1. Introduction
Mastering the art of prompting is crucial for harnessing the full potential of Vurvey's AI tools. This guide will equip you with strategies to craft effective prompts, enabling you to extract valuable insights from our datasets and maximize the capabilities of our AI agents.
2. General Prompting Principles
Be Clear and Specific
- Start with a clear objective or question
- Provide relevant context and details
- Break complex tasks into smaller, manageable steps
Examples:
Bad: "Analyze sales."
Why it's bad: Too vague, lacks context and specific objectives.
Good: "Analyze Q2 2023 sales data for our top 5 products, focusing on regional performance."
Why it's good: Specifies time period, product range, and analysis focus.
Great: "Analyze Q2 2023 sales data for our top 5 products. Compare regional performance, identify year-over-year growth trends, and suggest factors contributing to any significant changes. Present findings in a bulleted list with supporting data visualizations."
Why it's great: Provides comprehensive context, specific analysis points, and desired output format.
Use Structured Formats
- Utilize numbered lists or bullet points for multi-part requests
- Specify desired output formats (e.g., table, list, paragraph)
Examples:
Bad: "Give me information about our customer satisfaction."
Why it's bad: Lacks structure and specific requirements.
Good: "Provide a summary of our customer satisfaction scores for Q2 2023. Include average scores and top complaints."
Why it's good: More specific and structured, but could be more comprehensive.
Great: "Analyze our Q2 2023 customer satisfaction data:
1. Calculate overall satisfaction score and compare it to Q1 2023 and Q2 2022.
2. Identify top 3 areas of improvement and top 3 areas of decline.
3. List most common customer complaints and their frequency.
4. Suggest 3 actionable strategies to improve satisfaction scores.
Present findings in a report format with bullet points and a summary table."
Why it's great: Highly structured, specific, and requests actionable insights with a clear output format.
Leverage Examples
- Provide sample outputs or formats when possible
- Use "few-shot" learning by giving multiple examples before your main request
Encourage Reasoning
- Ask the AI to "think step-by-step" or "explain its reasoning"
- Use phrases like "walk me through your approach" for complex analyses
3. Working with Vurvey Datasets
Specify the Dataset
- Always reference the dataset by its exact name
- Mention relevant time periods or segments if applicable
Example:
Bad: "Look at the customer data."
Why it's bad: Doesn't specify which dataset or what to look for.
Good: "Analyze the 'Global Customer Satisfaction 2023' dataset, focusing on APAC region responses in Q2."
Why it's good: Specifies the dataset, region, and time period.
Great: "Using the 'Global Customer Satisfaction 2023' dataset:
1. Compare APAC region responses in Q2 to Q1, and to other regions.
2. Identify trends in satisfaction scores across different product categories.
3. Correlate satisfaction scores with customer demographics (age, gender, income).
Provide insights on what's driving satisfaction in APAC and areas for improvement."
Why it's great: Clearly defines the dataset, provides specific analysis points, and asks for actionable insights.
Define Analysis Parameters
- Clearly state the variables or features you want to analyze
- Specify any filters or conditions to apply to the data
Request Specific Visualizations
- Specify the type of chart or graph you need
- Clearly state what data should be represented on each axis
4. Interacting with Vurvey Agents
Using the @ Symbol
- Use @AgentName to interact with specific agents (e.g., @DataAnalyst, @CustomerService)
- Engage multiple agents in a single conversation by using their @names
- Agents can read and respond to each other when prompted, but you must wait for one to respond before prompting the next.
Examples:
Bad: "Someone analyze sales and come up with marketing ideas."
Why it's bad: Doesn't specify which agents to use or define clear tasks for each.
Good: "@DataAnalyst, please analyze our Q2 sales data. (Wait for response, then after response, prompt the next agent) @MarketingSpecialist, can you suggest campaign ideas based on the analysis?"
Why it's good: Engages specific agents with clear tasks, but could provide more direction.
Great: "@DataAnalyst, please analyze our Q2 sales data, focusing on:
1. Top-performing products and regions
2. Year-over-year growth trends
3. Customer segments driving the most revenue
Why it's great: Orchestrates multiple prompts with specific, interconnected tasks, promoting in-depth, human-level detail and problem-solving.
Collaborative Problem-Solving
- Ask one agent to review or comment on another's work
- Use agents with complementary skills to tackle complex problems
Provide Contextual Information
- Give agents necessary background information
- Specify any constraints or guidelines the agent should follow
5. Advanced Techniques
Combine Agents and Datasets
- Leverage multiple agents or datasets for complex tasks
- Clearly define the role of each agent and the relevance of each dataset
Example:
Bad: "Use our data to come up with a new product."
Why it's bad: Vague, doesn't specify datasets or provide clear direction.
Good: "@ProductDevelopment, use the 'Consumer Trends 2023' dataset to suggest ideas for a new smart home device."
Why it's good: Specifies the agent, dataset, and general task, but could be more comprehensive.
Great: "@MarketResearcher, analyze the 'Consumer Trends 2023' and 'Smart Home Market 2023' datasets:
1. Identify top 5 emerging trends in smart home technology
2. Highlight unmet consumer needs in this space
3. Analyze competitor offerings and identify gaps
Present a comprehensive report synthesizing all inputs and recommendations."
Why it's great: Orchestrates multiple agents and datasets, defines clear tasks for each, and creates a comprehensive workflow for product development.
Implement Chain-of-Thought Prompting
- Guide the AI through a logical sequence of steps
- Ask for intermediate outputs or explanations between steps
Utilize Hypothetical Scenarios
- Present "what-if" situations to explore potential outcomes
- Use this technique for predictive analysis or strategy development
6. Troubleshooting and Optimization
Embrace Iterative Refinement
- If the initial response isn't satisfactory, rephrase or provide additional context
- Use phrases like "That's helpful, but can you focus more on X?" or "Could you elaborate on Y?"
Encourage AI Uncertainty Expression
- Prompt the AI to express uncertainty when appropriate
- Ask for confidence levels or potential limitations in the analysis
Leverage Feedback Mechanisms
- Use Vurvey's built-in feedback tools to rate responses
- Provide specific feedback on what was helpful or could be improved
7. Best Practices for Specific Tasks
For Data Analysis
- Start with broad questions, then narrow down based on initial insights
- Always ask for statistical significance or confidence intervals when relevant
For Content Creation
- Provide clear guidelines on tone, style, and target audience
- Ask for outlines before requesting full content pieces
For Predictive Modeling
- Clearly define the variables and time frame for predictions
- Ask the AI to explain its methodology and any assumptions made
Best practices for text clarity and brand identity:
Specify font styles: Mention "clear, legible text" or specific font types (e.g., "futuristic sans-serif font").
Use descriptive text: Instead of asking for a specific logo, describe its key elements (e.g., "sleek beverage can with 'NeoQuench' text").
Emphasize clarity: Use words like "crisp," "sharp," or "high-contrast" when referring to text or logos.
Provide context: Describe where and how the text/logo appears (e.g., "on a prominent billboard" or "on product packaging").
Use simple, made-up brand names: AI models often struggle with real brand names, so using simple, fictional names can yield better results.
Specify size and prominence: Use terms like "large," "prominent," or "focal point" to ensure text and logos are noticeable.
Request multiple variations: Generate several images and select the best one for text clarity and brand representation.
Remember, while these practices can improve results, AI image generators may still struggle with producing perfectly legible text or accurate logos. For brand-specific needs, consider using a specialized visual generator agent.
Remember, effective prompting is a skill that improves with practice. Experiment with different techniques, learn from each interaction, and don't hesitate to ask for clarification or refinement. The more you engage with Vurvey's AI tools, the more proficient you'll become at extracting valuable insights and solutions.
8. Tool Prompting Best Practices
Using Image Generation Tools
Vurvey offers several Image Generation Tools to generate images based on your prompts:
- Stable Diffusion (OpenArt)
- DALL-E (OpenAI)
- Imagen (Google)
To generate images, simply use the Tools drop down and select the image model you prefer and type your prompt in the input section:
This will generate several images based on the prompt for you to see. From there, you can fine tune using the Image Studio by clicking on one of the images, or you can rewrite the prompt and try again with changes you desire.
Using Social Media and Web Listening and Scraping Commands
Vurvey offers several Tools to do social listening and web search based on your prompts:
-TikTok: Will search TikTok for your query or hashtag and pull the 30 most relevant videos to your search and provide analysis and insights on them directly from TikTok, along with links to some of the videos as examples.
-Reddit: Will search Reddit for your query or hashtag and pull the 30 most relevant posts to your search and provide analysis and insights on them directly from Reddit, along with links to some of the threads as examples.
-LinkedIn: Will search Linkedin for your query or hashtag and pull the top relevant posts and articles to your search and provide analysis and insights on them directly from LinkedIn. LinkedIn often updates their terms and privacy settings so this tool does have outages from time to time.
-Youtube: Will search Youtube for your query or hashtag and pull the 30 most relevant videos to your search and provide analysis and insights on them directly from Youtube, along with links to some of the videos as examples.
-X/(formerly Twitter): Will search X for your query or hashtag and pull the 30 most relevant posts to your search and provide analysis and insights on them directly from X, along with links to some of the posts as examples.
-Instagram: Will search Instagram for your query or hashtag and pull the 30 most relevant posts to your search and provide analysis and insights on them directly from Instagram, along with links to some of the posts as examples.
- Web Search: Will search the web for an answer and not LLM training, giving you access to current information and or the ability to do research on anything on the internet today.
- Fast Scrape: Add a URL after this to pull all the data from that web page into your datasets (will land in “Scraped-Data” dataset and allow you to talk to this information using your agents.)
- Deep Scrape: Add a URL after this to pull all the data from that web page and all of the data on the pages found in the subsequent links found on that web page into your datasets (will land in “Scraped-Data” dataset and allow you to talk to this information using your agents.
Crafting Effective Image Prompts
Creating the perfect image prompt can be challenging. Vurvey's conversation feature can help you refine your prompts for better results. Here's how to use it effectively:
1. Start a conversation with Vurvey's AI assistant
2. Describe the image you want to create
3. Ask for help in crafting an optimal prompt
Example:
Human: I want to create an image of a futuristic cityscape with flying cars and holographic billboards. Can you help me write an effective prompt for image generation?
AI: Certainly! I'd be happy to help you craft an effective prompt for your futuristic cityscape image. Here's a detailed prompt you could use:
"A breathtaking futuristic cityscape at dusk. Sleek skyscrapers with curved, organic shapes reach towards a purple and orange sky. Flying cars with neon lights zoom between buildings. Massive holographic billboards display vibrant advertisements. The streets below are bustling with pedestrians and autonomous vehicles. Ethereal light beams crisscross the sky, adding to the sci-fi atmosphere. Hyper-realistic style with dramatic lighting and rich details."
This prompt provides specific details about the elements you want (flying cars, holographic billboards), the overall atmosphere (futuristic, dusk), and the desired style (hyper-realistic with dramatic lighting). Feel free to adjust any details to better match your vision.