Since Open AI’s ChatGPT was released last November, it’s arguably become the most talked about bot since the T-800. But before we face the inevitable future in which we must choose to either embrace our robot overlords or battle sentient toasters, we have an opportunity to teach this real-life HAL 9000 to serve the good of humankind—or at least to get some decent copy out of it.
What Is ChatGPT?
An artificial intelligence tool developed by OpenAI, ChatGPT accepts prompts from users and provides responses derived from the scope and diversity of text its programmers supplied as training data, most of which dates prior to 2021. It can follow the thread of a conversation, building on previous interactions and refining each response.
The research preview of ChatGPT is free to use, as it is technically still in beta testing. As such, there are some limitations in terms of the length and number of questions you can ask, as well as in the accuracy of the data it provides. Funmi Looi Somoye, of PCGuide, writes that “while there is no definite word limit, many users have reported the bot breaking down when over 500 to 1000 words were submitted” (https://www.pcguide.com/apps/chat-gpt-word-limit/). Additionally, according to OpenAI, the bot—as of the update to GPT-4—can only generate approximately eight thousand words of a conversation in one model and twenty-two thousand words in another model. After a certain length of text beyond this point, the text generated by the bot starts to suffer in its ability to create effective responses.
For writers who can leave their technophobia to entertainment, the chatbot’s ability to generate mostly plausible passages of text may solve a broad range of problems. But whether you choose to ask it to find words that rhyme with “orange” or to help you with A/B testing by iterating headlines and marketing content, the prompts you feed this language-generation tool can be hacked to promote stronger responses.
Here are ten tips to keep in mind before you have your first conversation with the robot.
1) Use Precise Language
You can use a conversational tone, but try to be concise to keep the chatbot focused. Although ChatGPT can check for grammatical errors in short passages, too many mistakes in your prompt will lead to miscommunication, just as they would with a human reader.
You want to keep the prompt short enough to avoid confusion, but provide enough detail to direct the bot accurately. A targeted request, like “Write a brief email introducing my company to a potential sales lead,” will return more successful results than broad requests, like “Write me a sales email.”
2) Take Advantage of the Chat
No matter how great your first prompt is, you can “teach” the bot how to provide better responses for you through tiered questioning. ChatGPT was designed to have conversations, so if you “engage in some back and forth,” writes Forbes Senior Contributor TJ McCue, you can fine tune the answers you receive.
For example, Tom Demers, of Search Engine Land, suggests that writers seeking to use ChatGPT to improve their SEO might start by asking “What are some examples of topics that popular websites in the youth basketball niche talk about?” Then follow up by asking for clarification on specific responses, or ask it to build on that information by providing you with keywords related to that topic.
3) Ask the Program for Help
If you’re stuck on how to use the program, start by asking the bot what it can do. Ajay Yadav, of Simplified.com, suggests the prompt “Can you list out the copywriting frameworks that you can write in?” From there, you can choose the structures that best suit your purpose and follow up with specific prompts for those frameworks. Or you can use the list it provides to brainstorm your own copy.
Pro Tip: If you’re interested in using AI art, but struggle to create successful prompts for programs like Midjourney, choose a simple image description, like “astronaut walking on Mars,” and ask ChatGPT to expand the sentence into something more unique and descriptive, suggests Baris Sen of TextCortex AI.
From the Creators
4) Manage Your Expectations
Even the AI’s creators recognize ChatGPT isn’t qualified to take your job. According to a Tweet by OpenAI CEO Sam Altman, “it’s a mistake to be relying on it for anything important right now.” Although ChatGPT has amazing capabilities for language generation, it’s important to remember that it’s a new tool, and its limitations are still being tested. In the same thread, Altman cautions that ChatGPT’s tone and technically correct phrasing may mislead users into believing it knows more than it does. “The danger is that it is confident and wrong a significant fraction of the time,” he tweeted.
Be careful what you ask for, and keep in mind that the OpenAI tech that powers Bing’s upgraded search engine has been “plagued with problems like factual errors and unhinged conversations,” according to a February article by CNET. Consider content output from conversations with ChatGPT drafts at best, and always check them for accuracy and bias before publishing or sharing.
Pro Tip: If you need to reset the conversation, ask ChatGPT to “ignore all previous prompts in this conversation,” writes Sarah Tamsin on her technical blog, SarahTamsin.com.
5) Don’t Underestimate the Skill Involved in Writing a Good Prompt
To get the most out of the tool, remember that it is a machine, which uses the text you input to inform its processes. “Writing a really great prompt for a chatbot persona is an amazingly high-leverage skill and an early example of programming in a little bit of natural language,” tweeted Altman. Despite the conversational experience of using ChatGPT, the responses are not coming from another human who understands nuance and context or makes allowances for ambiguous statements. The chatbot is going to be much more literal in its interpretation of prompts.
6) Be Prepared for Change
On February 1, OpenAI announced ChatGPT Plus, a subscription plan that will be available for $20 per month and promises subscribers faster response times and improved access during peak times (https://openai.com/blog/chatgpt-plus). For now, the company plans to continue to offer free access to the original recipe, but since they view ChatGPT as “a research preview” launched to “learn more about the system’s strengths and weaknesses,” it may not persist in its current form long term.
From Experienced Writers to New Users
7) Get Out of Beginner Mode with Prompt Engineering
In early January, self-published author Rob Lennon went viral when he tweeted ten advanced techniques for using ChatGPT. His prompts focus the chatbot’s responses by giving it a topic, audience, goal, and style of writing. Longer, more complex prompts can lead to more accurate starting responses, which users can then improve further by continuing the conversation.
On March 1, he tweeted he had completed a first draft of a full book using a sequence of ChatGPT prompts. He admitted that “this book is not great. It could really use some story, some heart, and a punchier writing style,” but he writes he considers that a part of the process. Ultimately, he hopes to use AI-generated content to complete a fast draft in a day that could then be revised for style and readability over the course of a few weeks, instead of the months or years that sometimes accompany drafting from scratch.
Pro Tip: Take prompt engineering a step further and give ChatGPT a role to play. By telling it to respond as an expert in a certain field, or even in the style of another program, you may get a more sophisticated response, according to GitHub (https://github.com/f/awesome-chatgpt-prompts).
8) Use a Tested Formula
Tamsin, a digital content creator, experimented with OpenAI’s Beta API before the ChatGPT interface made AI more accessible. Like Lennon, Tamsin has noticed that longer prompts generally lead to better responses, but rather than put everything into a single prompt, she divides her requests into separate entries that build on each other. She details the technique on her blog, but the five-step formula is fairly straightforward: “context, task, instruction, clarify, refine,” specifically in that order.
9) Consider the Ethical Implications
Paranormal mystery author Leanne Leeds uses ChatGPT to generate titles and plots for her mystery series. But she has concerns about the potential for using it to copy other authors’ style, she told The Verge. “That, for me, is an ethical line. I may like Jim Butcher … but I’m not going to take my stories and have them rewritten in his voice to rip him off. You could, if you were ethically okay with that, with this technology and what it allows you to do.”
Given that AI programs can access expansive training databases from which to generate their responses, definitions of permission and plagiarism will have to evolve. Vanessa Arnold, of Neuroflash (https://neuroflash.com), discusses the complexity of copyright and the use of ChatGPT in an article from February 21, “ChatGPT Copyright: Everything you need to know.” Until legal consequences are in place, it will be up to authors to identify the lines they won’t cross.
10) Get More Help
AI chatbots like ChatGPT are constantly evolving, and the program’s uses are only limited by your creativity. Continue experimenting and exploring the program with the tips and resources below.
- GripRoom (https://griproom.com) provides a ten-step guide to better prompts and follows up with twelve ideas for types of content ChatGPT can create.
- Zapier (https://zapier.com) offers six tips for writing effective prompts, and provides examples of “basic” and “better” prompts.
- Marcus Ramsey, founder of the AI newsletter The Brink, provides an extensive list of the best ChatGPT resources.