Navigating AI in the Writing Industry—Part 3: What AI Actually Does With Your Words

Let’s face it, AI is everywhere now, from the likes of ChatGPT and Copilot, to audio converters, spell-checkers, and even plagiarism checkers. Yet, despite AI’s ubiquity, writers everywhere are left with burning questions as to how it works. For example, right now, many writers may be asking themselves:

• When a user inputs original work into any AI of any kind, what actually happens with that text?
• How safe are my manuscripts and original ideas from theft?
• What can I do about AI taking over? Is it going to cost me valuable writing work?

Even as I type right now, I have to ask myself, isn’t AI listening in?

Yes, it is. But what does this mean? If you haven’t already, see Parts 1 and 2, then continue following this series for the expose on what exactly happens with any text we input or share on ChatGPT or any other AI tool.

What do you think? Should AI be used in writing and, if so, where should we draw the line? Drop your ideas below.

As AI-generated content becomes more prevalent, expect to see new regulations and licensing models emerge. As you might know from Part 2 of this series, some publishing companies are already negotiating deals where AI companies would pay for access to copyrighted books, similar to how Netflix licenses films.

But in the meantime, writers must be proactive in protecting their work. AI is neither an unstoppable monster nor a miracle tool—it’s just another shift in the industry, like the internet was decades ago. Those who learn to navigate it wisely will be in the best position to thrive.

So How Do AI Tools Really Work?

When a book or any text is uploaded into ChatGPT, it is not stored, memorised, or used for future responses. Here’s how it works and why it would not reproduce content word-for-word:

1. Limited Data Retention

Originally, ChatGPT did not retain memory across conversations—meaning anything uploaded or discussed in one chat wouldn’t influence future interactions. However, OpenAI has since introduced memory features, allowing the model to remember certain details about users over multiple interactions.

That said, memory and uploaded content retention are separate things:

• Uploaded text (like a book)—ChatGPT does not store or memorize it permanently. Once the session ends, that text isn’t retained or used in future chats.
• User-specific memory—If enabled, ChatGPT can recall broad facts about you (e.g., your writing projects, business, or preferences) to improve responses across sessions. However, it doesn’t store or reproduce exact wording from past interactions.

So, while ChatGPT’s memory has evolved, it still doesn’t function like a long-term database for storing or retrieving previously uploaded text word-for-word. If you ever want a clean slate, you can clear the model’s memory via settings.

2. No Exact Reproduction

ChatGPT, the leader in generative AI programs, is designed to avoid regurgitating copyrighted material verbatim. Unlike an eBook reader or document viewer, which stores and displays content exactly as it was uploaded, ChatGPT works by analysing the text and generating responses based on patterns, rather than recalling specific passages. Even when asked directly, AI will not produce long, word-for-word excerpts from copyrighted texts.

3. Why Can’t AI Copy Books Word-for-Word?

• Training vs. Processing: The AI was trained on publicly available data before deployment, but it doesn’t actively learn from uploads or user interactions.
• Probability-Based Responses: ChatGPT generates responses based on probabilities rather than recall—meaning it predicts words in context rather than retrieving exact sentences.
• Copyright Protections: OpenAI has put restrictions in place to prevent large-scale text reproduction. Even if AI were trained on parts of a book in the past, AI would not recall it exactly or be able to produce it in full.

4. What Happens If Someone Tries to Upload a Book?

If a user uploads a book and asks for a summary, analysis, or key themes, AI can process that within the conversation and generate useful insights. However, AI would not store that book or be able to recall it later. If someone asks for the full text, AI would refuse.

5. The Misconception About AI “Stealing” Books

Some people fear that AI scrapes books and regurgitates them, but that is not necessarily the case. It certainly is not how ChatGPT works. Rather, here are a few facts that remain true for AI at the time of writing:

• AI models are not like search engines or PDF readers that store exact files.
• There are legal and ethical guardrails in place to prevent AI from copying large chunks of copyrighted material.
• The biggest copyright concerns are actually with AI training data, not with how AI like ChatGPT responds in conversations.

6. The Real Copyright Debate: AI Training vs. AI Output

The major legal and ethical debate isn’t about ChatGPT reproducing books—it’s about whether AI models should have been trained on copyrighted books in the first place. Some AI models have used large datasets (sometimes containing books without permission) to learn language patterns. This is why lawsuits have emerged, arguing that AI companies should pay licensing fees for training data—similar to how music streaming services pay artists.

In Conclusion

• Uploads are not stored or remembered.
• Chat GPT does not generate exact word-for-word copies of books.
• AI generates responses based on probability, not memory.
• The copyright debate is more about AI training data than AI-generated output.

Need Help Publishing Your Work?

At Harvard Ink, we provide expert ghostwriting, editing, publishing support, and copyright guidance for writers navigating today’s fast-changing industry. Whether you need help registering your work, refining your manuscript, or launching your book, we’re here to help.

For all your publishing and book polishing needs, contact https://harvardink.com or info@harvardink.com.

Photos (in order) by:

• Markus Spiske, 2018, Pexels.
• Tara Winstead, 2021, Pexels.