I’ll never forget the day our daughter was born.
It was February 14, 2007 during a massive blizzard. The snow was coming down so hard. You couldn’t even see the hospital entrance from the parking lot.
That’s the last thing you want to deal with as a new parent.
But even though Madison’s arrival was chaos, my wife and I didn’t just wing it.
We didn’t walk through our front door with this tiny human and go “Now what?”
No way.
We’d spent tons of time planning ahead.
Months before she was born, we were in full preparation mode. We’d painted the nursery, assembled the crib and rocking chair, stocked up on wipes and diapers, got a stroller and car seat, bought a high-chair, odor-locking diaper bin, diaper bag, and the list goes on and on.
You know, thinking back on it now, that was a lot of work!
But even though it was a lot, the planning really paid off.
When we brought Madison home, everything went a lot smoother (or as smooth as it can go with a newborn).
Now here’s what’s really interesting: AI can benefit from the same approach.
The Plan-and-Solve technique does exactly this. It tells the AI to first create a step-by-step plan. Then it executes that plan to get a great final result.
No more “winging” it. No more low quality responses.
The Plan-and-Solve technique is great for getting the AI to come up with a bulletproof plan before it writes your content.
Let me show you how it works.
How it Works: Plan-and-Solve Prompting
The Plan-and-Solve prompting technique is easy to implement in your prompts.
Step 1: Write Your Prompt as Usual
Start by writing your prompt like you normally would.
Now, don’t press ENTER yet. We’ve got to add this technique.
Step 2: Add the Plan-and-Solve Command
Now put something like this at the end of your prompt:
Before carrying out my request, write a step-by-step plan to carry out the request exactly as I specified to the best of your ability. Then execute the plan.
Or if you’re dealing with a “problem” then you could write:
Before solving my problem, write a step-by-step plan to solve the problem. Then, execute the plan to find the solution.
See how easy it is?
One thing you should know is that some AI models might already use this approach natively, but some might not. Also, for those that do, they might not use it all the time. So if this is what you want, manually prompting the AI to carry out the Plan-and-Solve technique is a good choice.
Plan-and-Solve Prompting vs. Chain-of-Thought Prompting
At this point, you might be wondering…
This technique sounds a lot like Chain-of-Thought prompting.
But it’s actually very different.
With Chain-of-Thought, you’re asking the AI to simply “show its work.” Or you tell it to “think step by step.” The AI’s basically figuring things out as it goes while showing its work.
It’s kind of like trying to assemble IKEA furniture without the instructions.
Plan-and-Solve prompting is more deliberate.
The AI must first develop a complete plan. Then it carries out that plan. It can still show its work during execution. But it’s working from a thoughtful strategy.
Not just making it up off the top of its head.
Advantages of Plan-and-Solve Prompting
- Improved Accuracy: By forcing a complete plan upfront, the model is less likely to miss steps or get sidetracked by intermediate calculations. It reduces the cognitive load at each step of the execution phase.
- Better for Complex Problems: For tasks with many dependencies (e.g., complex physics problems, logic puzzles, project planning), creating a global plan is far more robust than the sequential thinking of CoT.
- Easier to Debug and Verify: When an answer is wrong, you can easily check if the error was in the Plan (a flawed strategy) or in the Execution (a calculation mistake). This makes it easier to correct the model’s reasoning.
- More Structured Output: The response is naturally organized, making it much easier for a human to read and follow the model’s logic.
When to Use Plan-and-Solve
PS prompting is most effective for:
- Multi-step mathematical or scientific word problems.
- Logic puzzles and reasoning tasks.
- Complex instruction-following or task planning (e.g., “Plan a 3-day marketing campaign for a new product launch”).
- Code generation for complex functions or algorithms.
A More Advanced Variant: PS+ Prompting
The last thing I want to leave you with is an enhanced version of this technique called PS+ Prompting. When you use this variant, the plan the AI makes is even more detailed and higher quality.
Here’s how you use it:
When you’re creating your prompt, add something like “Also extract all relevant variables, equations, or requirements from the request or problem, and restate the plan using those variables.”
This’ll make the result even better!