Prompting is the practice of giving instructions or input to a language model like GPT-5, to guide its response. Think of it as a conversation starter, but with precision. A prompt can be a question, a command, or even a few keywords, and the way you phrase it can dramatically affect the quality, the tone and usefulness of the output.
Prompting isnât just about asking questions, though, itâs about knowing which model youâre speaking to, what itâs optimised for, and how to guide it effectively. GPT-5, for example, is highly obedient and precise, but that means vague or conflicting instructions can derail its reasoning. Earlier models might have guessed your intent, GPT-5 will try to follow it to the letter.
In this post, weâll explore how GPT models have evolved, what makes GPT-5 unique, and how understanding these differences can help you craft prompts that get the best results, whether you're coding, writing, analysing, or building agents.
Understanding GPT Models
The journey from GPT-3.5 to GPT-5 has been less about adding raw power and more about refining how these models think, follow instructions and also interact with us.
GPT-3.5 was a breakthrough for many uses; it could generate surprisingly human-like text, but would often need careful nudging. It also has a tendency to âhallucinateâ facts and would sometimes prioritise sounding fluent over being accurate. Prompts that were short and simple worked best but complex tasks involved a lot of trial and error.
When GPT-4 was introduced it had strong reasoning and context awareness. It became better at accepting layered instructions, and was able to produce more reliable and detailed answers. GPT-4 also introduced multimodal abilities in some versions meaning it could analyse not just text but images as well. However it relied on prompt clarity, vague instructions could send it off course completely.
GPT-5 by contrast is a highly obedient and precise model. Where earlier models might have guessed your intent, GPT-5 tends to follow instructions to the letter. This means you can achieve great accuracy with the outputs, but it means that if you give it a poor or conflicting prompt it can completely derail the output. GPT-5âs reasoning is more structured, and its optimised for multitasking across coding, writing, analyst and even multimodal workflows that combine text and image inputs.
In short the evolution from GPT-3.5 to GPT-5 has gone from âcreative but sometimes unpredictableâ to âstructured, obedient and powerfulâ. Understanding this shift is crucial, because prompting GPT-5 isnât about hoping it interprets your intent, itâs about telling it exactly what you want and in the clearest way possible.
Prompting for GPT-5
To get the most out of GPT-5, think about three key qualities that this model is based on:
- Precision , it will do what you ask, not what you meant.
- Multimodal input , it can handle text, images and even audio together.
- Reasoning depth , it carries context across longer conversations and can break down problems step by step.
These strengths bring some challenges. A single vague instruction like âHelp me with cybersecurityâ often produces a broad and unfocused answer. A clearer prompt would be:
âAct as a cybersecurity consultant. Explain three common phishing techniques in nontechnical language, suitable for a business audience in under 300 words.â
This gives GPT-5 the framing it needs to deliver something useful. Also take note of how weâve told it to act like someone, given it boundaries and a task, all built into the prompt.
When youâre writing prompts, keep a few principles in mind:
- Be explicit and structured, spell out the task and desired format.
- Use roles or personas to shape the modelâs perspective.
- Break down complex jobs into steps rather than asking for everything at once.
- Add examples or constraints if you want a specific style or output.
Instructional Modifiers: Steering GPT-5 with Simple Phrases
When prompting GPT-5 youâre not just asking questions, you are giving instructions. One powerful way to guide the modelâs behaviour is through instructional modifiers. These are short, directive phrases that you include in your prompt that include how GPT-5 thinks, responds and prioritises information.
Think of them as behavioural nudges. They dramatically improve the quality and the relevance of the output you get from your prompts.
Here are some modifiers you can start to build into your prompts:
- âBe quickâ. This encourages GPT-5 to respond concisely and with low latency, which is great for summaries code snippets or fast answers.
- âDo ultra thinkingâ or âThink deeplyâ. These activate GPT-5âs reasoning mode. It will take longer for an output to be delivered, however it will give you a more thoughtful, structured and multi-perspective response.
- âThink step by stepâ. When you write this in your prompt you are asking GPT-5 to break down its reasoning into clear, logical stages. This is great for problem solving or planning.
- âChallenge your assumptionsâ. Having this within your prompt encourages critical thinning and self-reflection which can be useful when you are approaching a complex or controversial topic.
Advanced Prompting Tips for GPT-5
Once you are compatible with basic modifiers you can start to take your prompting to the next level with these advanced techniques.
You can start to use what is called meta prompting , where you ask GPT-5 to evaluate or improve your prompt. This is useful when you arenât sure how to phrase your prompt. You can say:
_âYou are a prompt engineer. Review this prompt and suggest improvements: âinsert promptâ. _
You can also control how big the output you get from GPT-5 is. You can ask for a âbrief summaryâ or a âdetailed summaryâ or even set word limits. Such as saying: âSummarise this in under 100 words. Be quick.â
Donât forget to include roles or personas within your prompt. It can help to shape itâs tone, depth and approach as well. You can build a prompt such as:
_âAct as a gardening expert. Help create a crop rotation plan for an eight bed allotment that is in Glasgow, Scotlandâ. _
The takeaway
GPT-5 is the latest model being used by ChatGPT, and Microsoftâs Copilot, it can be a very effective model. Interacting with it is less like chatting with a helpful friend and more like briefing a brilliant strategist.
But like any strategist you need to give it a detailed and strong brief. The way you shape your prompts directly influences the quality of the output youâll receive. So by understanding how GPT-5 thinks and building your prompts in a way it will value you can unlock itâs full potential.
Top comments (0)