There’s always a lot of noise when a new model lands. Most of it is hype. The fastest way to find out whether a release actually matters to your work is to stop reading the threads and test it yourself.
So when ChatGPT 5 showed up, I didn’t reach for a benchmark chart. I reached for the same five prompts I run against every new model. They’re not flashy. They’re the jobs I actually do in a week: learn something fast, decide what to work on, block out the time, turn a mess into a plan, and get unstuck on a hard call. If a model wants a spot in my stack, it has to handle these well.
That’s the real test. Not “is it smarter on paper,” but “does it make my Tuesday easier.”
The five prompts I run on every new model
Copy these straight into ChatGPT 5 (or whatever you’re testing this week) and swap in your own brackets. Each one has a role up top, because giving the model a job to hold onto sharpens the output every time.
1. The learning coach
Role: Learning coach. As an expert in [area], I want to rapidly understand [subject]. Apply the Pareto principle: list the 20% of concepts that yield ~80% of the understanding. Provide a concise, ordered list with one-line explanations and a concrete example for each. End with 3 high-leverage next steps or resources and 3 common pitfalls to avoid.
This is my fastest read on whether a model can teach. Weak ones dump a Wikipedia summary. The good ones find the load-bearing 20% and tell me exactly where beginners trip.
2. The prioritization coach
Role: Prioritization coach. Analyze and prioritize: [insert tasks, projects, or ideas]. Score each on Urgency (1-5), Importance (1-5), and Effort/Time (1-5, where lower is better). Compute Priority = (Urgency + Importance) minus Effort. Output a markdown table with scores, rank, brief rationale, dependencies, and a “quick win” tag. Conclude with a 3-item action sequence.
When my list gets long and my brain gets foggy, this one turns a pile of “everything is urgent” into an actual order of operations. The dependencies column is where it earns its keep.
3. The time-blocking coach
Role: Time-blocking coach. Create a realistic, time-blocked schedule to complete [task or project] within [timeframe]. Use focused work blocks, dedicated lunch breaks, and a 15% buffer. Output a markdown table with Start, End, Block, Description, and Deliverable. Flag likely bottlenecks, and always include a simple catch-up block on Friday in case we fall behind.
A plan that ignores lunch and assumes nothing goes wrong is a fantasy. The 15% buffer and the Friday catch-up block are what let the schedule survive a real week.
4. The writing coach
Role: Writing coach. Turn this messy brainstorm into a clean outline and action plan: [insert raw ideas]. First, cluster related ideas into named themes. Then produce (a) a hierarchical outline and (b) an actionable step list (with owner and due date if provided). Call out gaps or open questions that must be resolved before drafting or execution.
I dump half-formed thoughts in and get back structure. The part I lean on most is that last line. It tells me what I haven’t figured out yet instead of papering over the holes.
5. The decision analyst
Role: Decision analyst. I’m stuck on [problem or decision]. Propose 3 very distinct options. For each: summary, pros, cons, likely outcomes, key risks, and confidence (0-100%). Recommend one path with detailed reasoning, then give a 3-step next-action plan plus the decision triggers that would prompt revisiting the choice.
This is the one that does the most actual thinking. “Decision triggers” is the line that separates a useful answer from a confident guess. It tells me what would have to change for me to reconsider.
What the test actually measures
Notice what none of these prompts ask: “are you the best model?” They ask “can you do my work?” That’s deliberate.
Run the same five against any new release and you stop arguing about leaderboards. You start noticing the only thing that matters, which is whether the output is something you’d actually use. Some models nail the learning coach and fumble the schedule. Some build a beautiful table and skip the risks. The gaps tell you more than any benchmark ever will.
It’s not the model. It’s your workflow. The model is a tool. A clear process is what separates beginners from pros.
That line is the whole point, and it’s why I don’t pick one AI and call it done. Different jobs reward different tools, and the real skill is knowing which one to reach for. That’s the entire premise behind how I think about AI tool comparisons: you don’t crown a winner, you match the job to the tool that’s best at it. The five prompts above are just the fixture I use to check the fit, and they’re a core part of how I approach AI for marketing in general.
So don’t take a new model’s launch post at face value. Take it for a drive. Five prompts, fifteen minutes, real work. You’ll know in one sitting whether it belongs in your week.
Try it this week
Pick the prompt that maps to whatever’s stuck on your plate right now and run it. Then hit me with the one that surprised you.
I run these prompts on every model so you don’t have to
Twice a week I send one short email with that week’s results: which of these five prompts a new model passed, which one it flubbed, and which tool quietly lost a job it used to own. You get the next batch of test prompts before I post them anywhere else.