Writing for AI
Computers are learning how to talk to people. At Procore we want to ensure our automated conversations feel familiar to our users no matter how they are interacting with our product.
Personality for our Conversational Experiences
The personality of our conversational experiences will mirror the personality of our products. We want to ensure our product feels familiar to users, no matter where or how they are interacting with our tools.
Our conversational experiences will be helpful, relatable, and clear. We will always let our users know they are interacting with an AI assistant.
Our conversational experiences will be helpful, relatable, and clear. We will always let our users know they are interacting with an AI assistant.
Conversation Design Best Practices
Regularly evaluating and adjusting our communication with LLMs is crucial for enhancing the user experience. This is because the design of conversations and interactions is constantly evolving.
Remember your Goal
Keep your goal in mind at all times.
- Clearly state how you will help the user accomplish their goal
- Keep your user journey in mind and move the user to their end goal quickly as possible
- Avoid temptations to add “fluff,” extra details, or wordy phrases that distract from the action necessary to complete the conversation
Remember the User’s Why
The main reason a user opts to communicate with an automated assistant instead of a human is to reach their goal faster. Conversation for the sake of conversation should be avoided.
When designing digital experience, aim for brevity. Only include necessary dialogues for users and avoid unnecessary ones.
When designing digital experience, aim for brevity. Only include necessary dialogues for users and avoid unnecessary ones.
- Be brief with the information it provides
- Limited to what user needs to accomplish their goals
Be Clear and Direct
Clarity helps users easily understand what your trying to communicate with your conversational experience.
- Say what the user needs to do in clear, simple language. Direct them.
- Say what will happen next - but be sure to stay on the topic at hand. Don’t skip ahead to future steps or users will be confused.
- Say one thing at a time. Don’t ask for more than one response or answer to a question inside a single message.
- Tell users exactly what they can and can’t do.
- If you need to hand off a user to a live person, be sure to clearly communicate how the handoff works.
Sometimes, it's important to provide clear instructions to users who may be new to conversational AI and may need help on how to respond.
- Use Quick Replies or buttons to keep the user on-script.
- Use language like “tap the button below”, ”select from the following options” or a menu of options to select from with “here are things I can do”.
- Where possible, provide suggested responses, like the user's phone number, email, and first name. Ask them to confirm, instead of asking them to re-type their information.
- If their answer is invalid, let them know why this answer wasn’t accepted and how to reply back with one that will be.
Examples of specific prompts/phrasing to prevent errors:
- Select from the options below
- Tap yes to continue
- Watch this video for instructions on uploading your photo
- Here's the correct format for entering your phone number
Make it Easy to Understand
- If the user inputs a command freely, remind them when it’s time for them to type (the bot should say, “Reply with your answer”).
- For longer messages, be sure to increase the wait or pause time to give users enough time to read the whole message.
- Offer the option to resume the conversation or perform a new task after it has ended, such as through a "restart" or "do something else" feature.
Keep it Short
- No more than 3 lines of wrapping in any message box
- Messages from your conversational experiences should be about the length of a tweet (140 characters)
- Send no more than 1-3 messages before asking a user to respond
Make it Scannable
- Value-add first
- Numbers first
- Bullet points
- Short, clear sentences
- Limit user actions to three
- Use quick replies or buttons
- Avoid using hyperlinks in text, use buttons or a list of options instead
Avoid Open-Ended Conversations
As a general rule, open-ended questions should be avoided because it’s hard for users to know what kind of answer will be permitted. If you do ask an open-ended question, make sure it’s super simple: like “How old are you? or “What is your name?” (not. “How old do you feel?”)
UX writing is often front-load messages, putting the most important information first. But for conversational interactions the most important information needs to come at the end, so it's the last thing a user hears or reads.
A user should be able to complete any task by only remembering the last sentence they hear or read.
UX writing is often front-load messages, putting the most important information first. But for conversational interactions the most important information needs to come at the end, so it's the last thing a user hears or reads.
A user should be able to complete any task by only remembering the last sentence they hear or read.
Error Handling
To ensure a smooth conversational experience, it's important to handle the 4 types of errors that may arise. Make sure to create prompt responses for each type of error.
- No input - I hear nothing
- No match - I didn’t understand what you said.
- Misrecognition - I understood the wrong thing
- Task failure - I can’t resolve your issue even though I understand it
Have the Last Word
Questions and statements should always be final. Your conversational experience should confirm with the user the conversation is over, nothing more is expected of them.
When a user is asked to type or press a button, the conversational experience should always respond with a confirmation. Even at the end of the conversation, the AI should be last to respond, confirming to the user that the conversation is over and nothing is expected of them.
When a user is asked to type or press a button, the conversational experience should always respond with a confirmation. Even at the end of the conversation, the AI should be last to respond, confirming to the user that the conversation is over and nothing is expected of them.
Conversation Design Patterns
It is important for conversational experiences to maintain a consistent personality that adapts to different contexts.
This helps to create a sense of familiarity for the user when they request information or interact with us. By establishing a clear interaction pattern that users can easily understand and rely on, we can ensure that they stay on track and achieve their goals.
This helps to create a sense of familiarity for the user when they request information or interact with us. By establishing a clear interaction pattern that users can easily understand and rely on, we can ensure that they stay on track and achieve their goals.
- When the conversational experience cannot answer the user’s inquiry, we always… (pass them off to the appropriate service phone number or live support)
- Always follow up with the user after 90 seconds (or another time increment) have passed.
- Before proceeding with any action make sure to confirm with the user first. For example, we may say "Here’s how to set up your timesheet," "Let's upload your drawings," or "Let's find you that tool," depending on the user's needs.
- Only ask the user a maximum of five questions before giving them the thing they requested. (like: a product recommendation, the answer to their question, or appointment confirmation)