Thread by @cyberandy

I worked on my first experiment with ChatGPT Plugins applied to SEO! I created a search engine for an eCommerce site using WordLift's Product Knowledge Graph (PKG) as a starting point. Here is what I learned:

#SEO #ChatGPT #WordLift #eCommerce

🧵 1/6

1. You can use @LangChainAI to experiment with the idea (here is the documentation 📖: python.langchain.com/en/latest/modu…) and choose any of the supported LLM. #Opensource is amazing 🤩.

2. You only need two files: a JSON manifest and the OpenAPI specs.

2/6

I built the search engine using @JinaAI_ by indexing schema:Product in the PKG WordLift created by accessing the Google Merchant APIs.

🔎 nowrun.jina.ai/?host=grpcs://… (you can play with the search engine here)

3. You need all your prompting skill to create the JSON manifest.

3/6

1️⃣ …orest-0da2ee70f.2.azurestaticapps.net/ai-plugin.json here is mine;* description_for_model* is where you instruct the model to use the API (it is not trivial, and POST API is still not fully supported by LangChain)

2⃣ …orest-0da2ee70f.2.azurestaticapps.net/openapi.json this is the OpenAPI, always be wise and very concise.

4/6

4. You can achieve similar objectives by using @LangChainAI Agents 🤖 (ref: python.langchain.com/en/latest/modu…).

Here below, you can see that I was able to instruct the model to make the query to the search end-point 🎉.

Long story short: by design, all foundational models hallucinate and suffer from a data cut-off. By using plugins in ChatGPT (or agents in @LangChainAI with any LLM!), we can guide their completions and let them access up-to-date information.

Happy SEO Automation!

END 🧵