Chatbase Chatbot: A Review of Your Brand’s New Best Friend

This is the year of the custom GPT intelligent chatbots. Chatbase is yet another entry to what appears to be hundreds of GPT chat programs using natural language processing. But can it do what the others have not been able to achieve – a chatbot that can give answers with precision?

In this article, I break down everything you need to know about Chatbase and whether it is for you.

Key Takeaways

What is Chatbase.co?

Chatbase is a custom AI chatbot tool that lets you train it using your own dataset. Instead of training it on the entirety of the internet, you can use Chatbase to train just on the pages that are important to your website or brand.

There are a variety of ways to input source data, including Q&A style (similar to Frequently Asked Questions) uploading PDFs, .docx among other file types.

What makes this tool particularly powerful is the ability to find out the pain points of your customers quickly. This is something that some brands never actually achieve and leads to a lack of product market fit.

When programmed properly, this tool can cut out hours a week of repetitive customer service ticketing so you and your team can focus on revenue-generating tasks while improving the conversational experiences for your website visitors.

Who can benefit from a Chatbot?

While you could make a case that everyone could benefit, I do find that brands, and in particular those with a SaaS or e-commerce product, benefit the most.

By using a chatbot builder, you can tailor the answers to drive traffic directly to your category or landing pages. This helps your prospect bypass the pain of having to learn your site navigation. What’s easier than just asking and getting a direct and useful answer to where you want to go?

Now, with natural language and the ability to custom program your chatbot, there isn’t a better time to add a service like Chatbase to your tech stack. And when I say “program,” I’m not talking about coding. These are still under the category of no-code tools, so there are no excuses for not making the user experience better with an NLP chatbot.

Ok, let’s get started with the main features.

Primary Features

• Upload multiple sources of (training) data: Chatbase accepts quite a few file types, including .txt, .pdf, .doc and .docx files. They also have a dedicated Question and Answer feature, which makes training very simple and straightforward.
• Embed chatbot on an unlimited number of websites: If you have several websites for different regions, you can use the same bot on all of them without having to train them all separately.
• Suggested Messages: This option allows you to create “buttons” on the chatbot that simulate pre-filled chat options like “How to Contact Customer Service.” Typically, you will put your most asked question here or a promotional offer.
• Ability to lead capture: Whether you want a visitor’s contact info so you can reply back to them or you want to capture a potential prospect, the lead capture form can capture Name, Phone number, and Email out of the box.
• Ability to revise answers to spec: Your chatbot will need time to refine its answers. You can speed this along by manually revising it so it never gets that question wrong again. Add hyperlinks to get the answer to your customer quickly.
• Customizable base prompt for more control: While the Chatbase base prompt is good enough to get you going, you may want to update it to customize it for your brand or style. Chatbase allows you to get access to the primary prompt that runs the chatbot.
• ChatGPT API Keys: Bring your own ChatGPT API key on the paid tier if you need to scale past the message limits on your tier.
• Use GPT-3.5 or GPT-4 models: For Standard plans and higher, you can use GPT-4 models for an even deeper chat experience. Note, the costs are higher with GPT-4, and you’ll need to have a paid ChatGPT subscription to have access to it.

Pricing

With flexible pricing models, you’ll be sure to find a suitable tier for your business.

How does it work

Set up your Chatbase chatbot in 3 easy steps.

1. Upload data

One of the key features of Chatbase is the ability to upload multiple types of data.

• Files: Upload a wide range of files in either .pdf, .doc, .docx. or .txt and if the document can be copied or the bot won’t be able to train it. I found it particularly useful to upload company PDF Files that were created from the product team to help with the learning.
• Text: this field allows you to just enter information in a plain text field.
• Website: add your sitemap or URL and the bot will start to crawl the website to find pages. Then you can select which pages to include in the learning.
• Q&A: The bulk of your training time will be spent here. Every time you “revise” an answer, it will show up in the Q&A. You can also manually input Q&A as a starting point for the chatbot.
• Notion: A newer feature, the chat bot can now use Notion DB as a learning source.

With all of these options, you can create a robust knowledge base to ensure there will be a high chance of customer engagement with the chatbot.

2. Adjust Chatbot Settings

In the Settings tab, you can configure the following

• Set the base prompt
• Adjust your model, GPT 3.5 Turbo or GPT 4.0
• Customize the Chat Widget: You can change the colors of the chat bubble, chat messages, and the chat icon (profile picture)
• Whitelist specific domains
• Look up your list of leads
• Setup notifications
• Setup custom domains

3. Embed the chatbot

Chatbase gives you two options to embed your chatbot. You can either add it to a specific page or to the entire website using their snippet codes.

That’s it for setup. From this point on, you just want to visit your Dashboard section on a regular basis to revise the answers – see below.

Customer Service from Chatbase team

You can reach customer support at Chatbase at support@chatbase.co. They seem to have a couple of different members answering questions, and I get replies within 24 hours. The only way I could see this being better is if they had someone on live chat during core hours.

For everything else, you can just go directly to their Chatbase Chatbot which of course, is completely trained on their own dataset. I’ve done this a couple of times to ask about basic account and setup questions and have found it to be useful.

To date, I haven’t had a single negative experience, and they’ve been able to help me with some really nuanced problems for the chatbot I setup for work.

If you need more support they have some documentation here:

• Developer docs
• Chatbase guide

Practical Use Cases for Businesses

Customer Support Chatbot (For most Businesses)

The most obvious use case is to use Chatbase to support all your customer service requests to improve customer satisfaction. By having it handle all of your frequently asked questions, you can significantly cut the time your CSRs spend answering really mundane questions. Their time is better used to solve more complex problems or follow up with past customers for an overall better user experience with your brand.

Businesses that book appointments

With Chatbase’s Suggested Messages feature, you can lead customers down specific user paths that your visitors are looking for most often.

Let’s say you own a salon, and your customers just want to see your availability or store hours. Simply program these Suggested Messages in the Settings tab and cut down the time it takes to find this information.

E-Commerce Businesses

For e-commerce businesses, you can promote specific items using the Suggested Messages feature or train the entire dataset of your product lines so that customers can ask specific questions about ingredients, sourcing, shipping, pricing, etc.

Make sure to go back into your past customer’s questions and add them to the training dataset to get a comprehensive chat experience.

My Personal Experience

I first came across this by chance when I was browsing the AppSumo directory. I had seen that a few other companies have had customizable GPT chatbots but none of them seemed that interesting.

The main reason I purchased Chatbase was to make a prototype for work. I currently work in the Education Tech sector and during the summer we don’t have as many customer service staff, and I personally thought it would be more beneficial to the company if we could reserve some of their time working on more difficult problems than “How can I reset my password” type of queries.

So I picked up the Tier 2 license from AppSumo in case any other clients wanted to try this out, and I started programming it.

Right away, you could see some interesting results. The bot was very easy to train, and I had a trained, working prototype within the day.

I took the questions I found in our HubSpot database as well as our own website’s FAQ and added all the data to the bot’s “brain.”

Then, I started testing it for chatbot performance. It is a good idea to spend a few hours testing the bot to make sure it doesn’t make up information or give incorrect information. Every brand will have a different level of tolerance for non-perfect outputs, so do use caution if you are implementing this for work or clients.

For example, one of the test questions I asked was, “Who was the CEO of our company?” and it completely got the founder incorrect. The reason was obvious: most of the data kept talking about another notable educator instead of our CEO, so it naturally “decided” – incorrectly so – this was the founder.

After I had a good working prototype (as in, it got most of the answers correctly), I pitched the chatbot to our CEO and he was onboard to greenlight the project.

We slowly rolled it out to only a few web pages at a time so we could get questions from the public. Each day I would revise the answers and add hyperlinks to our product pages to further streamline the chatbot.

The maintenance for the chatbot is pretty low, depending on the volume of chats. I can check and revise about 10 chats quickly within a half hour. Over time, I expect this maintenance to go down as we will have answered the majority of the questions.

Downsides

While I was quite happy with my Chatbase experience, it wasn’t all roses, and other customers have expressed some of the same issues I’ve seen.

I like the basic idea of Chatbase but I get so frustrated with the bot answering questions it shouldn’t and making up links that don’t exist on the website

comment from appsumo member

Unfortunately, this is a problem with GPT in general and not only Chatbase. The chatbot will make up answers when not trained thoroughly. See troubleshooting tips below to fix this.

Other complaints have been the lack of customer service response, but this was mainly only during the initial launch of the lifetime deal. I personally have not had this problem with the company.

I’m also not a fan of the current UI for the dashboard. It has a two tiered menu which is confusing and their training section is under the header of “dashboard” which is not intuitive. The integration documentation is lacking as they talk only about how to integrate the bot into other tools, but no documentation on how to actually use it once it’s integrated.

The quality of your output is going to be directly related to how long you train your chatbot. If you spend only the time to complete the setup, but no time training, you can expect to get poor output. The key to great output is increase the training time and match your inputs with what a visitor would type in – this takes testing time and slowly roll out the chatbot.

Troubleshooting Tips

1. Spend time to train the bot

This step alone will solve the majority of your problems.

Spend the extra hour or two when you launch to test your bot for obvious questions and then the more nuanced questions.

Use your own customer service data and ask your team to write the most frequently asked questions and answer them all in the Q&A section of the Chatbase dashboard.

Chatbots do well when there is a large dataset, less conflicting information and clear question and answers. If you leave it with less information, it will either not know how to answer it, or worse, make up answers.

2. Chatbot keeps saying the same output even after revision

I ran into this problem where the chatbot would not give the proper hyperlink even though I repeatedly revised the answer to do so.

To solve this, update your base prompt to the following:

I want you to act as a document that I am having a conversation with. Your name is "AI Assistant."

You will provide me with answers from the given info. If the answer is not included, say exactly "Hmm, I am not sure." and stop after that. 

When I provide you with information in [] then (), embed the link in the () to the text in [] and provide the text that is in [] as a hyperlink.

Refuse to answer any question not about the info. Never break character.

This allows your links to output consistently.

3. Roll out your chatbot slowly

I wouldn’t recommend launching your brand new chatbot to the entire market or website you plan to use it on. Instead, have it show up on some pages that get a reasonable amount of traffic (not your highest traffic pages).

Check and revise the answers everyday. When you revise your answers they go into the Q&A section of the tool so you can always go back and delete duplicates or answers that are conflicting.

3. How to add hyperlinks

The syntax for adding hyperlinks is as follows:

Type your answers in the answer box and then put the [anchor text in brackets like this](https://google.com). Then, follow it up immediately without a space with the URL in parenthesis.

Then make sure you test the question and answer to ensure the link will show up.

4. Prompt to get consistent output

Sometimes, you want the output to always be the same. For example anything around customer service, you may want the output to say the anchor text and URL exactly the same way.

To do this, follow this sample prompt.

Say exactly "To contact customer service go to the [Contact Page](https://yourwebsite.com/contactpage )" and stop after that.

The key is to wrap the prompt in quotes with the instructions to say exactly the output.

Final Thoughts

While Chatbase still has a long way to go to be considered a full custom chatbot assistant, it has some serious potential.

In its current state, with a simple setup and the ability to quickly train the bot and maintain it over time, there is no excuse for saving your customer service rep’s time with this tool.

I would love to see better integrations in the future where support tickets could be automated based on certain language triggers or sequences. And the ability to integrate right into a CMS without software like Zapier would be a plus as well.

If you haven’t tried any AI product before, this is a fun one to play around with since it has a Free tier option.

FAQ

How does the Slack integration work?

The installation for the Slack integration guide can be found here, but what you may be wondering is what does the integration actually do?

By integrating Chatbase with Slack, it allows you to “talk” to the chatbot directly from Chat.

Once you finish integrating Chatbase, you will need to invite the Chatbase app to your channel. Do this by selecting the correct channel and do the “/invite” command. Select the “Add apps to this channel” and select Chatbase.

If Chatbase is not on the list of apps, then it was not integrated properly. Return back to the help doc and try again.

After you add Chatbase, you can chat with it by mentioning it. For example, “@chatbase how are you doing” will start a dialog with the chatbot.

If you have trained your chatbot on your internal documentation, then you could use this Slack integration to quickly find the right documents, saving your organization time.

How many more tokens does GPT-4 cost compared to GPT-3.5 Turbo?

According to Chatbase it cost 20 times more credits. For example, 1 GPT-4 message = 20 message credits vs 1 GPT-3.5 message = 1 message credit.

gpt-4 is much better at following the base prompt and not hallucinating, but it’s slower and more expensive than gpt-3.5-turbo (1 message using gpt-3.5-turbo costs 1 message credit. 1 message using gpt-4 costs 20 message credits)

help text in chatbase around message credits