How to Write Better Prompts for AI Tools

How to Write Better Prompts for AI Tools
Photo: Openverse

The result of any AI tool can be improved by modifying the instructions given. A quality prompt is well-defined, has a clear role, specific task, specific rules and a clear end product. This will reduce the time wasted on chasing and copying back and forth, and will save you 20 minutes on each piece of work.


Exploring the individual nuances of prompt writing.The art of prompt writing.

There are only four elements to creating a great prompt that will steer the computer towards the correct answer. When you omit important elements, the system tries to deduce what you're looking for, which typically results in a lengthy and generic answer. Provide precise context otherwise there's a possibility of bad guessing and a lack of value in the output.


Each text prompt must follow these 4 components within them in order:


Role: Indicate to the tool what kind of user it should behave like; for example, the user may be a proof reader, senior manager, or a math tutor.

Task: Identify exactly what you want them to do (use an action verb, such as write, summarize, and extract).

Context:Give the background information, target audience and any source material required for the role.

Output: Want or need the answer presented in the manner as printed in five bulleted points, or a short paragraph?

How to Write Better Prompts for AI Tools
Photo from the web


Give Clear Role and Context Instructions
A setting resolves things and tells your software, "Now talk this way, with these words.A setting constrains things into a set of words and a point of view the software must use for your purpose. It speaks formally and with exactness if requested to play the part of a certified financial accountant. If you try to tell it to behave as a Middle Grade teacher it uses illustrative analogies and simple words.

However, context provides the boundaries that make the output relevant to the actual situation. Contextual information such as who the text is for, how long it takes to read and what voice it should have. For example: If you know your client is a busy person and reads nothing but 10 seconds, then tell them that it is for him/her. Read about what machines do with these text inputs – prompt engineering on Wikipedia.

Apply Rules and Treatments that are Severely Restricted.
As well as telling the tool what it's required to do, it's helpful to tell it what it shouldn't do. Negative constraints prevent the software from "filler-words" or generic introductions or additional comments that you have to cut out later. These limitations will save you 5-10 minutes of clean-up on each document.

These are some rules you can add to your regular messages verbatim:

Avoid introducing your text with phrases such as sure here is your text or in conclusion.
Limit entire word count to a maximum of 200 words.
Write and speak at an eighth grade reading level in simple terms, everyday language.
Make up and attribute new information that is not in the provided text.
Output should be presented as a bulleted list and not bold titles.
Show precise formats where applicable, using Examples where appropriate.
It is better to give one example within your prompt than to just write a long instruction. This way, you'll let the system know exactly what you mean as far as the style, length, and layout. This technique is referred to as "few shot prompting" and is really effective for tasks needing complex formatting of the data.


For instance, if you already have some examples of your previous work that you want to quote, copy/paste them above the main question you are addressing. Append it with EXAMPLE INPUT and EXAMPLE OUTPUT to let the software know how. Next, enter your new text below the label New Input. The software will generate the structure of your sample, but won't alter your original content. With an understanding of the generative AI systems, you will know the reason why the same input patterns yield the same output styles.

Optimize and Enhance Results
Take the first thing you've created as just a draft and not a finished product. If they do not get it right, it is NOT time to come up with a fresh response, or to open a new window. Instead, follow-up with an instruction which identifies the specific error and maintains the rest of the conversation.

Explain precisely what should be changed within the text, for instance, make paragraph two shorter, or write all formal jargon in plain language. You can also prompt the tool to compare itself and tell you what it didn't include in the original text. Typically, two rapid rounds of edits would result in the final draft that you need.

Comments

Popular posts from this blog

CVS to split off drugstore business in major breakup

Men's Health Outlines Goblet Overhead Press for Shoulder Width

Amazon Cloud Revenue Beats Most S&P 500 Companies