Hi @JeffImel  Excited to hear that you’re giving the data source integration a whirl! 

@Troy Tessalone - have you tried a ChatBot with this feature yet? Any insight you could share? TYSM!

@Liz_Roberts 

I have not as it’s a paid feature that requires upgrading.

Hi

I have the same issue, my chatbot will not give answer to specific questions which can be found in the data source. 

I look forward to a speedy response otherwise my upgrade is wasted. 

Thanks

Rowland

Hi there @Deputy,

You might be running into a bug here. I recommend reaching out to our Interfaces team regarding this issue.

You can send a ticket via this link: https://eap.zapier.app/interfaces.

Hopefully, this helps! 

I submitted a ticket last week on this issue and have not heard anything back.  As of a few minutes ago, you can ask my chatbot pretty much any topic and it has an answer. My datafile is not being used at all.

Still no worries.  This is bleeding edge functionality. It will get worked out. Always does.

Jeff

Hi @JeffImel,

I apologize if our Interfaces team hasn't gotten back to you yet. They're currently managing a Beta product, so their response times might not be as swift as our Core Zapier Support. But rest assured, they're working hard to get back to you as soon as they can.

We really appreciate your understanding and patience during this time. Thank you! 

What is the best way to format a data source for Zapier’s Chatbot?

A .txt file - how does the Zapier Chatbot “read” a data source in .txt format? Should I divide subjects or topics into paragraphs?  How many carriage return lines between paragraphs? If I make a list inside of a paragraph may I use bullet points or should I leave the out so that it is pure text with no symbols?

Jeff

Hi @JeffImel,

I’m so sorry, but I have exhausted my resources trying to find an answer to your question. I suggest reaching out to our Interfaces team, who you're currently in contact with via email. They'll be able to provide further assistance.

Thank you for your understanding.

SHOUT OUT TO THE ZAPIER CHATBOT TEAM!!!!!

GREAT JOB!

The Chatbot is now only answering questions about first aid. When I ask a question outside of first aid I receive a very polite answer stating: 

I apologize for the confusion, but as a wilderness first aid instructor, my area of expertise is in providing first aid information and guidance. I do not have access to instructions on changing the oil in a car. I recommend referring to your car's owner's manual or consulting a qualified mechanic for specific instructions on how to change the oil in your particular vehicle. Is there anything else I can assist you with regarding wilderness first aid?

EXCELLENT!

I will now work on formating my data source in a .txt format with a couple of carriage returns between subject paragraphs.

I really value the work the Zapier Chatbot team put in to make all of this work!  I appreciate y’all!

Jeff

The Zapier Chatbot using my data source is working as advertised.  Amazing.  I have embedded it into a sandbox web page built with Elementor and it works fantastic.  I am beta testing it with a group of former students of my school.

Great job Chatbot and Interfaces Teams!

Jeff

That’s fantastic news, @JeffImel!  

Really appreciate you taking the time to share this wonderful feedback, I’m so pleased the Chatbot component is working well. I’ve passed on your comments here over to the relevant teams involved so that they don’t miss it! 

It’s great to see that the development of the data source feature is making progress and that the ChatGPT feature in Interfaces is starting to draw information from the file into account when answering questions.

However, when we tested it out during the the last days, we noticed that there are always a few questions where the chatbot will not draw its answers from the knowledge base even if the question included there is copied as the prompt - regardless of the language and the format of the file uploaded as the data source. While it is always the same questions that will not yield an answer when entered as a prompt, we have no clue why the chatbot is having trouble with them in particular and not with the others. To make our observations more comprehensible, we are attaching the json file uploaded as the data source along with a screenshot of some of the results we got when we copy-pasted questions from the knowledge sheet into the prompt/user message field of the chatbot (I have copied the content of the json file to a word document because I cannot upload the json file here directly). The directive we gave the chatbot was: 

Please answer the questions posed to you in the language they are raised by a user. Answer the user’s questions in a friendly manner from the perspective of the customer service of Ersatz-Pilot. Ersatz-Pilot is a passenger rights website that pays out flight compensation directly to its customers in exchange for their claims against their respective airline which arise when either the airline cancels the flight, or the flight is delayed by more than 3 hours, or the passengers cancel the flight (in the latter case, they are at least entitled to a refund of the unused taxes and fees in their ticket).

In the connected data source you will find information on Ersatz-Pilot and its services in English in a json file in the format of a question as key directly followed by the corresponding answer as the value. Each pair of question and answer is separated from the next through a comma. The user may ask a specific question or about a more general topic. Please search through the questions and answers to find the most relevant information.

Any idea of why the chatbot is incapable of answering some (but by far not all) questions even if they are included in the data source word by word would be highly appreciated. Likewise, it would be amazing to hear about any way of structuring the directive and the questions and answers in the data source that does not run into this kind of problem. 

I have deployed my chatbot to a beta group of customers for the last two weeks. My customers love it!

I am experiencing the same results. Information outside of my data source is included sometimes. 

I too would like to learn more about structuring the directive and formatting the data source so that the chatbot will use my data source.

Jeff

Hi @RefundPilot and @JeffImel!

@RefundPilot - For missing answers I wonder if it could be that it wasn’t able to find a match as there were commas within the answers. In the directive you shared above it mentions that “Each pair of question and answer is separated from the next through a comma” so I wonder if that’s what prevented all the questions from being read as it may have read a comma within the text and assumed that it was the start of the next question. Can you try changing the character that separates the question and answer pairs to see if that helps?

@JeffImel - There’s a guide on how to use data sources that covers the best practices on how to structure the directives which you may find helpful: Use a data source with your Interfaces AI Chatbot. 

Hope that helps. Please do keep us updated on how you both get on. 

Hi @SamB ,

Thank you for your response. We tried uploading json files as the data source and gave according instructions in the directive so that the Chatbot regards the keys as questions and the values as answers to rule out any hiccups due to the commas. Initially, this gave us the same results as before (i.e. often the Chatbot would respond that he didn’t know or rather did not find anything in the data source even if we posed the exact same questions listed in the json files as keys). However, we retried it today and now this message is not repeated. So generally we would have reason to assume that the Chatbot is finding information in the data source, indeed, by now. Unfortunately though, the quality of the answers is poor. More often than not, it is the exact same opposite of what is included in the data source. We have the strong suspicion that even though the Chatbot is not notifying us that it is unable to draw information from the data source, in fact it does so and simply guesstimates its way through the questions based on the closest related information it can find on the backup copy of the internet from 2021 it regularly taps to answer queries. For example, it will describe services of our company based on information our competitors have been putting out even though we handle some processes quite differently and have made that clear through the data source.

Since the Interfaces team probably worked on the Chatbot integration in the last weeks, we are wondering whether in fact the feature has been disabled that would ask the chatbot to flag a question as unanswerable based on the data source rather than jump right to the general source?

At any rate, the question remains what we can do to make the Chatbot acutally take in the information provided in the data source. Is there a best practice emerging in terms of the ideal file type that should be used, the directive to read through the file type that should be given? Any feedback on this matter would be highly appreciated.

Best regards

Sorry to hear that @RefundPilot. In that case I think it might be best to get in touch with the Interfaces team about this. They can further investigate and troubleshoot what might be causing the these issues with the data source not being referenced by the Chatbot correctly. And are better able to advise on the ideal file type and directive to use for your use case here.

Keen to know how you get on with this so please do let us know whether the Interfaces team and you are able to get a solution in place!

Hello, sorry but there are no answers regarding how did you solve this issue, i have tried with differnet forms from txt, docx, pdf and the bot is still not taking any information i have provided in the document.

I have also tried indentation to group concepts within that documetn but still, the answer is always, i have not been provided with the necesary infotmation to provide that answer even though its clearly in the document.

We have hired this from 2 different accounts, just to test, and nothing is working. Please let me know how you solved the issue if that was the case, but since there is no solution provided i can only guess this is still having  bugs in order to provide answers based on custom information.

Hi @CrisCan, welcome to the Community! 

I’m so sorry to hear that it’s not working for you either. I took a quick look at the email correspondence with the Interfaces team but, I’m not certain that it was fully resolved as the Interfaces team were waiting on a response back from them. 

I’d recommend contacting the Interfaces team directly so they can look into this further with you. You can do that here: https://eap.zapier.app/interfaces

If you are able to finally get to the bottom of this with them please let us know. Keen to ensure you’re all set!

Still not resolved.  Ticket submitted.  Honestly, you shouldn't’ charge to upgrade if it's this buggy.

Really sorry to hear that @Rgbrown. The chatbot component for Interfaces is in Beta at the minute, and while it’s still in active development bugs and glitches are a bit more likely during this phase unfortunately.

I can see you’ve got a conversation going with the support folks in the ticket you submitted, so if you have any follow up questions on that it would be best to continue the conversation with the Support team to get those answered. I can totally understand the disappointment caused when running into this issue following an upgrade, so I’ve passed your comments here to the relevant folks on this end so we can work on improving this experience in future.

If we can help with anything else in the meantime, please do feel free to reach back out to the Community here. 

Hi @SamB ! I’ve been facing the same issue, and I’m a bit upset that the paid version has this kind of bugs still. I tried to create a second bot and embed in my website, and it completely changes the instructions after embedded. I’ve submitted two tickets already, can someone pls help me?

Thanks and happy celebrations to you all! <3
Marcos
 

I’m so sorry for missing your message here previously, @MarcosSahade! 

I did some checking on the support tickets you mentioned and it appears that from Support’s side it looked as though the issue had since righted it self. Can you confirm whether that’s definitely the case?

Looking forward to hearing from you! 

Hi Sam! Yes, it’s ben solved!! Thanks =D

That is fantastic news @MarcosSahade! I’m so glad it’s fixed!  

Please do get in touch again in Community, or with our helpful Support team, if you need any further assistance in future - we’re always happy to help!