The rise of AI chatbots created the art of prompt engineering. Put simply, prompt engineering is the practice of optimizing your AI chatbot inputs, whether you are asking a question, doing deep research, or generating images and videos, among other things. No matter how advanced their underlying models, chatbots aren’t great at predicting exactly what you want from them, so a good prompt can absolutely be the difference between getting what you want and wasting time dealing with frustrating, irrelevant answers. I’m here to walk you through the process of writing great AI prompts, no matter what you’re trying to do.
Clarity, Context, Specificity: The Prompting Trifecta
Direct communication is the key to any good human relationship, but it’s equally important when prompting an AI. This means good prompts should be specific, incorporate relevant context, and use clear language above all else. Even simple requests that slightly miss the mark can lead to unhelpful responses. Similarly, small changes to prompts can make a big difference.
For example, if you ask an AI chatbot where you can get Chinese food from in your town, it will give you a list of restaurants. However, if you ask a chatbot what the best Chinese-American restaurant is (clearer) within five miles of your house (more specific) to get orange chicken from (more relevant context), you will get a better response.
Pay especially close attention to the responses you get as feedback for future ones. Even with the prompt above, which is clearer, has relevant context, and is more specific than the first version, an AI chatbot might still suggest restaurants that are too expensive. So, you can further refine your prompt to say you want only Chinese-American restaurants within 5 miles of your location that serve orange chicken for under $15.
Prompt engineering goes beyond just writing, too. You need to use the mode or model most appropriate for your prompt. For example, if you’re chatting with ChatGPT and you need it to solve a tough math problem, opt for its Thinking model rather than its Instant model. You should also attach any relevant files to prompts when possible, since an actual picture is more helpful than a description.
Don’t Trust Stale Data: Make AI Search the Web First
If your question requires up-to-date information, make sure the chatbot you are using searches the web. Chatbots usually search the web automatically, but you might need to toggle a web search feature or ask them to do so directly. If you don’t, you can get incorrect answers. For example, a chatbot might tell you the current pope is Pope Francis, rather than Pope Leo, if it doesn’t consult the internet.
(Credit: OpenAI/PCMag)
Another way to improve your prompts is by specifying the source or timeframe you want. For example, with a medical question, you might want information only from reputable medical organizations. Meanwhile, if you want to know the best weapon to use in a video game, you might care only about the answer in the context of the game’s latest update.
Lastly, don’t hesitate to use complex reasoning models when you ask questions. Take ChatGPT’s Thinking model, for example, which is designed to solve complex problems. When you engage it, ChatGPT spends more time thinking about your prompt and can likely then provide a more informed answer.
Deep Research Demands Extreme Specificity
Prompting is arguably most important for deep research because you can end up wasting tens of minutes if you use a bad one. Thankfully, many chatbots help you refine deep research prompts before they start researching. ChatGPT and Copilot ask follow-up questions, while Gemini presents you with a customizable research plan. Don’t ignore these.
Before you even get to follow-up questions or research plans, however, make sure to go into as much detail as you can and to a greater extent than for other types of prompts. For example, if you ask a chatbot what lotion to use on your hands, mention whether your hands are cracked and dry. But if you’re doing deep research on the same subject, it’s worth going the extra mile to include the average temperature of the spaces you spend your time in, as well. Deep research involves accessing hundreds of different sources, so it’s easy to taint reports with irrelevant findings when you aren’t as specific as possible.
Lastly, you can use deep research reports as the basis of further deep research. For example, say you generate the aforementioned report on hand lotions, but the recommendations are too thick and greasy. You can then attach the report and write a new prompt that instructs a chatbot to find lotions with characteristics similar to those suggested in the report but lighter. This way, even if a report isn’t quite what you wanted, you can still use it to your advantage.
Precision Beats Poetry in AI Art
We have dedicated guides on how to use AI to generate the best possible images and videos, but, briefly, the most important thing you can do is be descriptive. However, descriptive doesn’t mean flowery—it’s still essential that your prompts are clear and specific.
For example, prompting a chatbot for “a picture of a gorgeous tree with branches fanning outward like spiderwebs” is descriptive, but it’s not clear or specific. Instead, you are better off with a prompt like “a picture of a tall Japanese maple tree with many branches that go off into different directions, each of which has many bright red leaves.” This is descriptive but also precise.
Get Our Best Stories!
Your Daily Dose of Our Top Tech News
Sign up for our What’s New Now newsletter to receive the latest news, best new products, and expert advice from the editors of PCMag.
Sign up for our What’s New Now newsletter to receive the latest news, best new products, and expert advice from the editors of PCMag.
By clicking Sign Me Up, you confirm you are 16+ and agree to our Terms of Use and Privacy
Policy.
Thanks for signing up!
Your subscription has been confirmed. Keep an eye on your inbox!
Unlike with other types of prompts, it’s worth submitting prompts for images or videos multiple times. Each time an AI generates media, the results differ slightly. So, even if your prompt doesn’t shine at first, it might still do so after a few generations.
Include Clear and Relevant Documents or Images
Although some chatbots have impressive image recognition abilities, that doesn’t mean they’re flawless. As such, the image you send to a chatbot is just as important as the text of a prompt. For example, if you ask a chatbot to identify a part in a picture of a computer, make sure the image you send is free of glare and reflections. Furthermore, if your prompt is relevant only to one part of an image, cropping it can help.
The reflections on the glass side panel of this computer may cause problems in your prompts. (Credit: Ruben Circelli)
Similar advice applies to documents you provide a chatbot. Rather than upload a whole book, send just the sections relevant to your prompt. Also, whenever possible, upload either the document itself or its text rather than a picture of it. Chatbots are better at reading and understanding text than they are at extracting it from images, so it’s better to eliminate possible points of failure when you can.
Guide the Style or Get Generic Results
In creative writing tasks, chatbots tend to be repetitive, often writing about similar topics or in the same style. One way to combat this is to be specific and lay out the particular style, subject, and tone you want. I also recommend asking for creative writing in a new chat and including the instruction not to rely on memory or previous responses.
Recommended by Our Editors
If at all possible, provide examples. If you want a Shakespearean sonnet, don’t just say that; provide a particular sonnet and ask the chatbot to generate a new one in that style, alongside your other requirements. If you want something unlike other popular works, write up a few lines yourself as an example.
Match the Model to the Mission
First and foremost, you need to engage an appropriate model. However, this isn’t as simple as it sounds. For example, although ChatGPT’s Thinking model ideal is already ideal for complex reasoning, you can also turn on extended thinking when using it. So, take care before you send your prompt.
(Credit: OpenAI/PCMag)
If you’re using AI to solve a problem, you should ask it to show its work. This helps guard against incorrect answers because you can see how a chatbot came to its conclusion. Furthermore, if you need help studying, just getting the answer won’t do much.
Like with document analysis, it’s best to send complex reasoning prompts as text. For example, rather than snapping a picture of a textbook and asking for help with the fourth problem, write out the problem itself in your prompt. Chatbots can process images, as mentioned, but even the smallest mistake in complex reasoning can significantly skew the resulting response.
Prompting Is a Skill, Not a Shortcut
Even when you manage to dial in your prompt perfectly to get a true, high-quality response, it’s important to remember that prompt engineering is a skill you develop, not a checklist to follow. The above advice is a great starting point, but you still need to build up a feel for how chatbots react to your requests and adjust them accordingly. It’s a long-term learning experience that is highly likely to continue evolving.
Chandra Steele contributed to this story.
Disclosure: Ziff Davis, PCMag’s parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
About Our Expert
Ruben Circelli
Writer, Software
Experience
I’ve been writing about consumer technology and video games for over a decade at a variety of publications, including Destructoid, GamesRadar+, Lifewire, PCGamesN, Trusted Reviews, and What Hi-Fi?, among many others. At PCMag, I review AI and productivity software—everything from chatbots to to-do list apps. In my free time, I’m likely cooking something, playing a game, or tinkering with my computer.
Read Full Bio
