Refining Prompts for better AI generated Content

Introduction to Refining AI-Generated SEO Content

In a previous article, we explored how Kemb AI tools can be used to generate optimized titles, descriptions and meta descriptions in bulk. The tests were done with sample data from Amazon Fashion Products 2020.

Knowing that the tools are designed with general prompts for SEO description generation, we decided to refine these prompts to align more closely with a specific brand style and a target audience in mind. 

Dataset Overview: Structuring Inputs for AI

To generate optimized content, we only require a dataset including some essential input-columns such as: 

• Product URL

• Name

• Image URL

• Product Details

• Categories

• Meta Keywords

Important: Most product catalogs certainly contain information like this, at least in part. Of course, you don’t have to have all this data available, but the more input you already have, the more “guidance” you can offer the AI to create the new desired text. 

We limited the dataset to a manageable size, focusing on a selected set of rows to showcase the results. The tools, which are freely accessible, can generate an overview for up to 100 rows in a single session (in the free version), making them a practical solution for batch processing.

Once the dataset is selected, our tools offer various options for the analysis, on the basis of which the AI generates the output. For instance:

• SEO Text Generation: This feature is particularly effective for text-based columns, such as Product URL and Product Name, creating optimized titles and descriptions.

• Image Analysis: Leveraging the Image URL column, this feature generates better-tailored titles and descriptions based on visual data.

After the results are calculated, users can conveniently download the output as a CSV file, allowing for seamless integration into workflows or further refinement.

Refining Prompts for Improved Results

Initial Test: Using Default Prompts

We did an initial run of the tool without altering the prompts and the results obtained can be seen in the column Original Tool (which is probably already sufficient for most people if they are looking for a quickly scalable option). 

For more specific requirements though, , we randomly created an audience and brand style to use on the app definition afterwards. This additional information was integrated into the app’s general instructions for subsequent runs. The following scenarios outline the changes made:

Scenario 1: Adding Audience Descriptions

The results can be seen below in the results table (“Adding Audience” column)

Audience Profile:

• Adults (likely 18–45), with a focus on those interested in apparel and traditional styles. 

• Individuals who value aesthetic consistency, blending clothing styles with lifestyle choices.

• Individuals who appreciate fashion as an expression of cultural identity and tradition and have an interest in eco-friendly, ethical, or regionally sourced products.

• Customers that seek high-quality, well-branded clothing, leaning towards Indian and modern fusion fashion.

The AI generated descriptions emphasizing cultural and lifestyle elements, such as:

• “Highlight its cultural roots.”

• “Designed for the modern woman who cherishes tradition.”

Scenario 2: Incorporating  brand tone and style 

To align descriptions with our fictional brand, the following tone and style were introduced (see “Adding tone” column below):

Style description:

• To consider the brand to be  Imaginative and Evocative ( the messages should  transport customers into the world of the products); 

• Confident yet Warm: speaking with authority about the quality, but maintaining a welcoming charm;  

• Playful with Depth- Infuse the brand with a sense of curiosity and subtle humor, showing customers that it is approachable.

Scenario 3: Adding specific power words to the descriptions

Scenario 3, focused on incorporating power words, was implemented to test how emotionally engaging and persuasive language could enhance the appeal of the generated content (see “Adding Power Words” column):

Ahead of the game, Amazing, Amplify, At the top, Attractive, Awe-inspiring, Beautiful, Defiance, Drop-dead, Effective, Elegant, Elite, Enchant, Epic, Fearless, Fortunate, Foxy, Genius, Good-looking, Gorgeous, Guts, Handsome, Hero, Jaw-dropping, Jubilant, Kick ass, Knockout Legendary, Lucky, Magic, Mind-blowing, Moneymaking, Notable, Noteworthy, Optimal, Prosperous, Quick-witted, Remarkable, Sassy, Saucy, Ultimate, Unbeaten, Undefeated, Valor, Vanquish, Victory, Wealthy, Wonderful, Wondrous, You, etc.

The Results

Exemplary AI Tool Output (I):

Exemplary AI Tool Output (II):

Refining input positively impacting outpot quality

Some descriptions appeared consistently regardless of the adjustments, focusing on the product’s intrinsic details. For instance:

• “Featuring a vibrant red color adorned with intricately detailed white embroidery.”

• “This stunning red embroidered kurta.”

• “Stunning yellow chiffon saree.”

With the audience definition added, some changes appeared, such as:

• Highlighting cultural and lifestyle elements: “Highlight its cultural roots.” and “Making it perfect for festive occasions or casual outings.”

• Addressing specific demographics: “Designed for the modern woman who cherishes tradition.” and “Individual who appreciates high-quality, ethically crafted apparel.”

These outputs reflected themes like blending clothing styles with lifestyle choices and appealing to those interested in traditional styles.

Adding the tone and brand style stood out in statements that told a story or showed confidence:

• Storytelling elements: “Enhances your natural beauty but also adds a touch of sophistication to any occasion.”, “Tell a story of craftsmanship and elegance.”, “Creating a beautiful contrast that captures attention effortlessly.”

• Confidence in quality:  “At the top of its game in contemporary style.”, “Wear it with confidence and let your style shine.” 

The inclusion of power words added flair, but in some cases, the frequency felt overused. However, when applied thoughtfully, they had better emotional appeal. 

Advantages of AI-Generated SEO Content

The standout feature of wearekemb tools is their ability to perform these changes at scale, allowing users to quickly iterate and fine-tune descriptions to achieve the perfect fit for their brand and audience. This efficiency offers a substantial advantage compared to manual text generation, enabling rapid changes and optimization.

Conclusion: Prompt Refinement as an Art

The AI Tools offer a powerful solution for generating optimized SEO content. However, refining prompts to resonate with your audience and reflect your brand style is key to unlocking their full potential. Small tweaks in audience descriptions, tone, and word choice can transform ordinary outputs into compelling, brand-specific content.

Whether you’re managing e-commerce descriptions or crafting meta tags, leveraging AI for bulk generation—combined with thoughtful customization—can be a game-changer in your SEO strategy. Start testing our free tools today and do not hesitate to not hesitate to contact us at any time if you have comments or ideas about the tool or would like a completely individual solution. Our experts will be happy to help you design the ideal solution for your application!