This post is super practical. It’s all about helping you get ChatGPT-5 working for YOU.
Like 700 million other people, I've spent the last few days living in the messy reality of adjusting to the biggest model launch of the year.
The issue is not "it can't do hard things"—it can. I think the frustration I’ve been reading across the internet runs deeper: it's about losing a thinking (and for some—seemingly feeling) partner you'd come to rely on.
Overnight, OpenAI swapped the AI that knew your work style for a routed system that sometimes thinks deeply, sometimes sprints, and sometimes silently downgrades without warning. It wasn't just that workflows broke—it was that the rhythm of collaboration people had built over months suddenly felt foreign because it was suddenly disintermediated by a router.
I've watched teams lose momentum as they work through the change. I've seen LOTS of comments that amount to “it used to work and now it’s missing [nuance, emotion, thoughtfulness, insert-your-adjective]. Bottom line: the trust that makes AI useful—that sense of "this system gets what I'm trying to do"—became unpredictable.
I've been tracking this from three angles. First, the review—The Complete ChatGPT-5 Review: 5 Real-World Tests and the Playbook—where I pressure-tested the model on work that actually matters and showed where it shines and where it stumbles. Then the reliability note—GPT-5 at Scale: Why Reliability Slipped—which explained how routing, caps, and model retirements created that "something feels off" moment so many people experienced. Yesterday's executive briefing—The ChatGPT-5 Implementation Playbook—brought it into organizational reality: how to roll this out without breaking what your team depends on.
This piece is the missing fourth leg: getting your thinking partner back. Not workarounds or coping strategies—specific moves that make GPT-5 feel like it's working with you again, not just responding to you. I'm going to address the ten most common frustrations I'm hearing (the routing guessing game, chat vs API mismatches, long-context false promises, JSON that drifts, tool-call overconfidence, guardrail friction, and more) and show you exactly which ones Custom Instructions can fix, which ones need API precision, and which ones are simply how this system works now—so you can design around them instead of fighting them.
Why this matters: GPT-5 isn't a single intelligence anymore; it's a policy decision wrapped in a router. If you don't guide it intentionally, it will default to patterns that might not serve your actual work—shallow when you needed depth, verbose when you needed clarity, diplomatic when you needed honest pushback. That's not sustainable for leaders making consequential decisions or builders shipping on deadlines.
The approach here is practical but human-centered: restore the thinking relationship that was working, on purpose. I'll share the exact Custom Instructions that rebuilt that trust for me, the two simple macros that change everything (Depth and Concise), when it's worth paying for reasoning versus speed, and how to keep long-document work grounded with specific citations. The goal isn't perfect prose—it's getting back that reliable thinking partner who challenges weak assumptions, remembers context, and helps you make better decisions faster.
And one more thing: we’re not done talking about ChatGPT-5. I’ve had requests for specific on-screen ChatGPT-5 demos where I share my prompting, requests for ChatGPT-5 specific prompt libraries, and requests for a deep dive on data analysis in ChatGPT-5 (one area where I think it shines). I’ll be tackling all of those over the next week or two! This model is immediately one of the most-used in the world, and it deserves that kind of attention.
But for today—dive in and let’s get ChatGPT-5 working for YOU!
Listen to this episode with a 7-day free trial
Subscribe to Nate’s Substack to listen to this post and get 7 days of free access to the full post archives.
