What are Generative AI chatbots?

Generative AI chatbots are advanced dialog systems that can generate answers and texts independently with the help of artificial intelligence and machine learning. They use large amounts of data and learn from interactions to simulate human-like conversations. These chatbots are not limited to pre-programmed responses, but can generate new and unique responses that respond to the user’s context and intentions. They are used in a wide range of applications, from customer service to support for complex tasks such as programming or writing texts.

Generative AI chatbots use knowledge sources that we humans provide them with as the basis for their answers. These can be web pages, PDF documents or other data sources.

How are generative AI chatbots developed?

Generative AI chatbots are based on a combination of an AI language model, knowledge sources and fine-tuning.

The well-known ChatGPT 3.5 or ChatGPT 4.0 is often used as the language model. However, it is also possible to use open source models such as Falcon. Fine-tuning is usually done with the help of prompts. Suitable sources of knowledge are websites, PDFs, databases or other sources from which the chatbot should obtain its knowledge.

Create knowledge sources for Generative AI chatbots

In order for your Generative AI chatbot to answer user queries as correctly and accurately as possible, it is important that it has usable and comprehensible sources of knowledge. A source of knowledge that may seem useful to us humans is not necessarily comprehensible to a chatbot. When creating knowledge bases for your Generative AI chatbot, there are a few best practices to consider.

Document type

Ideally, you can provide your Generative AI chatbot with the knowledge sources as a PDF document. However, it is also possible to use other file formats.

Alternatively, website URLs are also possible. However, if you use websites as a knowledge base, you should ensure that the website complies with the following text requirements. These text specifications differ in part from the classic specifications from user experience research. A well-performing website on Google or with your human users is therefore not automatically useful for a Generative AI chatbot 😉

Text format

Initial experience shows that texts in the form of continuous text are the most suitable. PDFs are usually the easiest to understand. It is important that they are real PDFs and not scanned documents saved as PDFs.

Furthermore, all content should be contained in the continuous text and not be additionally contained in images, tables or other graphics. It should really be a continuous text with complete sentences and no simple lists.

Ideally, your body text should contain different sections, with each section containing the same important keywords. In this case, it is not a problem if there are repetitions that are logical in terms of content. Most generative AI bots scan the individual sections of text, including the title, separately. It is therefore important that the keywords are repeated in every effective paragraph.

If you work with links, these must be written out as part of the body text and not linked as hyperlinks in the text.

Best Practices

The following best practices have proven to be very successful in recent months:

• PDF as file format (no scans)
• Continuous text
  • No tables
  • No lists (no bullet points)
  • No pictures or graphics
  • No alt texts on images
  • Whole sentences
  • Links written out as part of the body text (no hyperlinks)
• Sections must contain keywords
  • Titles and their corresponding paragraphs are scraped as separate parts. It is therefore important that the keywords appear in every effective paragraph so that the bot does not just find the titles. Each paragraph must also be written in complete sentences with reference to the title.

Even more tips

There are many more tips for your Generative AI Chatbot projects in the article on my learnings about Generative AI Chatbot projects.

And of course you can always contact me directly if you have any questions. Just send me a message via WhatsApp or e-mail.

Or come directly to my WhatsApp group – where I regularly post use cases, news, best practices, events and much more about chatbots, ChatGPT and co.