How AI-Powered Descriptions Helped Our Client’s PPC Ads Break Through Google’s Shopping Feed

Note: Due to our contracts, the client in this case study must be anonymous.

Client Background

Our client is an animal health retailer and technology services company with veterinary clinics across the United States. They provide various animal health products, including pet medications, practice management solutions, and related services to support veterinarians and pet owners.

To drive incremental sales for their veterinary clinic partners, our client partnered with Greenlane to launch a large pilot project that involved Google Search, Display, Shopping, and Meta campaigns and spanned 24 veterinary clinic storefronts across the United States.  

The Challenge

The Google Shopping campaigns for this pilot project were limited in reach and visibility. The pet pharmacy space is highly competitive, with multiple sellers offering identical medications and products, varying price competitiveness, and more across prescription and non-prescription items. Our client’s ads struggled to break through the clutter, leading to low impression share, underwhelming click-through rates (CTR), and low sales. But why?

After analyzing performance data, our team discovered a significant issue—Google’s algorithm struggled to match our client’s ads with relevant searches due to the quality of their product feed. Incomplete and auto-populated, “boiler-plate” product descriptions on thousands of their product listings gave Google no additional context or keywords to match beyond the item description. Since Google Shopping relies heavily on product descriptions for keyword matching, our listings were deprioritized among competitors with richer, more detailed content in their product feeds.

Here’s an example of one of their “boiler-plate” product description before working with us:

^{Capsule size, color, shape, markings, or banding may vary from pictured image. Capsule – Quantities that exceed 180 days supply will not be processed and will create a delay in your order.}

However, our client’s product feed contains over 8,000 SKUs, and we identified over 4,500 product SKUs needing optimization. Manually optimizing descriptions would have been slow and expensive. The pilot project would have been over before the team optimized and implemented 4,500+ new product descriptions. Additionally, Google requires strict accuracy and compliance with pharmaceutical products.

The Solution

Instead of manual labor, we leveraged AI prompt engineering to revamp product descriptions at scale without sacrificing quality.

Here’s how we implemented this AI-driven strategy:

AI Prompt Engineering for Ads Optimization

To optimize thousands of product descriptions, we used an AI prompt engineering approach that combined fine-tuning, iterative testing, and batching to monitor the quality of the outputs.

We began by utilizing trusted pet-focused resources like Chewy, PetSmart, and other reputable sites to gather information on identical products. We instructed the AI to cross-reference these platforms, extracting key attributes such as product features, ingredient lists, and benefits. The team felt that including content around which health conditions and symptoms each medication treated would provide Google Shopping with the correct information to match products to user searches better. For example, all dog dewormer products should include the term “deworming” and a list of the types of worms (heartworm, tapeworm, roundworm, etc.) they treat. 

Once the AI-generated product descriptions were based on this data, we reviewed the outputs and identified both positive and negative description examples. The positive examples highlighted well-written, user-friendly descriptions that aligned with best practices, while the negative examples revealed vague or non-compliant language that needed improvement. We used these examples to fine-tune the model, reinforcing what worked and correcting areas where the AI strayed from the ideal description. Additionally, we clarified vocabulary to include and avoid, ensuring the tone was explicitly focused on pets rather than being too broad and potentially referring to humans.

Lastly, we included a carefully crafted disclaimer at the end of each product description to limit potential liability concerns and maintain transparency at scale. This statement clarified the product’s intended use for pets, which guided Google Shopping to match our products with pet queries rather than human queries for the same medications and to ensure that pet parents ultimately confer with their veterinarian for medical advice.

Our final step was to spot-check descriptions for accuracy and to ensure no hallucinations or incorrect medical information was included in our descriptions. Team members spot-checked one another’s work for accuracy, and we deemed the work complete when the group felt confident in the quality and correctness of the information provided by the AI tool. 

Example of an improved product description:

^{Alendronate (from Sodium) in Almond Oil Suspension SF is specially designed for dogs to promote bone health and manage conditions like osteoporosis. This formulation delivers 5mg of Alendronate per mL in a palatable almond oil base, making it easy to administer. The 30mL bottle provides an ample supply for your pet’s ongoing treatment needs. This product is intended for dogs only. Always consult your veterinarian before use to confirm it’s suitable for your pet’s specific condition.}

The Results

The new descriptions were packaged into a Supplemental Feed for Google Merchant Center to override the original feed inputs. The impact of AI-driven product description enhancements was immediate and measurable:

• +52% increase in impressions and +50% increase in Shopping clicks after descriptions began rolling out in late October vs. an equal time frame prior.
  • Google used more detailed descriptions to match products to appropriate user searches.
  • Clickthrough rates were maintained across time frames, indicating no reduction in user interest or intent from the larger available search pool.  
• Impression Share and Click Share also improved, +9% and +13%, respectively, indicating that campaigns qualified for more of the available Shopping searches and we secured a greater portion of available traffic. 
• The second wave of description updates reinforced these gains, pushing impressions and clicks to their highest levels at the start of November.

Conclusion

This AI workflow resulted in optimized product descriptions that followed best practices and improved ad performance. It maintained accuracy across thousands of listings, all in just a few weeks.