Put 'Google' on your Website with AI Search
Everybody knows they can use OpenAI to create an AI chatbot, but did you know you can use them to create your own search engine? The dirty little secret is that it provides better search results than Google. In addition it will only show results from your website. It's kind of like having your own Google search engine, but only for your website.
Most people probably think about transformers and the chat/completions API when they're asked about OpenAI. But there's another API, that allows you to take a natural language query and search for whatever matches that query. This API is called the "embeddings API", and you can feed it text such as for instance; "Coffee mug with image of Frank Zappa", which would return results such as for instance ...
- Frank Zappa coffee mug
- Ozzy Osbourne coffee mug
- Mozart coffee mug
- Coffee mug with your own profile picture
- Etc ...
The above are just examples, but illustrates how AI can match natural language queries towards individual pages on your website.
The point is that the AI embeddings API can leverage "intent" during the search process. Instead of relying upon exact keyword matching, it asks itself how the information in its database "semantically matches" the query. This implies a tractor will be matched for a search for "strong vehicle" and a motorcycle will be matched for "racing vehicle", etc. This is why it would match Mozart and Ozzy to Frank Zappa, simply because the search algorithm understands they're both musicians, and is therefor associating your query with these musicians, returning Ozzy first because he's a contemporary artist just like Frank Zappa was. The quality of the search results becomes 1,000 times better compared to keyword search, because it's capable of finding "relevant pages" based upon "the intent" of the user. To understand the process, imagine a librarian being asked for a copy of the Book of Mormons, and answering the following.
I am sorry, but we don't have the Book of Mormons, but we do have the Bible, the Torah and the Quran. Are you interested in any of these instead?
For a user looking for the Book of Mormons, the above would be the "closest match", even though these are completely different books, with completely different titles. Basically, this provides you with a search and user experience that dwarfs anything else out there, such as Elastic search, etc ...
Navigation for E-Commerce
If you're doing drop shipping or e-commerce, and you've got a lot of products, you already know how difficult it can be to build navigation that allows your users to easily find the product they're looking for. If you've got 100 products in 15 different categories, then your only real option is to create a hierarchical navigation menu. This was the way Yahoo used to work 30 years ago and it was a mess. Sometimes you had to wade through half a dozen clicks, and very often you'd simply give up and go somewhere else.
Then came Google and allowed users to search using "natural language", which combined with their PageRank algorithm created an experience that was simply superior to Yahoo. Today Google is worth 100 times more than Yahoo, and some of my readers might not even know what Yahoo even was, while I betcha all my users have heard of Google.
By leveraging the embeddings API, prompt engineering the AI, and adding a RAG database referencing your pages on your site - We can add a "personal Google user experience" to your website. And in fact, some of our clients are already using their "AI chatbot" like this. Below is a screenshot from one of our clients.
The above is a Shopify website - But we can install this on any E-Commerce system or website. And in fact, for the above query the AI didn't find any exact matches, but realised that unisex wrist watch was the closest match in its database. For a customer looking for a Diesel wrist watch, the above would probably be a viable alternative, resulting in a sale of a different article because of the AI's ability to understand the user's intent.
AI Search Case Study
Between Sessions have a library of more than 2,000 PDF files. These PDF files are tools for psychologists and therapists. Dr. Lawrence Shapiro came to us because his users had trouble finding relevant PDFs, and he wanted to leverage AI to make it easier for his therapist clients to find tools they could leverage in their sessions with their clients. It was absolutely crucial for Dr. Lawrence that the AI chatbot didn't give lengthy responses, but rather worked as a "search engine" for his library of tools.
We imported all these PDF files, created embeddings for them, and prompt engineered the AI chatbot to instead of returning lengthy answers, it would just return a summary of each individual PDF and the URL where you could download it.
Dr. Lawrence was so happy with the result he sent us this email ...
Wrapping Up
AI search can be a valuable addition to your site, especially if your users are struggling to navigate it. Club de Mode from the first screenshot above for instance, increased their sales by 338% when they installed their AINIRO chatbot. Such figures are almost "fairy tale figures" from an e-commerce perspective, but they're 100% real. If you'd like to talk to us to see how we can help you improve your navigation using AI, you can contact us below.
