Best Practices for Prompt Engineering with Sphinx Mind

Learn how to craft effective prompts to get the most out of Sphinx Mind, your AI-powered marketing assistant. This guide will help you understand the nuances of interacting with Sphinx Mind to receive insightful marketing data and analytics.

‍

How Prompt Engineering Works with Sphinx Mind

‍

Sphinx Mind is designed to provide comprehensive reporting and real-time optimization. However, the quality of these insights is directly proportional to the quality of the prompts you use. A well-crafted prompt can yield in-depth analyses, while a poorly constructed one may result in vague or irrelevant information.

Understanding the mechanics of prompt engineering is crucial for maximizing the utility of Sphinx Mind. This section will delve into how Sphinx Mind interprets prompts and how you can tailor them for optimal results

‍

Rules of Thumb and Examples

‍

1- Be Specific

Being specific means asking for exact metrics or dimensions. This helps Sphinx Mind to fetch the precise data you're interested in, rather than providing a general overview that might not be as useful.

Less Effective :

❌ Show me the performance for our Google ads campaigns

‍

More Effective :

✅ Show me top converting campaigns from Google Ads last week

✅ Retrieve cost, clicks, and impressions broken down by campaign name for this week

✅ Fetch the average session duration for our website this month

‍

If you're uncertain about the metrics or dimensions in your active data source, simply type '@' in the chat box to begin your search. Navigate through the options using your keyboard's up and down arrows, and confirm your preferred selection by pressing 'Enter'. Alternatively, you can also make a selection using your mouse.

2- Specify Date Range

Use date presets like "last week," "this month," "last year," or specific date ranges like "July 2023" to get more targeted data.

Less Effective :

❌ How much did we spend on Facebook Ads?

‍

More Effective:

✅ How much did we spend on Facebook Ads last week?

✅ Show me the total clicks in July 2023

✅ Get me sessions by date from 2023-06-18 until 2023-06-25‍

3- Request Format of Data

Requesting data in a specific format, such as tables or charts, allows you to visualize the information more effectively. This can be particularly useful when you're sharing the data with team members or stakeholders who prefer visual representations.

Less Effective :

❌ Visualize performance by device category for last week

‍

More Effective :

✅ Show me a pie chart of transactions by device category for last week

✅ Fetch 'sessions,' 'sessionsPerUser,' and 'date' from GA4. Then create one chart where sessions are represented as bars and sessionsPerUser is represented as a line

✅ Generate a table of campaign name, impressions, clicks for last month

‍

4- Relax your Filters

Be flexible with filters unless you have a specific need. For instance, if you're requesting the bot to retrieve performance for your 'Summer' campaign, it's more effective to ask for campaigns that 'contain summer' rather than campaigns that 'equal summer', unless your actual campaign name is solely 'summer'.

Less Effective :

❌ How much did we spend on Summer campaign this month?

‍

More Effective :

✅ How much did we spend on campaigns contain "summer" this month?

‍

5- Avoid Excessive Granularity

Requesting data with too much granularity can lead to two issues: First, it can exceed Sphinx Mind's token limitations, causing incomplete or truncated responses. Second, it can overwhelm you with too much information, making it difficult to derive actionable insights. Sphinx Mind is designed to summarize complex data efficiently, so take advantage of that feature.

Less Effective :

❌ Break down the performance of our Google Ads campaign by age, gender, location, device, and time of day for the last six months

‍

More Effective :

✅ Provide a device overview of our Facebook Ads performance, focusing on CTR and conversion rate for this month

‍

6- Mention Source Name for Multiple Data Sources

If you're pulling data from multiple platforms like Google Ads, Facebook Ads, or Google Analytics, specifying the source in your prompt ensures that Sphinx Mind fetches data from the correct platform.

Less Effective :

❌ Fetch me spend per campaign name

‍

More Effective :

✅ Fetch me spend per campaign name from Facebook Ads

✅ Retrieve the bounce rate for our website from Google Analytics

‍

7- Use Metrics and Dimensions

Metrics are quantitative measurements like "clicks" or "impressions," while dimensions are attributes that can be used to segment metrics, like "age group" or "device type." Using these in your prompts can help you get more nuanced insights.

Less Effective :

❌ I want to see some numbers

‍

More Effective :

✅ Show me the conversion rate segmented by age group

✅ Display the average session duration broken down by device type

✅ Fetch the click-through rate divided by ad placements

‍

8- Use Double Quotes for Precision

Utilizing double quotes in your queries can add a layer of precision, ensuring that Sphinx Mind focuses on the exact term or phrase you're interested in. This is particularly useful for filtering data or when you're dealing with terms that could be ambiguous.

Less Effective :

❌ Show me spend by campaign name with summer

‍

More Effective :

✅ Show me spend by campaign name contains "summer" for the last 14 days

‍

9- Add Steps for Long Prompts

For complex queries that involve multiple tasks or stages, breaking down the prompt into numbered steps can make it easier for Sphinx Mind to process and execute your request. This is particularly useful for comprehensive reports or multi-faceted analyses.

Less Effective :

❌ Create a comprehensive weekly report using Google Analytics with tables, charts, and insights for various metrics and dimensions.

‍

More Effective :

✅ Create a comprehensive weekly report using Google Analytics. Here are your tasks for the weekly report, detailing the data from "last week":

1. General Overview Table: Create a table with the following fields only: 'sessions,' 'activeUsers,' 'screenPageViews,' and 'eventCount.' The date range is last week compared to the previous week. Provide your insights below the table.
2. Combo Chart Representation: Fetch 'sessions,' 'sessionsPerUser,' and 'date' from GA4. Then create one chart where sessions are represented as bars and sessionsPerUser is represented as a line. Provide insights below the graph.
3. Pie Chart Representation: Build a pie chart with 'sessions' as the metric and 'deviceCategory' as the dimension. Analyze the data and include your insights below the visualization.

‍

‍

10- Give Examples for Complex Prompts

When your query involves multiple steps or a specific format, providing an example can help Sphinx Mind understand your expectations better. This is particularly useful when you're asking for a detailed report or analysis that involves multiple metrics and dimensions.

Less Effective :

❌ Create a monthly report for branded queires from Google Search Console data

‍

More Effective :

✅ Create a comprehensive monthly report using Google Search Console. Here are your tasks for the monthly report, detailing the data from "last month". 

Output: Blended Table for "Brand" and "Non-Brand" 

1. Strictly retrieve the following fields: "impressions", "clicks", "position", "ctr" with the filter "query" containing "optimization". Label these as "Brand" in a "Type" column. 
2. Retrieve the same fields like above but with the filter "query" doesn't contain "optimization".  Label these as "Brand" in a "Type" column.  
3. Combine the results from both "Brand" and "Non-Brand" into a single table using the "Type" column.
4. Provide a summary analysis, highlighting key differences between brand and non-brand queries. Remember results should be in one table.

‍

In addition to these best practices, we also offer a Prompt Library filled with a variety of examples to inspire you. Feel free to explore it to get more ideas on how to interact effectively with Sphinx Mind.