Learn how AI prompting really works. Discover advanced prompt engineering techniques like personas, chain of thought, few-shot prompting, and trees of thought to get better results from ChatGPT and large language models.
Let’s clear something up. AI prompting isn’t “just asking ChatGPT to do stuff”. It’s programming with words. Large language models (LLMs) don’t think the way we do. They are prediction engines, sophisticated autocomplete systems. When you write a prompt, you’re not simply asking a question. You’re starting a pattern. The model completes it based on probabilities.
If your prompt is vague, it guesses.
If your prompt is structured, you hack the probability.
That’s the foundation of prompt engineering.
Prompting is a call to action, but more than that, it’s instruction design.
When you write:
“Write me an email about X”
Who’s writing it?
Nobody.
That’s the issue.
Without structure, perspective, or context, the model defaults to generic output. That’s why basic ChatGPT prompts often sound bland. The model is statistically averaging everything it has seen before.
To improve AI output, you need to reduce ambiguity.
One of the most effective AI prompting techniques is assigning a persona.
Instead of:
“Write an email about X.”
Try:
“You are a senior performance marketing strategist. Write an email about X.”
Why does this work?
Because you’re narrowing the knowledge source. You’re defining:
In prompt engineering terms, you’re constraining the probability space.
Better constraints = better output.
Large language models are eager to please. They rarely say “I don’t know.” If information is missing, they fill the gaps.That’s where context comes in.
Compare:
“Give me birthday present ideas under £30.”
vs
“Give me 5 birthday present ideas under £30 for a 29-year-old who loves sport and recently started playing basketball.”
MORE CONTEXT= LESS GUESSWORK.
A powerful addition to any AI workflow is this line:
“If it’s not in the context and you can’t find the answer, say ‘I don’t know.’”
This dramatically reduces hallucinations and increases reliability.
In AI prompting, specificity is the difference between average and accurate.
Modern AI models can browse the web and access external data. That helps solve the “frozen in time” limitation of training cut-off dates.
But there’s a trade-off.
The more tools models have, the more we trust them. And if they retrieve outdated or low-quality sources, you get confidently delivered misinformation.
That’s why structured prompting and validation matter even more in advanced AI workflows.
Another key part of prompt engineering is formatting.
Don’t just ask for content. Specify:
Formatting reduces randomness. It tells the model exactly what “good” looks like.
This is still zero-shot prompting: one instruction, no examples.
But we can go further.
Few-shot prompting means providing examples of the output you want.
Instead of describing tone, structure, and style, you show it.
This is one of the most powerful techniques in ChatGPT prompting because it drastically reduces interpretation gaps.
Chain of thought (COT) prompting tells the model to think step by step before answering.
For example:
Before writing the email, think through:
This increases:
Modern reasoning models often integrate this internally, but explicitly structuring thinking improves repeatable AI systems and complex workflows.
If the chain of thought is linear, trees of thought (TOT) explore multiple reasoning paths.
Instead of asking for one answer, you ask for:
This avoids average answers and encourages diversity of ideas.
For brainstorming, strategy, or creative direction, this is significantly stronger than basic prompting.
Also known as the “playoff method”, adversarial validation creates internal competition.You can simulate:
Each produces or critiques drafts. The best elements are synthesised.
Why does this work?
Because AI models are often better at editing and critiquing than producing first drafts.
By engineering friction, you engineer quality.
All of these advanced AI prompting techniques - personas, context, few-shot prompting, chain of thought, trees of thought - boil down to one thing: Clarity.
Persona forces you to define perspective.
Context forces you to define facts.
Chain of thought forces you to define logic.
Few-shot prompting forces you to define quality.
If AI output is messy, it’s rarely because AI “failed”.
It’s usually because the prompt wasn’t clear enough.
The real skill isn’t clever tricks. It’s structured thinking.
At Studio 34, we don’t just use AI tools. We design AI systems. Because when you can clearly explain a process, you can scale it.
And that’s how ideas perform.
Complete this form to receive the Prompting Guide.
