ChatGPT Voice Can Now Talk to Your Files and Projects
By Wendy Keir | EmpowerAi™
OpenAI has made what looks, at first, like quite a small change to ChatGPT Voice.
You can now upload a file during a voice conversation and talk to ChatGPT about what is inside it. Voice also works within ChatGPT Projects, which means the conversation can draw on recent Project chats, source material and the instructions already associated with that Project.
I think the second part is probably more important than the first.
Uploading a document and asking questions about it is already familiar in ChatGPT. Being able to do the same thing by voice is useful, particularly for people who find talking easier than typing, although it does not fundamentally change what the AI can do.
Projects change the nature of the interaction because they give the conversation somewhere to belong.
A Project can contain the documents you are working from, the conversations you have already had and instructions describing how ChatGPT should behave within that particular piece of work. Voice can now enter that environment and use the context that is already there.
That starts to feel much closer to an ongoing working relationship with a body of knowledge.
Voice no longer has to sit outside the work
Voice interfaces have often felt slightly separate from the serious work happening elsewhere.
You might talk to an AI to brainstorm an idea or ask a quick question, but once you needed to work through a detailed report, compare documents or refer back to previous decisions, you usually returned to a keyboard and screen.
Files and Projects reduce that separation.
Imagine an author working on a nonfiction manuscript.
The manuscript sits inside a Project alongside research notes, the book structure and previous conversations about the argument. Instead of opening a fresh chat and explaining everything again, the author can speak inside the existing Project.
They might say that chapter six still feels disconnected from chapter three and ask ChatGPT to help them think through why.
The AI already has access to the relevant material and the recent Project context.
The conversation can begin further along.
That matters because a surprising amount of friction in AI comes from having to reconstruct context. We repeatedly explain what we are doing, which document matters, which decisions have already been made and how we want the AI to approach the work.
Projects were already intended to reduce that repetition. Voice now gives us another way to interact with the same accumulated context.
Talking through a document is different from asking it a question
There is something slightly different about speaking when you are trying to understand an idea.
Written prompts tend to encourage us to formulate the question before we ask it.
Conversation allows us to discover the question while we are talking.
You can begin with a vague sense that something does not make sense, explain what you are thinking, change direction halfway through and gradually arrive at the actual problem.
That is very close to the way many experts work with another person.
An author may not know precisely why a chapter feels wrong. A coach may have a framework in front of them and want to explore how it applies to an unusual situation. A consultant may need to talk through several pieces of evidence before understanding what the client really needs.
The ability to speak directly to source material makes that kind of exploration easier.
OpenAI's current voice technology is designed for natural interruption and follow-up rather than treating every exchange as a rigid sequence of question and answer. Adding files and Project context gives that conversation something substantial to work from.
That combination is where it becomes interesting.
This could change how readers interact with nonfiction books
For me, the most obvious publishing implication is around books that are designed to help a reader think, learn or apply something.
A conventional book can explain the author's idea very well.
It cannot know which part the individual reader did not understand.
The reader may finish a chapter and think, "I understand the principle, but I have no idea how this applies to what is happening in my business."
At the moment, they either work it out themselves, search somewhere else or perhaps never apply the idea.
A conversational layer changes that.
If the book and its approved supporting material exist inside a governed environment, the reader can talk through what they have read.
They do not necessarily need to ask a beautifully structured question.
They might say, "I think I understand this, but my situation is slightly different because..." and then explain what is happening.
The AI can help them work through the author's framework using the source material that is already available.
Voice makes that considerably more natural.
This is where AI Book Companion™ becomes less like a chatbot
I have never thought the interesting part of an AI Book Companion™ was simply allowing someone to ask questions about a PDF.
That can be done relatively easily.
The real value comes from preserving the author's thinking while allowing the reader to have their own conversation with it.
That requires several layers.
The system needs the author's source material. It needs to understand the frameworks and the relationship between them. It needs instructions covering what the Companion is allowed to explain, what it can help the reader apply and where its authority stops.
It also needs continuity.
A reader may have discussed one chapter last week and return today with a question that depends on that earlier conversation.
If the system treats every interaction as completely new, some of the value disappears.
Projects already provide a structure for keeping related files, conversations and instructions together. Adding voice means that persistent context can now support a spoken interaction as well.
That is quite close to the direction I have been expecting AI Book Companion™ to move.
The book remains the intellectual source.
The technology creates another way for the reader to stay with the thinking.
There is still an important difference between context and governance
Having access to a book does not automatically mean an AI represents the author accurately.
This is a distinction that becomes more important as these systems feel more natural.
Someone can upload almost any book into a general AI system and begin asking questions about it. The AI may give useful answers.
It can also interpret the material through its wider knowledge, introduce ideas that were never part of the book or confidently extend a framework beyond the point where the author would have stopped.
A reader may not be able to tell when that has happened.
A governed Companion needs stronger boundaries.
The approved sources need to be clear. The instructions need to define how the author's ideas should be handled. The AI needs to know when it can make an interpretation and when it should tell the reader that the author's material does not provide an answer.
Voice does not remove any of those requirements.
If anything, it makes them more important because spoken AI can feel particularly authoritative.
When someone hears an immediate, fluent response, it is easy to experience that answer as more certain than it really is.
The architecture underneath the conversation therefore matters just as much as the conversation itself.
Expert businesses could use the same model
The implications extend beyond authors.
Many expert-led businesses already have substantial bodies of knowledge scattered across documents.
There may be frameworks, programme materials, client resources, workshop notes, articles and operating procedures.
Projects provide a way of bringing some of that material into a defined environment.
Voice then provides a different way of accessing it.
A consultant preparing for a meeting could talk through the relevant documents while getting ready rather than sitting at a desk asking written questions.
A trainer could discuss an exercise and ask the AI to locate the part of the programme where a particular concept was introduced.
A founder might keep strategic documents inside a Project and use voice to think through a decision while the AI refers back to earlier conversations.
The usefulness depends on the quality of the material and the instructions surrounding it.
A folder full of documents does not automatically become intelligence.
Someone still needs to decide what belongs there, which sources carry authority and what the AI is expected to do with them.
That is quite similar to building a good AI agent.
The technology is increasingly capable of holding the conversation. The quality of the system comes from what has been designed underneath it.
There is an accessibility implication as well
Voice also changes who can use these knowledge environments comfortably.
Typing long prompts is not natural for everyone.
Someone with dyslexia may find it easier to explain their question verbally. A person with visual impairment may prefer spoken interaction. Another reader may simply understand ideas better when they can talk them through.
Even people who are perfectly comfortable typing may find voice more useful when they are reflecting rather than producing something.
Books have traditionally required the reader to adapt themselves to the format.
Interactive AI begins to make the format more adaptable to the reader.
That does not mean voice will replace text. I doubt most readers will want an entirely spoken experience.
The interesting possibility is moving between them.
Read the chapter.
Ask a question by voice.
Look at the relevant passage.
Continue the discussion later inside the same Project.
The technology starts to sit around the reading experience rather than requiring the reader to leave it.
The book is gradually becoming a working environment
This is the part I find most interesting.
For a very long time, publishing has treated the finished book as the end product.
Once the manuscript is edited, designed and published, the author's work is largely finished. Readers then take responsibility for interpreting and applying what they have read.
AI is beginning to change that boundary.
A book can increasingly sit inside a larger environment where the reader can continue exploring the ideas.
Files provide the source.
Projects provide continuity and context.
Voice gives the reader a more natural way to engage.
None of those things individually creates a particularly radical change.
Put together, though, they begin to alter what a digital book experience could become.
The reader is no longer limited to receiving the author's thinking in one direction.
They can begin to work with it.
Three Key Insights
- ChatGPT Voice can now work with persistent knowledge rather than isolated conversations.
Files can be discussed directly in Voice, while Projects allow spoken conversations to draw on existing chats, sources and Project instructions. - Voice makes AI interaction more useful for exploratory thinking.
Authors, readers and experts can talk through an idea while the AI refers to the source material rather than having to formulate every question as a written prompt. - AI Book Companion™ increasingly has the technical building blocks for a natural reader experience.
Source material, continuity, instructions and spoken conversation can now exist much more closely together, although careful governance is still required to preserve the author's intellectual boundaries.
Three Questions I'm Thinking About
- Will readers eventually expect to be able to discuss a nonfiction book while they are reading it rather than searching elsewhere when something is unclear?
- Could Projects become the underlying environment for personal knowledge systems built around individual authors and experts?
- As voice becomes more natural, how do we make sure the reader always understands where the author's thinking ends and the AI's interpretation begins?
Sources
OpenAI — ChatGPT Release Notes: Files and Projects in ChatGPT Voice
OpenAI — ChatGPT Business Release Notes
OpenAI — Realtime Conversations Documentation
EmpowerAi™ | Real AI. Real Impact. Smarter Business Decisions.
AI Book Companion™ helps expert authors transform a completed book into a governed, interactive knowledge asset built around their own ideas, methodology and boundaries.