If a visitor question contains certain keywords, how to define a custom
reply for the bot to answer?
Trigger > custom response.
And how to train the bot to provide product recommendations?
We are a WooCommerce store, rather let the bot provide non-business related
information, I want the bot to push the product url to the visitor
How to fix bot’s wrong answers?
Sometimes, the bot give complete wrong answer
Even your own bot make mistake, it tell me your support email is info@answerly.io
How to manually fix this to prevent it happening again?
I just checked your dataset, and the bot will provide incorrect answers because you are using the wrong approach for listing a product.
The best way to handle multiple products is to create a Google Sheets table with categories, descriptions of the product, and links to the article. The description is what will match the user query, and you can specify in the business description that you always want to send a link when the user matches the product they are looking for.
Our chatbot is providing inaccurate information because we recently released a significant update that has changed the entire knowledge hub, and we are still writing the new documentation.
Do you have a recommended way to export WooCommerce products to Sheets?
I have about 300 products, with 11 language translations, about 3K product pages.
You said that there is a change to knowledge hub, do you mean your own Answerly knowledge hub? and that is why the bot say your support email is info@answerly.io?
I have another two questions:
One instance, I asked the my own bot, can you recommend any Glucoamylase?
The bot said yes, but the bot made up a url www.enzymes.com/glucoamylase
This site is neither managed by me, nor it is a valid URL
How to avoid the bot making a fake url like this?
Another instance, I told the bot to provide product URL.
It replied, as a OpenAi bot, I can not surf the internet, something like this.
How to avoid it telling I am a “openai bot” etc.
I guess exporting the 300 products in English is fine, GPT should be able to recognize the language and answer accordingly.
The change in the knowledge hub has brought a lot of updates and completely reworked the previous one. We still need to train our chatbot with the recent updates.
If you follow the Google Sheets advice with description, category, and URL, it should not make up the URL when the description of the product is matched. Hallucination occurs when the information is not found or is duplicated / overlapping.
The third problem could be caused by the quality control features. Try to disable them when you are building the dataset, specifically the one that says “Prevent Unrelated Conversations”.
ChatGPT is not always perfect, but if you build the right dataset, you will get very good results with near 100% correct responses.
We are working to improve the ability to easily find and improve your dataset with features like fine-tuning quality scores.