AI tools such as ChatGPT are immensely powerful, but it can be hard to get the best from them. A few simple tips can improve your ‘prompts’ and deliver better results
Generative artificial intelligence (AI) technology, such as ChatGPT, has scarcely been out of the headlines for most of the past year. Controversy covers everything from whether the AI models were trained on the work of artists and writers without their permission and the possibility of the technology making masses of people redundant, through to whether it could somehow bring about the end of human life.
The latter becomes incredibly unlikely once you understand what generative AI actually is: a tool that analyses a vast database of existing content – words, pictures, videos, etc – then uses complex algorithms to produce something similar. Despite the name, it isn’t intelligent in a meaningful way, and, as we’ve discussed at length before, it’s no substitute for a human writer. It just makes good guesses about the words or pictures users are looking for.
That’s not to completely diminish its capabilities – its ‘guesses’ can be remarkably accurate and useful. But left to its own devices, a generative AI tool can misunderstand the topic, deliver lifeless prose and occasionally invent its own ‘facts’, among other issues. Avoiding those pitfalls and getting the best out of generative AI is all about understanding its strengths and weaknesses, as well as knowing how to guide it to the output required.
- AI hallucinations: why generative AI invents facts and how to manage it
- Journalism vs The Machine: the journalists win
- The A-Z of artificial intelligence
The instructions users provide are known as prompts and can be as simple as a single question – “Explain the meaning of the term EBITDA” – or as complex as multi-paragraph prompts that tell the AI to write an article and specify how long it should be, what it should include, the desired writing style and more. In other words, once you know what you are doing, you can give it a brief to help guide it to produce what you need.
The following tips will make a big difference to the quality of responses you get from generative AI.
- Define the task precisely
Unless you are using a tool that has been ‘trained’ for you or your organisation, the AI has no idea what you are trying to accomplish. Therefore, the more precisely you define your task, the more useful the results will be. Tell the AI that you need help drafting an article and you will get ideas for structuring your writing or what points to include; tell it you are brainstorming a new product and the AI will lean towards generating as many ideas as possible.
- Provide context
Adding context will further improve the responses. You will get different suggestions, or a different emphasis, if you specify an article for experts, rather than one for non-specialists. Likewise, brainstorming a new product might yield different results if you say it’s for a start-up, rather than for a multinational. A key difference from using a search engine is that you can add a lot of context – entire paragraphs – and the AI will alter its response accordingly.
- Specify the desired output
We’ve touched on this a little already: article drafts and brainstorms are different types of output. But you can be much more precise. Do you want a bullet-pointed list or a detailed breakdown of each point? Do you want suggestions for data points you can include or just an outline of the argument? Generative AI is particularly good at summarising documents but, again, you will get better results with more specific requests. Do you want a short summary or a point-by-point explanation? Are you looking for the most significant statistics on a topic or everything related to one narrow aspect? Be clear about your desired output and that’s what the AI will aim for.
- Give it a persona
One popular way to shape responses is to assign the AI a persona. For example, telling the AI to give feedback on your business ideas from the perspective of a customer will probably deliver results that are quite broad, written in everyday language and which emphasise a non-expert’s concerns. Ask for the response of a marketing expert and you’ll get content that focuses on how to sell the product, that uses more jargon and which emphasises comparisons with similar products.
- Refine your results
An important difference between tools such as ChatGPT and search engines is the chat interface. If a set of search-engine results does not contain what you are looking for, then you need to start again with a new query. A generative AI chatbot remembers your conversation, so you can ask follow-up questions. If you don’t understand part of the AI’s response, ask it to explain. Similarly, you can ask for a different tone, request more detail, better sources or tell the AI you think it’s made a mistake. This is perhaps the most powerful capability of these tools.
The most important thing to keep in mind is that this technology is still very new. It makes things up, so check its facts. And it is worth experimenting with different contexts and personas to see how they change results. Repeat the same response multiple times and the answer will vary – there’s an element of randomness in these models. You can also ask AI to explain how it got a result, which will help you refine your prompts in future.
The capabilities of these tools are changing fast. In ChatGPT, you can already save permanent instructions that the AI will apply to every query. That means you could assign it a permanent persona or tell it to always tailor results to your business sector, for example.
In time, we’re likely to see more pre-defined prompts, or even AIs that will know what to ask so they can define the task without the user needing any expertise in prompting. However, even as that happens, there is still value in being able to tailor prompts to your requirements. The true value of generative AI is that it can be completely personalised. Getting the best out of it will always require some understanding of how to ask the right question.