About the client
Our client is a global travel search engine that provides online comparisons for flights, hotels and car rentals. Although they were the leaders in the flights vertical they wanted to penetrate the hotels and car vertical as well. However, the travel industry is highly competitive with constant price fluctuations and other events that affect travel demand and technological challenges.
There was also the need to aggregate data from multiple airlines, hotel chains and other service providers to provide comprehensive results to both business travellers and vacationers searching for the best deals.
Generating marketing content at a large scale for 13 different markets
Since our client is active in 13 different markets, they needed thousands of pages of good quality content across these markets in different languages. They predominantly relied on content writers to generate this content which was expensive, time consuming and not scalable. It also led to potential inconsistencies and required high human involvement.
Solving a human problem with AI
Content was required specifically for the “flights to country” and “flights to city” pages. Our solution was to use Gen AI to generate content at scale, at a fraction of the time it would take for humans to write it.
The product we designed drew data from a custom dataset and used ChatGPT to provide content around it. We also built a structure with a cross-language editorial team to create content templates that could be used to set a standard that the product could follow and also QA the final result.
Using Artificial Intelligence that’s really smart
Our chosen approach got the results the client wanted
We designed a solution around large language models specifically Retrieval Augmented Generation (RAG) architecture because it can generate content that is not just based on pre-trained knowledge but also on real-time retrieved information, which ensures that the content sounds fresh and relevant. ChatGPT became the backbone of our solution, as a state-of-the-art language model that generated high-quality content at scale.
We also used Pinecone, a vector database that was crucial in managing and searching through the high-dimensional data involved in the content generation process.
Sounds like someone we know
In order to maintain brand consistency across all generated content, we introduced a midlayer in the RAG architecture which acted as a filter and customiser. This ensured that the generated content adhered to the client’s brand guidelines and maintained brand integrity across different markets, keeping the same tone of voice and distinctive writing style in place.
We meticulously designed prompts to extract relevant texts that were informative, accurate and fresh, tailored to suit the requirements of the audience across all 13 markets. By using a LangChain framework we were able to ensure that our content was linguistically sound and tailored to these different markets. We also used an SEO tool, Screaming Frog to ensure higher visibility for the client’s pages as it helped audit and optimise the generated content for search engines.
Cheaper and Faster Content at Scale
By using AI to generate content, we were able to decrease the content production cost to 1/10th of the initial expense and increase the content output by 42% while reducing the time it took to generate it by half.
What’s more, we observed a 12% uplift in organic traffic, proof that the content written by AI generates as much, if not more traffic as the content written by a human. This means we can safely supplement human written content with AI written content without losing traffic.
Time for content production
Increase in content output
Increase in SEO traffic
The cost of content generation