to make the response easier to understand.
AI can organise scattered thoughts, improve clarity, identify questions that have not been considered, adjust tone, and help someone communicate with greater confidence. A loosely expressed idea can become a structured requirement, a lengthy response can become easier to understand, and a difficult message can become more measured.
Used this way, AI acts as a communication aid without replacing either participant. The dynamics become more complicated when that assistance is repeatedly applied to the interpretation and continuation of the same conversation.
When AI Becomes Part of the Exchange
What begins as help with drafting one message can gradually become part of an entire exchange. A response is analysed, alternative interpretations are generated, questions are suggested, and another message is drafted. The next response may go through the same process.
This can happen with requirements, feedback, proposals, negotiations, revisions, questions, objections, or recommendations. In many cases, the process helps people explore unfamiliar or complicated subjects more thoroughly.
This can happen even when neither person is communicating poorly. The sender may genuinely see a new possibility in an AI-assisted response, while the recipient may reasonably assume that a detailed new message contains new matters requiring attention. The pattern can therefore emerge even when both people are communicating in good faith and AI is doing what it was asked to do.
When Repetition No Longer Looks Like Repetition
Ordinary repetition is usually easy to recognise. When someone asks the same question several times using similar language, the recipient can see that the conversation has returned to an earlier point.
Generative AI makes this less obvious because it is very good at producing linguistic variation. The same concern can return as a question, a counterargument, a revised proposal, a comparison of alternatives, or a detailed recommendation. Vocabulary changes, structure improves, examples are added, and reasoning may become more elaborate.
The resulting message can feel new even when its underlying meaning is familiar. Sometimes this reformulation is valuable: presenting an idea differently may create exactly the understanding that was missing. A new framing can also expose an assumption or possibility that neither person had previously considered. But once the issue has already been understood, repeated reformulation can create the appearance of progress without materially changing the conversation.
This is where an important distinction emerges: reformulation is not always progression. A useful question for either participant is therefore not simply whether the message is different, but what is genuinely new in it.
Generating Is Easier Than Evaluating
AI changes the effort involved in exploring alternatives. Someone can ask for five other approaches, several objections, or multiple possible responses and receive them almost immediately. This is useful for brainstorming, but it creates an important asymmetry between the people involved: the effort required to generate an idea can be much lower than the effort required to evaluate it.
An AI system might produce several technical alternatives in seconds, while assessing them could require knowledge of an existing system, investigation of dependencies, security review, testing, budget analysis, or consultation with other people. The same principle applies outside technology. A contractual suggestion, business proposal, policy alternative, or operational change may be easy to describe but considerably harder to assess responsibly.
The imbalance also appears when documents are repeatedly revised. Creating another version can be quick, while the recipient may still need to determine what changed and whether those changes are significant. AI may then be used again to compare versions, summarise differences, and identify which areas deserve attention. This can save review time, but important changes still need to be verified. The ease of creating another version does not remove the cost of establishing what changed.
When the Surrounding Context Begins to Change
Repeated AI-assisted revision can introduce another, potentially more consequential issue. When AI is used to revise one part of a message or document, the resulting rewrite may also alter surrounding wording, assumptions, or emphasis—particularly when the instruction does not explicitly limit the scope of the revision.
These modifications may make the revised material appear more coherent, but not every change necessarily represents something either person deliberately decided to change.
This becomes particularly important when an evolving document passes through several rounds of AI-assisted revision. A requirement, proposal, plan, specification, policy, or strategy may begin with one version. Feedback is provided on a particular point, AI is used to incorporate it, and another version is produced. Further responses lead to further revisions.
After several rounds, the latest document may differ significantly from the original. Some differences may reflect genuine decisions, newly discovered information, or changed constraints. Others may have emerged from the way surrounding context was rewritten during successive revisions.
Once such a change enters the document, it can influence the next exchange. The recipient may respond to it as though it were deliberately introduced, and that response may become input for another AI-assisted revision. This does not mean every AI-assisted revision will produce contextual drift. It means that when a document matters, the origin and intention of significant changes matter too.
What Has Actually Changed?
As AI becomes more common in person-to-person communication, one useful skill may be learning to evaluate semantic change rather than textual change. When a new message or document arrives, the recipient can consider whether new information has been introduced, an assumption has changed, a constraint has been removed, the desired outcome is different, or a decision has actually been made.
When documents are involved, another question becomes equally important: Was every meaningful change intentional? It can be useful to ask why a particular change exists. Did one person change the requirement? Did new information require the revision? Did a professional identify a constraint? Did both people agree on an alternative? Or did the wording or surrounding context change during AI-assisted revision without either person explicitly requesting that change?
Understanding the reason behind important changes helps preserve the connection between what the people involved decided and what the document eventually says. This is not about distrusting AI-generated content. It is about maintaining human ownership of meaning. AI can propose, organise, rewrite, and explore, but the people involved still need to know which ideas they have accepted and which changes they intend to carry forward.
Keeping the Humans in the Same Conversation
When AI sits between two people, maintaining continuity becomes increasingly important. AI can help formulate the next message, but the humans involved still need to know which points are settled and which questions remain unresolved.
That may mean identifying significant document revisions, explicitly confirming changes that affect scope, meaning, responsibility, or expected outcomes, and avoiding the repeated reopening of points that have already been decided. When several AI-assisted written exchanges begin producing more variations than clarity, a direct conversation may sometimes be the simplest way to clarify the issue before continuing.
The purpose is not to remove AI from the communication process. AI can help people articulate ideas, understand difficult responses, explore alternatives, and communicate more confidently. The goal is to use that assistance without losing the continuity of the human conversation.
What Changes When AI Sits Between Two People?
Much of our discussion about generative AI focuses on a person communicating directly with an AI system. AI-assisted human communication has a different structure. One person’s thoughts may be organised by AI before reaching another person, whose response may then be interpreted through AI before the first person replies. When documents are involved, AI may also influence how shared material evolves across those exchanges.
The intentions, decisions, disagreements, constraints, and consequences still belong to the people involved. AI influences the language and possibilities through which those human elements travel. This makes it increasingly important to recognise when a message has been reformulated rather than meaningfully changed, when continued generation is producing alternatives without progress, and when surrounding context has changed without a deliberate human decision.
AI-assisted communication can make ideas clearer and more accessible, and it can help people consider possibilities they might otherwise miss. The challenge is preserving continuity and human intent while benefiting from that assistance.
When AI sits between two people, communicating well may therefore require more than producing a better next message. It may require both people to retain a shared understanding of what was said, what was decided, what changed, and why.