The Power of Staying in the Thread
In my last post, I teased how fast that blog came together. Here's the answer — and how it worked.
In Roast This Post, I shared how one little prompt — "roast this" — helped pull my pitch deck out of the land of polite mediocrity. I left you with a teaser: Did I use generative AI to write that blog post too? And, if so, how long did it take?
Answer: Yes. Time: About 20 minutes. All-in. Start to publish-ready.
This is a case study in what happens when you use the same AI model, in the same thread — and let it remember.
The Setup and Result
I spent several sessions with ChatGPT refining a pitch deck for Anthralytic using a simple "roast this" prompt for each slide. A few days later, I decided to share that experience so others can harness the power of the prompt and I wanted to us AI as a writing partner. Instead of starting fresh, I stayed in the same chat thread and prompted:
"Write me a blog post about my experience using AI to help me with my pitch deck. Use the history from our work on the deck. The main point is that the prompt I found most useful was 'roast this slide.' Give examples."
The result: A solid first draft that captured my voice and our the experience that I could tweak. In 10 seconds. Just from a prompt.
When I sat down to begin, I didn't start with "So I'm working on this blog piece..." I didn't have to re-upload files, re-explain my style, or remind the AI what we'd been working on. I just said "Write about our experience." It pulled from our back-and-forth and delivered something that sounded like me about the experience. No ramp-up. No context-setting.
The first draft incorporated my distaste for buzzwords and preference for sharp, direct language. It remembered which slides had worked and which ones I'd scrapped. It even knew my annoyance with a triangle graphic that made no sense. It was a great first draft.
Then I worked to tweak it using ChatGPT again as a partner. When I wanted to add more context but couldn't recall what I'd written in an earlier version, it pulled up the exact text: "This is what you said on April 27. Want to reuse that phrasing or make it sharper?" It felt like working with someone who'd been in every meeting, not starting over with a new assistant every time. The reason this worked? I stayed in the thread.
So, 10 seconds of first draft generation and about 15 minutes of tweaking, and five minutes of editing. Voilà!
Two types of AI memory
1. Thread continuity
Everything you and the model say in one chat remains in its “field of view,” so you don’t have to start from scratch. That’s why a simple prompt could lean on all our earlier pitch-deck work.
How it works — Each new prompt resends the whole thread (up to the model’s limit) to the foundation model—GPT-4-turbo, Claude 3, Gemini, etc.—which processes the text as tokens (tiny chunks of a few characters).
Capacity — GPT-4-turbo can manage a thread of up to about 300 pages (~ 128 k tokens), Claude 3 Opus about 500 pages (~ 200 k), Gemini claims several thousand pages (~1 M), and Perplexity just takes whatever limit the chosen model has—past those limits, older messages are summarized or dropped.
Persistence — Close the browser, come back to the same thread days later: the entire thread is still usable. Delete the chat and the history vanishes.
Tip — Keep each big project in its own thread so the model can instantly reuse prior slides, notes, and edits—no copy-paste, no lost context.
2. Cross-conversation memory
This lets an AI remember useful details between different chats—preferences, ongoing projects, writing style—so you can open a new thread and pick up where you left off.
Who has it — ChatGPT Plus offers the most mature system (view, edit, or erase what it stores). Gemini Advanced has a newer, still-growing version tied into Google Workspace.
Who doesn’t — Claude keeps no history by design. Perplexity, DeepSeek, and LLaMA 3 are stateless; any “memory” comes only from threads you save manually.
Bottom line
For deep work inside a single project, thread continuity is usually enough. If you need the AI to remember preferences across multiple projects, you’ll want cross-conversation memory—available today in ChatGPT Plus and, increasingly, Gemini Advanced. Other tools excel at one-off tasks but forget you once the window closes.1
When Staying in the Thread Pays Off
This isn't just about blog posts. The same principle applies when you're building anything over multiple sessions. No need to starting each conversation with "So I'm working on this presentation..." or "Remember, my audience is..."
The AI already knows. It's been there for the whole process.
This becomes particularly valuable when you're:
Refining presentations over multiple sessions (it remembers which slides worked)
Developing content series with consistent voice (it knows your style by now)
Building on strategic documents (it recalls your key decisions and reasoning)
Iterating on creative projects (it understands your aesthetic preferences)
The time savings add up, but what really matters is the quality of continuity. No creative fatigue from explaining yourself repeatedly. No drift in tone or approach. No losing track of what you decided three iterations ago.
Privacy Considerations
Memory features involve trade-offs. Consider your comfort level with data persistence and start new threads for sensitive topics. You can manage memory settings in most platforms, but it's worth understanding what each tool retains.
Key Takeaway
This isn't revolutionary—it's practical workflow optimization. The "magic" isn't in the AI's memory capabilities, but in understanding how to structure your work to maintain context efficiently.
Whether you're using ChatGPT, Claude, or other tools, staying in the thread eliminates the friction of constantly rebuilding context. It's a small change that compounds quickly for anyone doing iterative work.
Working on mission-driven projects and looking for ways to streamline without sacrificing quality? That's exactly what we focus on at Anthralytic. This newsletter shares practical insights on impact strategy, evaluation and AI tools—no hype, just what actually works.
Note, in the world of AI things move fast. What is true today may not be tomorrow.