Most people have developed their own Google search strategies over the years.
THEREFORE:
Most people talk to an LLM the same way they’d talk to a search engine.
They type something.
They get something back.
If it’s not right, they rephrase and try again without a real system.
THEREFORE:
Responses do not get better with each entry to the conversation and the real power of the LLM is not being utilized.
This is because:
An LLM is much more capable at understanding the nuance in human language than Google.
The problem isn’t the LLM.
The problem is the loop the prompter is using.
Or more precisely, the lack of one.
There’s a decision making framework developed by a U.S. Air Force fighter pilot named John Boyd.
He designed it to describe how skilled pilots out maneuvered their opponents in dogfights.
The framework is called the OODA loop: Observe, Orient, Decide, Act.
I’m going to show you why it’s the best mental model for talking to an AI and will give you better prompting results.
What the OODA Loop Actually Is
Boyd’s insight was that in any competitive environment, the person who cycles through decisions faster and more accurately wins.
It doesn’t matter how smart one is if one can’t process what’s happening and respond effectively.
The four steps:
Observe — Take in what’s in front of you. No filtering yet. Just look.
Orient — Make sense of what you observed.
This is where your knowledge, experience, and judgment come in.
Boyd considered this the most important step => it’s where you build your understanding of the situation.
Decide — Choose a course of action based on your orientation.
Act — Execute the decision. Then the loop starts again with a new observation.
It’s a cycle, not a checklist.
The output of each Act feeds the next Observe.
The loop keeps running until the problem is solved, or the pilot lands the plane, or the prompter’s LLM interaction is complete.
Why This Maps Perfectly to One Chat Turn
Here’s what a single LLM conversation turn actually looks like when one is doing it right:
Observe — LLM responds.
The prompter reads the full response.
Not a skim.
The whole thing.
They’re taking in raw information — what it said, how it said it, what it included, what it left out.
Orient — THIS IS KEY! >> Beginners will typically skip orientation entirely.
Orientation is where the prompter asks: Why didn’t this hit the mark?
Was the prompt too vague?
Did LLM misunderstand the audience?
Did I leave out a key constraint?
Was the output format wrong?
Is this actually close and just needs one adjustment, or is the whole direction off?
Why this is important: One can’t make a good decision (step three) without this step.
THIS IS KEY! >> And one can’t skip it by going faster.
Always Remember: “Slow is smooth .. smooth is fast”
Decide — Based on your orientation, you choose what to change.
Not everything.
The one thing that matters most.
Maybe you add more context.
Maybe you specify the output format.
Maybe you reframe the task entirely.
One variable at a time.
Act — You write and send your next prompt.
Not a blind rephrase.
A targeted adjustment based on a reasoned decision.
Then you observe what comes back. The loop runs again.
Elevate from Beginner to Power User
Be careful of this pattern:
prompt, read, retype, repeat — with no clear theory of what went wrong or what’s being changed.
This is guessing with extra steps, not iteration.
The OODA loop fixes this because it puts Orient where it belongs — between observation and decision.
It forces the prompter to build understanding before the prompter acts.
THIS IS KEY! >> And when one acts .. one acts on that understanding .. not on frustration.
Boyd’s original insight was that pilots who cycled through the loop faster and more accurately won engagements.
Speed matters .. but not at the expense of orientation.
A fast, wrong decision is worse than a slightly slower, right one.
The same is true in a LLM conversation.
The goal isn’t to send more prompts.
It’s to send better ones.
What This Looks Like in Practice
Let’s say someone asks an LLM to write an email to a prospect who went cold after a demo.
The LLM gives the prompter something generic. Professional, but it could have been written for anyone.
A beginner might rephrase: “Make it less formal.”
An OODA operator pauses and orients:
LLM didn’t know anything about this specific prospect..
It had no context for why the demo went well, what the prospect’s hesitation might be, or what the next step should be..
THEREFORE:
!!The output was generic because the input was generic!!
Then decides:
I need to add specific context .. what was discussed in the demo, what the prospect said, what I want them to do next..
Then acts:
“Rewrite this email for a prospect named [Name] who runs a 12-person HVAC company.
In our demo last Tuesday he said he was interested but needed to check with his partner.
The goal of this email is to make it easy for him to forward to his partner with context.
Keep it under 150 words.”
That’s not a better rephrase.
That’s a different prompt — built from observation, orientation, and a deliberate decision.
The LLM now has what it needs.
The Takeaway
The OODA loop isn’t complicated.
BUT: It requires something most people aren’t used to doing => pausing between stimulus and response.
THEREFORE:
Read the full output.
Diagnose what’s wrong and why.
Make a deliberate decision about what to change.
Send a targeted prompt.
Do that consistently and you’ll get more out of ten prompts than most people get out of fifty.
Want to learn how to build systems like this — not just use them one conversation at a time?
Register to find out more.