Optimization of Google Shopping Feeds with OpenAI’s ChatGPT: A Comprehensive Guide

Many online retailers face the problem that creating and updating product data is time-consuming and error-prone. Especially in niche markets or with rapidly changing assortments, it can be almost impossible for retailers to continuously generate high-quality data. However, insufficient or outdated product information often leads to poorer search results and a lower conversion rate.

The solution to this problem could lie in generative Artificial Intelligence (AI). OpenAI’s ChatGPT can be used to enrich existing product data with generated content or extract attributes from the description. By implementing an AI-supported feed creator, texts can be automatically generated and optimized, which improves data quality and significantly reduces the effort for data maintenance.

The use of ChatGPT to optimize Google Shopping feeds is an exciting example of how theory and practice can converge in the world of artificial intelligence. Therefore, we provide a deeper insight into the development of the tool here.

First, a feasibility test was conducted in Python to understand how well the existing product data is suited for automated processing. The reason for this decision was the existing OpenAI Python Library, which initially saves a lot of development work. This test helped to recognize the limits of the current data quality and identify the need for more precise data preparation.

After the feasibility test, we proceeded with prompt engineering. Here, we developed special input prompts that OpenAI’s ChatGPT could use to generate the desired product descriptions. This step required a deep understanding of the product features most relevant to customers, as well as SEO principles that maximize visibility in search engines. A particular hurdle was that the AI response had to have the correct JSON format. This problem has now been solved by OpenAI, as there is now the response_format parameter that clarifies this.

With the insights from the initial tests, further optimizations were made to the workflow. The goal was to increase efficiency, improve the reliability of the generated data, and bring the entire process to production readiness. This included technical adjustments to the code, the integration of additional data sources, and continuous monitoring of our system’s performance.

The final phase before full implementation involved extensive testing with various product categories and customer feedback, as well as performance monitoring. These tests were crucial to ensure that the GPT-Feedcreator not only works effectively in theory but also in real-world application. They helped to eliminate teething problems and make the application more user-friendly. The development of the GPT-Feedcreator is a good example of how innovative AI technologies can be practically applied to solve real problems in e-commerce. Through this process from theory to practice, we have created a solution that not only simplifies product data maintenance but also sustainably strengthens our customers’ market position.

The application of artificial intelligence to improve product data can be particularly impressively illustrated using a concrete example from an online shop for women’s fashion. In this case study, we focus on optimizing the data of a shirt from the brand “VIA APPIA DUE”.

The original product data was relatively simple and not fully geared towards search engine optimization. The product title was “VIA APPIA DUE Shirt – red multicolor – Size 46”. Although this information is basic, important details that could improve both search engine optimization and customer experience were missing.

Through the use of the GPT-Feedcreator, the product data was comprehensively revised. The product name was updated to “VIA APPIA DUE Women’s Round Neck 3/4 Sleeve Shirt Red Multicolor Size 46”, which is not only more precise but also more appealing to potential buyers. Additionally, the product description was expanded to highlight the shirt’s features and benefits, such as the comfortable fit and striking design.

The revised data not only emphasizes important selling points but is also designed to be more SEO-friendly. This leads to better visibility in search engine results and potentially a higher conversion rate. The following table shows a direct comparison of the original and optimized product data:

This case study demonstrates how targeted use of AI-supported technologies can significantly improve product data to enhance both user experience and technical discoverability. Such optimizations are essential for success in a highly competitive online market.

The GPT-Feedcreator is a Google Sheet equipped with Apps Script that handles communication with the OpenAI interface. This tool allows users to efficiently improve and expand their product information by leveraging the advanced capabilities of OpenAI’s GPT models.

1. API Key: First, you need an API key from OpenAI, which enables access to the AI platform.
2. OpenAI Assistant: Set up an OpenAI Assistant specifically configured for your requirements. This assistant can be fed with PDF documents describing your purchasing strategy. Creativity is key here – your own briefings, blog posts, and specific documents work well.
3. Configuration: Enter the Assistant ID in the configuration spreadsheet (“config” tab) of your Google Sheet.
4. Preparing the Shopping Feed: Prepare a CSV file with your product data, ideally using your best feed generator.
5. Setting Customer Abbreviation: Set an abbreviation for the customer to be used in the training data.
6. Adjusting the Briefing: Adjust the briefing as needed to meet the specific requirements of your feed.

After configuration, the workflow begins with importing the CSV file using the ‘Import CSV’ button. The product data is then displayed in the ‘Input Feed’. Use the ‘Improve Products’ button to start the AI-assisted optimization of product data. The improved data will appear in the sheet after a while. Mark products selected for the final feed as ‘Approved’. These are then transferred to the final ‘out’ feed using the ‘Process Approved’ function.

Continuous improvement of the generated product data is crucial to maximize the effectiveness of AI integration. Through regular reviews and adjustments of the AI model, the accuracy and relevance of the data can be steadily increased. This ensures that the product data always meets current SEO standards and is optimally optimized for Google Shopping. For particularly successful results, you can activate the ‘Training’ option to use this data for future improvements. Note that this process overwrites old training data. Automated processes can be scheduled via the script settings, so that individual functions of the script are executed at set times.

It’s important to emphasize that the GPT-Feedcreator is a powerful, but not error-free tool. Especially at the beginning, it is advisable to closely monitor the outputs before they are adopted into the production environment.

Given the advanced capabilities that the GPT-Feedcreator offers, one might wonder if traditional feed generators like Channable, FeedDynamix, and DataFeedWatch have now become obsolete. The answer is: Not yet. Although the GPT-Feedcreator automates the optimization and enrichment of product data, feed generators still play a crucial role in the overall process.

• Feed generators are specialized in creating and managing product feeds that are precisely tailored to the requirements of various distribution channels and platforms. They not only offer the ability to create feeds but also to efficiently manage and update them.
• GPT-Feedcreator on the other hand, is designed to optimize these feeds by improving product data based on AI-powered technologies.

The GPT-Feedcreator works seamlessly with feed generators like Channable to efficiently structure the data flow and ensure that only relevant product data is forwarded for optimization. This integration allows for the creation of customized feeds that are specifically tailored to the requirements of the GPT-Feedcreator. This way, even with large product catalogs, the most important products can be optimized first in a cost-effective manner.

How the Integration Works

1. Feed Creation: The feed generator (Channable, FeedDynamix, Datafeedwatch) first creates a specially adapted feed that is precisely aligned with the needs and capacities of the GPT-Feedcreator. This step is crucial to maximize the effectiveness of the entire optimization process.
2. Selection of Product Data: Only those product data that are eligible for optimization are included in the feed. This helps to save resources and make the process as efficient as possible.

Exclusion Criteria for Optimization

To determine which products should be included in the feed, specific exclusion criteria are applied. These criteria help to select products that can benefit the most from optimization. Here are some ideas on how to increase efficiency:

• Product Price: Products below a certain price level may be excluded as the effort for optimization could exceed the potential benefit.
• Sales Frequency: Products with low sales frequency might receive lower priority to concentrate resources on more frequently sold items.
• Stock Quantity: Products with low stock quantities could be excluded to focus on items with higher availability.
• Revenue: Items with low revenue contribution could also be exempted from optimization.
• Already Optimized Attributes: Products that already have optimized titles or descriptions may not need further processing and are therefore not included.

This targeted selection and pre-filtering of product data ensures that the GPT-Feedcreator focuses its capacities on the most promising candidates, thus generating maximum added value.

The GPT-Feedcreator has revealed impressive performance data in tests and practical applications, demonstrating its effectiveness and efficiency in optimizing Google Shopping Feeds. Here are the key data points summarized:

• Processing time: The Feedcreator takes an average of 33 seconds to improve individual product data. This is not lightning-fast but ultimately depends on the data depth and processing times of the model at OpenAI.
• Cost per product: Each product improvement costs approximately 0.16-0.25 euros, depending on the model used. It’s not free, but when compared to manual processing times, it’s still quite reasonable.

• Data processing: The tool is configurable for various CSV data and supports diverse content fields, ensuring broad applicability across different industries and product ranges.
• SEO optimization: In addition to improving product data, the GPT-Feedcreator can also be reconfigured without code adjustments, making it possible to optimize titles meta descriptions for SEO purposes.
• Training and customization: The Feedcreator allows learning from results and continuously improving the system. By marking high-quality data as “Training,” the quality of future outputs is continuously enhanced.
• Automation: Various steps of the process can be automated, minimizing operational effort and maximizing efficiency. The ability to schedule script functions offers additional convenience for users.
• Tip for better results: For product categories that vary greatly (e.g., clothing and kitchen items), it is recommended to first generate and train individual items from each product type. For significant content differences within a category, such as pants versus shirts, train an example item for each.
• Monitoring and quality assurance: Due to the high level of automation and the tool’s strong performance, users are advised to thoroughly check the outputs, especially in the initial phase, before they go into production.

The challenges of modern e-commerce, particularly in fast-paced and niche-specific market segments, demand advanced solutions for data management and optimization. The GPT-Feedcreator, supported by OpenAI’s ChatGPT, offers such an innovative solution that enables significantly improving the quality and efficiency of Google Shopping Feeds. By automating the process of data creation and optimization, online retailers can enhance their product presentations, increase visibility in search engines, and ultimately improve conversion rates. The integration of artificial intelligence into product data maintenance allows for more precise and appealing product descriptions that are tailored to the target audience’s needs. This not only leads to a better user experience but also to optimized resource utilization as manual interventions are minimized. The advanced features of the GPT-Feedcreator, including continuous improvement through training and adaptation, as well as the ability to automate specific processes, underscore this technology’s capacity to adapt to the ever-changing market requirements. Initial applications have shown that the Feedcreator delivers excellent results not just in theory but also in practice, increasing productivity while ensuring the quality of product data. In conclusion, the introduction of the GPT-Feedcreator in e-commerce represents a significant advancement that enables companies to successfully assert themselves in a competitive environment. Through the targeted use of AI-driven technologies, the landscape of online retail is being reshaped, leading to a more efficient, customer-oriented, and ultimately more profitable industry.