Beyond The Prompt: Getting Better Results from AI

When you ask AI to prepare a report, summarise a document, or analyse information, the response may looks polished, but still feel generic. This often happens because the request does not provide enough context. AI may not know the intended audience, the purpose of the task, which details matter most, or what the expected output should look like. 

When the result falls short, it is tempting to rephrase the prompt and try again. However, changing the wording alone may not solve the problem if the underlying gaps remain unaddressed. 

Getting useful AI output requires more than writing a good prompt. It involves providing the right context and reliable information, understanding the task's complexity, recognising the AI's capabilities and limitations, and reviewing the response before using it.

Start With a Clear Brief

When a request is vague, AI has to interpret what the user wants and make assumptions along the way. 

For example, asking AI to "prepare a report on customer satisfaction" leaves several questions unanswered. Who will read it? What should it cover? What should the report help the business understand? 

Before sending a prompt, consider including: 

• Task: What exactly should AI do? 

• Context: What background information or source material does it need? 

• Audience and purpose: Who will use the output, and what should it help them achieve? 

• Requirements: What limitations, priorities, or details must it consider? 

• Format: Should the response be a report, summary, table, presentation, or list of recommendations? 

Northeastern University's guide to effective AI practices highlights specificity, context, roles, and output formats as ways to make AI responses more relevant. OpenAI's prompt engineering best practices reinforces the same principle: clear instructions and sufficient context help guide more useful responses. 

Refine the Response 

Even with a clear prompt, the first response may not be exactly what you need. Complex tasks often require several rounds of refinement. 

Instead of asking AI to complete everything at once, break the task into smaller steps. For example, when preparing a business report: 

• Start by asking AI to identify the key findings. 

• Ask it to organise those findings around the report's objectives. 

• Request a draft, then provide feedback on what needs to be improved or clarified. 

You can also share examples of previous work or preferred formats to guide the result. Google Cloud's prompt design strategies similarly emphasises examples, task breakdown, and iterative refinement. 

Each round of feedback helps bring the response closer to the intended result.

Review Before You Use the Output

Even a well-written response can contain inaccurate information, overlook important details, or include references that do not exist. 

OpenAI's guidance on ChatGPT accuracy and limitations highlights that AI can present incorrect information confidently. Human review is therefore essential. 

Before using the output, check: 

• Accuracy: Are the facts, figures, and references correct? 

• Relevance: Does it answer the original request? 

• Completeness: Has it missed any important requirements? 

• Usability: Can the intended audience understand and use it? 

AI can also help identify gaps in its own response, but important claims should still be verified against reliable sources. 

The next time you use AI, start with a task where a better result could make a real difference to your work. Treat the AI response as a working draft, then bring in your knowledge of the task, your audience, and your organisation to shape the final result. With practice, you will become more confident in recognising where AI can help and where your own judgement needs to lead.