Now with over 180 million users and a fully-fledged premium offering, ChatGPT is becoming a key component in the week of an entrepreneur. Used well, the platform can write like you, spark game-changing business ideas, transform your personal brand and help you make tough business decisions.
But that’s only if you know what you’re doing; prompting ChatGPT effectively, avoiding ambiguity, and ensuring the output aligns with desired results. With average prompts comes average output, and no one wants those. To get exceptional results you have to communicate effectively with the model. Here are the most common ChatGPT prompting mistakes, laid out for you, so you don’t make them.
The most common ChatGPT prompting mistakes
1. Not being explicit
According to ChatGPT itself, users sometimes, “provide overly vague prompts, leading to generalized or non-targeted responses.” Don’t be vague, don’t be broad. If you’re not clear about what you want, you leave room for the model to interpret and guess. It’s not just vagueness that can lead to unsatisfactory or generalized responses. Colloquialisms can too. Avoid local idioms, terms that require explanation, or concepts with subtext.
ChatGPT works by making complex and highly intelligent predictions about which word to use next. Make the parameters upon which it makes this prediction abundantly clear.
Assume ChatGPT won’t interpret your sarcasm or read between the lines. Question every word you enter into the chat box, and use clear, universal language. Before you hit return, read it back. Try and see your prompt with fresh eyes and think critically about what you’re likely to get in return.
2. Giving complex multi-part questions
Within a single prompt, you should ask for one outcome. Multi-part questions have issues, including attention distribution, which is when the model distributes its “attention” unevenly, leading to uneven answers. Using multiple requests in one prompt leaves the model to prioritize questions over each other, based on a perception of relevance you might not agree with.
Multiple questions can introduce uncertainty about the overall context or theme. Even worse, if they’re not separated by lines, ChatGPT has to unjumble lengthy paragraphs and make calculations on which request to attack first. Think of prompting ChatGPT like briefing an enthusiastic intern. You wouldn’t unload a barrage of requests all at once; you’d give them clear instructions in priority order.
In general, one request at a time is preferable for more precise responses. After that, break down questions, add context and other information on separate lines, and rephrase any more confusing parts for clarity.
3. Not providing enough context
Whether you are human or robot, context matters for accurate and comprehensive responses. Just as you would provide context to any freelancer, team member or consultant before asking them for guidance, give ChatGPT enough information for it to calculate a decent response.
Don’t assume it will remember everything from previous interactions, because this can result in misunderstandings or irrelevant answers. Don’t make assumptions about the model’s knowledge. It’s not going to know about very recent events or hold specific details outside its training. Don’t assume it knows why you’re asking it something, that’s contextual information you can provide.
For best results, reference or recap prior questions so it knows which information to carry through. Remind it of key considerations within subsequent prompts. Be clear what information it should discard. Explain the purpose of your request and treat the model as a junior member of the team, not yet your MVP.
4. Unreasonable expectations
If you expect perfect answers every time, you’re in for a rough ride. Don’t make this basic mistake because it will undermine an otherwise fruitful relationship. Like any model, ChatGPT has limitations. Combining this mistake with even a hint of the others means you’ll have a disappointing and unfulfilling engagement.
Verify before trusting; don’t treat ChatGPT’s responses as definitive, especially in specialised areas like medical, legal or financial. Set realistic expectations and prepare to refine your prompting technique to get the best results.
Remember ChatGPT is not a human, so while it can simulate understanding and empathy, it doesn’t truly care, and it might not know exactly what you want. Not falling for the pitfall of mismanaged expectations means remembering all these potential flaws in the system, and working around them, whilst being quietly confident that it will continue to get better over time.
Final thoughts
Effective interactions with ChatGPT matter for the quality of output you receive. The better the output, the more you can do with less time and the more your work can reach new levels. Get good at prompting right now by knowing what not to do.
Avoid the common pitfalls of not being explicit, giving complex, jumbled questions, not providing context or setting the bar too high. Instead, be crystal clear with your prompts, lay requests out in a logical order, give ample context and be realistic about what might come out the other end. Continuous learning, building your familiarity and confidence in the model, will serve your business well going forward.
