How we built a FREE AI Chatbot for our Podcast Website

AI chatbots are powerful tools that can amplify the usefulness of any content-driven website. I believe that these tools should be used to enhance the abilities of team members, ultimately allowing them to provide better support to customers.

Using an AI chatbot for your podcast website can significantly enhance the overall listener experience. By providing an easy and interactive way for your audience to discover relevant episodes and access specific insights, the chatbot empowers listeners to get more value out of your content. Whether it’s summarizing key takeaways from episodes, helping users find specific guests or topics, or offering tailored advice based on previous discussions, an AI chatbot makes navigating and engaging with your podcast library more convenient and personalized.

But how do you build a free, low-code podcast chatbot? That’s what I sought out to discover.

I saw a lot of potential for an AI chatbot to enhance our WP Product Talk podcast website. I knew I could use a tool like Docsbot, but given that our podcast is entirely volunteer-driven generating no revenue at all, I wanted to avoid adding another monthly cost. Instead, I sought a virtually free, low-code way to create a content-driven AI chatbot to help you get more value from our ever-growing collection of product insights. If you’re curious about the project structure and associated costs, you can read more about it here.

Fortunately, I found a way, and it’s pretty straight-forward. You can already see the results by clicking on the chatbot icon in the bottom-right throughout our website. If you want to build a podcast chatbot like ours, then this article is for you.

Building the WP Product Talk Podcast Chatbot: A High-Level Overview

To build this chatbot, we needed just three key tools:

1. WP Import Export Lite plugin
2. OpenAI and its Assistant feature
3. OpenAssistantGPT (free plan)

This is the most budget-friendly, low-code approach I found to achieve what I wanted.

You might wonder, why not just use a WordPress plugin? Trust me, I looked into it. Despite the hundreds of contenders in the WordPress Plugin Directory, I couldn’t find a plugin that felt streamlined, modern, stable, and reliable enough for this purpose. I look forward to hearing your thoughts on that one!

That’s the overview. Now, let’s dive into the step-by-step guide for building this chatbot.

Step 1: Export Your Content in JSON Format

The first thing you need to do is export your podcast episode content in a format that the Assistant can best understand and process. Remember, we’re going for budget-friendly, so this is a manual process, not automated. Fortunately, the WP Import Export Lite plugin makes this pretty straight-forward.

Chatbots need well-structured content to provide meaningful, contextual responses. Typically, chatbots aggregate simple content and respond with search-like results. However, if you want your chatbot to associate specific pieces of information within a larger context, it’s best to prepare your content accordingly.

For instance, if you asked, “How can I improve my email marketing?” any LLM-based chatbot could easily provide a few paragraphs of advice based on our podcast episodes. However, if you asked for something more specific, like “Give me a summary of the episode Alex Denning appeared on, including the title, air date, link, and his top three tips,” then the chatbot needs to reliably link all the relevant details of Alex’s appearance.

See the formatted and detailed (and accurate!) response the chatbot gives you in the screenshot to the right. That’s where thoughtful formatting pays off. When done right, your chatbot can pull up a well-structured response that brings together all the meta-information for any episode or topic.

TIP: Including Transcript content for your Podcast Chatbot

Podcasts provide great content for AI to be informed by, but our chatbot is only going to be useful if we include the full transcript in the export.

One thing we do for every episode we create is we process the audio through Descript to generate far more accurate and helpful subtitles for Youtube, and an HTML transcript for the website. The challenge though, is that the Seriously Simple Podcasting plugin adds the transcript as post-meta, not as content. So if I just export the content of the episodes, the transcript is not included.

Fortunately, the WP Import Export Lite plugin has a really useful feature that allows you to include post-meta but output it with a PHP function. So I simply added the transcript_file to the export and used the readfile() PHP function to include the entire content of the HTML file in the episode export. You can see the preview of that export here:

Step 2: Create an OpenAI Assistant

Next, you’ll need to create an OpenAI Assistant. This is the core of the chatbot, where the magic of natural language understanding takes place. OpenAI’s Assistant is excellent at parsing nuanced queries and engaging in meaningful conversations, making it ideal for handling complex content like podcast episodes.

In a traditional RAG model, the assistant has to retrieve information from an external source every time a question is asked, which can be time-consuming and less accurate. By pre-loading the AI with content embedded into a vector-store, you can enable more contextual and precise answers without needing an ongoing retrieval process for every query.

For our purposes, the JSON export of our episodes is what we’ll be uploading to the vector store. Since I want the bot to be strongly informed of our guests, co-hosts, and blog posts as well, I’ll export each of those content types and add them to the vector store as well.

Fine-tuning your Podcast Chatbot Assistant

One of the advantages of using an Assistant to power your podcast chatbot is the features Assistants have over traditional LLM-based chatbots. I already mentioned how the vector store pre-loads the chatbot rather than having it query all the content every time. Additionally, Assistants can be “fine-tuned.” Fine-tuning involves training the model on your own data, making it more familiar with your unique content, tone, and subject matter. This leads to better, more accurate responses.

For instance, I needed to make sure our chatbot understood the finer nuances of the ever-important question of the popularity of various “Matt’s” in WordPress. After fine-tuning, our WP Product Talk Assistant is much more adept at understanding questions about specific podcast episodes and providing detailed summaries or key insights from them. The end result (like you see in the screenshot) is obviously way more factually relevant and a little tongue in cheek to boot 😉

Step 3: Use OpenAssistantGPT to Create the Bot and Embed It on Your Website

The final step involves creating the chatbot and embedding it into your website using OpenAssistantGPT. I’ll say that this was the most challenging part of this journey. Finding a low-code option was not hard, but you’d pay a lot for it. Finding a free option is not hard if you can code it all yourself. But finding a low-code and free option was very challenging.

In the end, OpenAssistantGPT is a great free option. It’s actually what made me look into OpenAI Assistants in the first place. It’s a free platform with minimal customization options, but it works well for our purposes—keeping everything simple and cost-effective.

With this free tool, you can choose to either use a “crawler” pointed at your website for the content, or integrate with your OpenAI Assistant. The crawler then is your standard LLM-based chatbot, but by integrating directly with OpenAI Assistants, you control all the parameters of your Assistant in OpenAI and then this platform is the shell that controls the chatbot and how its embedded into your website.

The biggest draw-back to this is they are incredibly stingy with their free options. The only thing you can customize with the free plan is the bot name, the “Welcome Message,” and the “Chatbot Error Message”. But in terms of fullfilling our needs for a low-code, free option to embed a podcast content powered AI chatbot on our website, this tool checked all the options.

Build your Own and Share your Results

So now you know two things:

1. How you can create your own low-code, low-cost AI-powered chatbot for your podcast website
2. That we have that same WPPT-content-drive AI chatbot here that you can learn a ton from

So if you’re reading this, and enjoy WP Product Talk, do us a favor and experiment with the AI chatbot. We’d love for you to take some screenshots and share the insights you gained from WP Product Talk. I’m certain we’ll all both learn a lot and most likely have a lot of laughs as well.