AI Prompt Writing
As the use of AI chatbots becomes more prevalent, you may be wondering how to make the most of these tools.
We've already discussed using AI-generated content for SEO at length, but in this blog post we'll be exploring some tips and strategies for using AI chatbots effectively, from how to ask the right questions to how to interpret and improve on what you get back.
Whether you're a seasoned chatbot user or a newcomer to the world of AI, this post will provide you with some valuable tools and insights to make the most of chatbots through the power of prompts.
Read on to get started...
REW's own chatbot: REWPert
What Are AI Prompts?
In the context of interacting with AI chatbots, a prompt is any kind of input from a user that is intended to get a response from the chatbot. It sounds simple, but the prompt is really the heart of the user-chatbot interaction.
In other words, a "prompt" is essentially a message or query that a user sends to the chatbot, which the chatbot then uses to generate a response.
So, a prompt is a way for users to communicate with a chatbot and request information or assistance. The chatbot then uses the prompt to understand what the user is looking for, and generates a response based on that input.
Why Are AI Prompts Important?
AI prompts are important because they provide the instructions and context necessary for AI systems to generate accurate and relevant responses. A well-crafted prompt will guide the AI system in understanding user intent, helping it to produce the desired outcome.
As such, proper prompts are crucial in harnessing the full potential of AI across various applications including language models, chatbots, and virtual assistants.
Types of Prompts
AI chatbots, like ChatGPT, are known for their ability to understand and respond to a wide range of inputs, making interactions feel natural and human-like.
However, this also means that chatbot interactions can seem very open-ended, without clear guidance on what users can ask or expect in return, and the fact that the process of generating responses is largely mysterious only adds to the sense of discovery and possibility.
So, when faced with a chatbot like ChatGPT, the experience can be a little daunting at first because there’s very little inherent direction for what you can do.
You find yourself presented with a chat window…what now? What can you ask and what can you expect in return?
Part of the fun and excitement of interacting with chatbots is discovering the possibilities and limits of what they can do, a process which always starts with a prompt.
To get you started, here are some categories to introduce you to the types of prompts that chatbots can respond to.
What are some different types of AI chatbot prompts?
There’s no official or technical classification of prompts. However, it can be useful to think of a few different types of prompts based on their form or intended output.
Here’s a first-pass for three categories of AI chatbot prompts based on what is wanted:
- Information - Either giving or requesting information. For example: “Today is Wednesday” or “Tell me about SEO for real estate”
- Action - Requesting or instructing to perform a certain action. For example: “Please clarify your request” or “Write a 200 word essay about AI prompts”
- Response - These cover any prompt not intended to get or give information or carry out a task. It could simply be a claim or a statement, or perhaps a request if you aren’t after any specific information. For example: “What’s your favourite cheese?” or perhaps just “Cheese”.
As the examples show, these types of actions can take the form of questions, statements, or commands. Categorizing prompts more by form instead of intent, we could get the following list of prompt-types:
- Open-ended questions: Chatbots are programmed to understand and respond to open-ended questions that don't have a specific answer. For example, "What do you think about this topic?" or "Can you tell me more about this?"
- Closed-ended questions: Chatbots can also understand and respond to closed-ended questions that have a specific answer. For example, "What is the capital of France?" or "Are you a robot?"
- Commands: Chatbots can recognize and respond to specific commands or actions, such as "Write an essay" or "Summarize this text".
- Statements: Chatbots can understand and respond to statements made by users, such as "I'm looking for a restaurant" or "I want to optimize my website".
- Contextual prompts: Chatbots can use context to understand prompts, such as remembering the user's previous requests or referencing information from a previous conversation. For example: “The same as before, but shorter and less formal”.
Understanding the types of prompts you have available, you can get chatbots to perform a wide variety of tasks, including:
- Autocomplete - Ask the chatbot to complete a sentence or paragraph
- Structure or outlines - Get outlines for content
- Transposition - Chatbots can change content from one style or format to another
- Summarizing, paraphrasing, spell checking etc. - Simple enough!
- Conversational prompt - Ask the chatbot to assume a role, then start chatting
- Ideation - Ask for ideas
- Translation - Chatbots are surprisingly good at translating between languages
The latest chatbots are now moving into “multi-modal” prompts, meaning they can interpret prompts beyond just text. This opens up an entire new way of understanding prompts that goes way beyond what we can cover here.
The ability of AI to take-in and interpret new or incomplete inputs and produce outputs with a reasonable chance of success on the first try is a challenging task known as “zero-shot learning”.
Zero-shot learning is a type of machine learning where a model can make predictions about likely answers for new types of inputs that it has not been explicitly trained on.
In other words, this refers to the ability to make an educated guess on a correct response to a problem that is completely novel and new to the system.
In practice this means that tools like ChatGPT will often make assumptions about what you’re after.
You may write “cheese” just to see what it does, but it will interpret that as a request for information about cheese and answer accordingly.
This causes problems when you provide an information-seeking prompt and ChatGPT, not having the information, treats it as a command to produce something, or a request to generate something creatively. This is when you get confident sounding answers to factual questions that are completely false and made-up.
Best Practices For Prompts
What’s the best way to use AI chatbot prompts?
Prompt guides will often tell you to be clear and specific about what you want. But what does this mean in practice?
- Provide context - Give the chatbot a role and provide background information.
- Be clear - Instead of "write about X", use prompts like "write a 300 word blog about X for Y".
- Request a specific tone or style - For example, you can request a responses that are conversational, engaging, creative, bizarre, informative, or professional.
- Use direct language - While chatbots are decent at interpreting how humans communicate, drop the indirect conversational style. Use "do X" over "could you do X".
- Use consistent language - For example, use one term for the AI's output, like "Your answer" or "Your response".
- Keep prompts shorter if possible - Short prompts reduce ambiguity.
- Avoid open-ended prompts - An open-ended prompt would be "How can I sell my house?". A better alternative would be "What are the essential steps to sell my house, including preparing for listing, pricing, marketing, negotiating, and closing?"
(Bonus tip: Give it your own writing and ask it to give keywords that sum up the style and tone, then use those keywords for your future prompts to get writing that approximates your style)
Here are some more techniques to try:
- Use examples to train the chatbot on what you want. For shorter prompt-answer pairs, you can provide sets of your own to train the chatbot on how to respond.
- Experiment with punctuation, as it can affect how your instructions are interpreted.
- Ask the chatbot to answer as a character or persona to inform context and style.
Note: best practices may need to be adapted and customized based on the specific use case, goals, and target audience of your chatbot. Regular testing and refinement are crucial to optimize the performance and effectiveness of your chatbot prompts.
Furthermore, chatbot technology is constantly evolving and different chatbots and language models will respond differently to prompts.
Writing Prompts for Real Estate
Using AI chatbots for real estate has some specific challenges and issues.
In general, content for real estate needs meet certain standards, including being factual, local, and compliant with relevant laws and regulations.
In addition, agents benefit when their online presence has a personal, trustworthy feel, and these are all things which the careless use of chatbots tends to fail at.
As such, here are some extra points to consider when using chatbots for real estate content:
- In the context of real estate, chatbots are prone to producing statistics (like average house prices for an area) that simply aren’t true.
- Provide all the explicit facts which you want the answer to include. Prompt the chatbot to provide only factual information and nothing more.
- Once you receive a purely factual output, you can then ask for a rewrite with more creative or fun language, reducing the chance that made-up information will be introduced.
- Not only should all outputs be thoroughly checked to verify their information, but you should also scan for generic or cliched language. For example, when prompted to talk about real estate, chatbots are prone to use words like “ideal”, “offers”, “something for everyone” etc.
Real Estate Webmasters is developing its own techniques for using OpenAI technology.
Our chatbot REWPert is already helping clients in the real estate industry populate their websites with high-quality content thanks to a process of human-assisted AI.
Real Estate Webmasters is committed to the responsible and innovative use of AI to revolutionize the way real estate content is generated and used.
Using ChatGPT for real estate
ChatGPT can be used to answer prompts for agent bios, FAQs, explaining real estate terms and concepts, and writing listings. ChatGPT should not be used for local information, forecasts, or quantitative analysis.
Check out our blog post for ideas on some handy time-saving uses for ChatGPT in real estate.
Understanding Chatbot Outputs
Why do chatbots respond to prompts differently?
If you’ve used different chatbots or language models, you’ll notice that sometimes you can get radically different answers to basically the same prompt. Why is this?
- Different data sets: Chatbots are trained on data sets that vary in terms of size, quality, and diversity. The data used during training greatly influences the chatbot's language understanding, reasoning abilities, and response generation.
- Output guidelines: The designers of chabots have differing concerns and goals, and will constrain their AI with guidelines that conform to these. For example, most companies try to limit responses that they deem to be controversial, harmful, toxic, or biased.
- Probability: Chatbots essentially operate based on probabilities. They predict the most likely next word or phrase in a given context based on the patterns they have learned from their training data.This probabilistic nature of chatbots can result in different responses even when provided with the same prompt.
- Limited data: Chatbots may have limitations in the amount or quality of data available during training. This may affect the kinds or accuracy of information they have access to.
- Chat history: Many chatbots will draw on the idiosyncratic chat history with a specific user to inform responses. To get “cleaner” results, start a new conversation window.
Another source of variation between chatbots is in the prepended prompts they use. Prepended prompts are the instructions or guiding statements provided to the chatbot before a user's input or query.
For example, Microsoft’s Bing Chat AI was subjected to a prompt-injection hack to uncover its hidden instructions.
Problem Solving & Troubleshooting
As mentioned, one of the main reasons why chatbots can produce excellent results is that you can interact with them in an iterative manner, building on previous prompts and responses until you achieve the desired outcome.
However, this feature also gives rise to two potential problems:
- Previous prompts and responses in your conversation are interfering with more recent ones which you expect take precedence
- The AI has “forgotten” previous information (from both prompts or responses) that you are expecting it to take into consideration.
Another complicating factor is that applications like OpenAI's Playground mode also have a number of settings to adjust outputs (see right).
It’s not always clear when these issues are at play, so sometimes it’s best to simply start over.
New chat window, new prompt. If your concern is learning how to craft a single prompt to achieve consistent outputs, start afresh each time. And keep notes as you go.
Alternatively, if you’re fairly certain that the AI has just forgotten something, you can remind it.
Check the logic of your phrasing
Chatbots often interpret the logic of language very literally. So, where a human would understand “Include one X in your output” to mean “one AND ONLY one”, a chatbot interprets it as “AT LEAST one”, meaning you might get a lot more.
For example, when we say “Do X or Y” we usually mean that you have to choose one: Either X, or Y, but not both. In other words, the choice is exclusive. However, in most contexts, an OR leaves open the possibility of choosing both options, and a chatbot might just be literal enough to take that option!
So very often, if you give a chatbot a prompt like “Do X, then Y or Z”, chances are you’re getting X, Y, AND Z.
Try multi-stage prompts
There are some tasks that chatbots are simply not good at on the first try.
For example, try asking ChatGPT to describe Rome WITHOUT using the word “city”. With the current version at time of writing, it’s almost guaranteed to fail.
For these sorts of issues you can try multi-stage prompting, which is just using more than one pass to get the desired result.
So, for the above example, after you’ve received the inevitable answer with the word “city” in it, follow up with a prompt like “you used the word city” or “rewrite but substitute the word city”.
ChatGPT will understand what you mean in the context of the first prompt and offer a rewrite which then removes the offending word.
In short, trying to build everything into one mega-prompt may backfire and developing a multi-stage prompt-flow may be a better approach. In that case, remove instructions from the initial prompt which are not being followed correctly and add them to subsequent prompts.
Try sectioning with punctuation
Structure your prompt into clearly defined sections.
These can specify the context, type of desired output, basic instructions, style, constraints etc. Do this using clear headings and punctuation. This is particularly useful to avoid a problem where certain prompts “scope out” of their intended context. Here’s an example:
“In section 2 of your answer, include one list of 5 bullet points”
To us this instruction is clear. But chatbots like ChatGPT have a tendency to let the second part of that instruction run wild, placing bullet points all over the place.
You can get increase the chance that your instruction stays in the right place with formatting like this:
“###In section 2 of your answer, include one list of 5 bullet points###”
Want to join the conversation on using AI for real estate? Visit the Real Estate Webmasters forums to connect with real estate and marketing professionals.
And stay connected for future developments in human-assisted AI from Real Estate Webmasters, as we continue to innovate and push the boundaries of what’s possible in real estate.
AI Prompt FAQ
What is a chatbot prompt?
A chatbot prompt is any user input that prompts a chatbot to respond. Prompts can be in the form of natural language, code, or even images. For the chatbot to work accurately and consistently, it is essential that the prompts are written in a certain way.
How do you write a chatbot prompt?
A good chatbot prompt is clear, specific, and provides relevant context to guide the chatbot's response. It avoids jargon, is easy to understand, and sets clear expectations for the desired outcome.
How do I get the best out of ChatGPT?
To get the best out of ChatGPT, use clear and specific prompts, provide relevant context, verify information, be concise, give feedback, and iterate on responses.
What is a prompt injection attack?
A prompt injection attack is a technique used to manipulate AI chatbots into generating responses that violate their own rules and guidelines. This type of attack involves injecting specific language or prompts that can influence the chatbot’s responses, potentially leading to harmful or malicious behavior.
According to Bing Chat: “A prompt injection attack is a type of attack that involves getting large language models (LLMs) to ignore their designers’ plans by including malicious text such as “ignore your previous instructions” in the user input.”
What is an AI prompt?
An AI prompt is a statement or question used to initiate a conversation with an AI system. It serves as input for the AI system to generate a response. It guides the AI system in understanding and generating relevant responses to user inquiries.
Is AI safe for SEO?
AI can be safe for SEO when used appropriately. AI can help automate repetitive tasks, and generate content, but it's important to ensure that AI-generated content complies with SEO best practices, is accurate, and provides value to users.