Tech Matters: Writing GPT prompts that work

Some Microsoft customers are complaining about the company's relatively new artificial intelligence, Copilot, an add-on to Word, Excel and other Office suite apps, saying that it doesn't stack up to OpenAI's ChatGPT. But Microsoft calls “user error,” according to a report by Business Insider. Either unhappy users are working with older versions of Office that cannot support the new AI integration or they're just not proficient at writing prompts. To use Copilot for business, you must have Microsoft 365 Apps for Enterprise because previous versions of Office — 2013, 2016, 2019 and 2021 — are not compatible. Copilot is designed to work within your Office environment, securely accessing email and files to provide task automation and analysis. If your company has purchased licenses for Copilot and you're not getting the results you were expecting, it is likely a problem with your prompts. Writing good prompts applies to all AI apps, not just Copilot, ChatGPT, Google's Gemini, Anthropic's Claude and the like. Whether you want to generate text, images or video, it all starts with a prompt. Because most new GPT users are asking for text-based output, we'll focus on input for written output. The process is similar across the AI tools: In an open window or panel, you'll see an area to type in your prompt. In some cases, you will be able to provide links to documents you'd like the tool to reference or you will be able to upload these documents. A word of caution: If you're referencing work documents outside of Copilot, make sure this complies with your company's privacy policy. Before you start a prompt, remember that different tools have their own cutoff points for accessing information on the web. It's an important thing to know if you're looking for recent news. So if you see the “Sorry, I don't have information after …,” do a Google search to augment your AI results. You can add this to your prompt. Varying cutoff dates corresponds to the training process that involves feeding it a large dataset of text from the web, books, articles and other sources […]