How to Export Transcripts From Custom GPTs (OpenAI GPTs Tutorial)

Key Takeaways

in this video I'll be showing you how to automatically export and analyze conversations from your custom gpts across channels like web chat widgets WhatsApp Instagram and more this has been a highly requested video because many of you have started deploying your custom gpts to these channels but once a way to track and analyze interactions with your users the truth is these conversational channels can be insanely powerful for businesses but without properly storing and analyzing this data you are losing half of the value of your gpts as this data can contain valuable information for use in future decision- making by the end of this video you'll not only know how to automate the exporting of transcripts from your custom GPT deployments but you'll also get access to my complete automation template for analyzing and extracting valuable data from them with the help of AI adding huge value to your GPT powered Solutions if you don't know who I am my name is Liam otley and I run my own AI development company Morningside Ai and run the largest community of AI agency owners in the world where my students learn how to make money by selling AI solutions to businesses just just like the system you're about to build let's get into it firstly we need a quick explainer of this build and how this system works this is going to be using the assistant API which you may be familiar from my other videos but we are creating these custom gpts that we can put onto our our WhatsApp Channel onto our website Etc but we're going to be calling that same API but this time in a different way that allows us to request and and pull all of the transcripts and messages from the interactions that our users have had with the API open AI allows us to get that information at any time we want but you need to know how to request that information from the API which is what we're going to be covering firstly in this video in this particular video we are going to be building on top of all of the work I've done in my prior videos such as my WhatsApp and Instagram deployments so in order to fully understand this and and how it's all working uh you need to watch one of those videos up here I'll put the Instagram one up so you can see that essentially we're going to be using that same system adding a little bit in that allows us to export the transcripts and then showing you the air table setup that we're using and also the make automations the way this works is that every time a conversation is started on our website on WhatsApp on Instagram wherever it is every time a conversation is started we are going to be saving what's called a thread ID and every time the conversation starts we generate a thread ID and in my prior builds we hadn't been doing anything with that thread ID apart from using it in the conversation but in this case we are going to be saving that thread ID to an air table and then we're going to be taking that thread ID and looking up all of the messages within that thread ID pulling them all out packaging them up analyzing them and then putting them all into a nice database and spreadsheet for us to look at and analyze later as always with my videos everything that you see here is going to be available for free on my resource Hub in the description so you can get the air table template which is the database you have the make automation template which is a a very complex one that I've created for you guys to all use and plug in and get started with immediately then we have a repet template and also a mini chat template so it's all templated you guys are going to get access to all of it and I'll walk you through how to use it now to get started you need to clone my ripple which is available on my resource Hub as I mentioned there'll be a link there that you can click and it'll open up this page and you'll be able to click this Fork button in the top right and that'll take you to a page that looks a bit like this once it's all booted up again as I mentioned in order to understand what's going on here you do need to watch the Instagram deployment the video which I will now Link in the description as well so watch that and come back and a lot of this will make a lot more sense and I'm going to scam over a few things like the API keys that I did Cover in depth on those videos as well but this particular Builder is is just a basic custom knowledge chatbot I have removed all of the tooling from this so all it is using is a knowledge based document with information about my AAA accelerator program so I've given it information on that and point of this chat bot is going to be able to ask questions about the pricing and the features and things like this now you won't necessarily see me using it in this because I've already created some transcripts for us to use later in the video but I'll give you a quick walkthrough of things here and how this is different from my prior videos and how this is going to allow us to export the transcripts so that we can analyze them and and manage them as well and the functions. P file is the main difference here which is this add thread function and this is interacting with an air table base that I've created that stores all of the threads which we get to in a sec aside from that we have this create assistant function here so if you want to modify this and change the knowledge base you can delete your knowledge base on the on the left this knowledge document and upload a new one and then you can change the name of the document here as I've done in my prior videos there is no tooling here all it is is a custom knowledge base there's no lead capture there's nothing the point of this is just to show you how you can export the thread IDs and then pull that information from there in a second and our main. PI file is another key difference here from my other videos and that's in this start route when we're calling to start the conversation as a mentioned we're going to start the conversation and then when we generate that thread ID we need to send that thread ID to our air table and save that to access it later and pull the information from and here is where we are calling the ad thread function to send that information to our air table in this start conversation function there is one other difference which is this platform argument that we're going to be including say if I deploy this application and I want to use it on messenger and I want to use it on WhatsApp we can actually specify on the many chat side of things which you'll see in a second we can specify which platform we are calling this this start conversation from so we are able to see in our air table where these conversations and where these threads are coming from if you are following along and want to copy this and get it working on your end then you do need to change these two Secrets up here I've gone over how to get your open AI key before and how to get your air table API key as well as mentioned I have covered that in my WhatsApp and Instagram videos jumping into our air table now you are going to be able to clone this entire base it'll be in the resource Hub as well there'll be a button when you click on the link that'll be up in this top corner you can click on that and it allows you clone the entire thing and start to make edits once you've cloned the air table base you need to do one thing in order to make this work on the repet side which is go over to the top right and you can do your API Keys as you do this as well and you can click on the web API documentation and if you go down to the Smith solar CRM if you scroll down to the threads table and if we go to create a record it will get this URL here that says threads at the end oh I need to get the whole thing copy that and then we can head back to our functions. piy and you need to delete this which will be my one and paste in yours and that will make sure that it's trying to send to yours and not mine you'll see there's a couple other tabs up here these are just from the other projects I like to just keep them all in one so I don't have 50 different air table bases for all these different videos but the only ones we worried about in this case is going to be the threads and the overview one in the threads table here I'll give you a quick orientation here are the thread IDs that are sent from our application when we call this uh add thread function we're passing the thread and the platform and that is being saved into our air table here we have the thread and then we have the platform as a nice little drop down select here air table is also automatically filling the created at field and then we have the real data that we're trying to extract out of these thread IDs which is the transcript and the topic and in this case I've added a status column and a date one as well so the status can be processed arrived or empty which I'll go into in a second and the date refers to what's in the overview here where you can get a high level overview of what kind of volume is coming through what kind of questions are being asked Etc on the left panel here I've made a few different views that allows us to switch between them on the screen I have this ready to process view which is basically anything that hasn't yet gone through our automation system to pull the transcripts out and label them and classify them and everything which you'll see in a second these are basically waiting to be processed hence the ready to process name then we have the processed ones what will appear here are all the transcripts that have already gone through our automation system come back and they've been labeled they've got the full transcript Etc and then here's an empty one cuz sometimes you get empty threads we want to keep them out of the data set and then you can see all threads here as well for these thread IDs to actually be getting created and sent off we of course need to call it within our application that's actually sending these messages and in this case we're using mini chat like with my other tutorials so the difference here between my other ones is that in this case we needing to go on this create thread and here I've added this question mark platform equals Instagram so in this case it's sending a a get parameter we're sending a a little bit of information attached to the URL and if you go back to our main.py file that is where we are expecting it here so it's from the request it's getting the platform key and it's saving it to platform and then we are sending off platform to our add thread function which is adding it and it goes back to air table and make sure that we're getting the right value in the platform column it's a little bit complex cuz there three different things going on here but basically whenever a conversation starts on WhatsApp or Instagram we're going to be sending off the name of the platform so that we can log it in our air table as well aside from that everything in here is the same as my other videos so I've gone off screen and I've gone through the conversation you can click the preview button up here and if you click the preview button I can copy this and send it to myself on Instagram and I've had a bunch of conversations and these have all been logged to air table already by using this function which is being called every time they start and so that is what you are seeing here which are all of the thread IDs from the conversations I've had are waiting to head our automation system so that they can get all the transcripts pulled which we getting into now we now need to call the assistant API and request the information that's inside that thread and process it and send it back to air table for that we'll be using make.com which is probably my favorite automation platform and luckily for you I've already created an entire template to make this as easy as possible for you so what you'll need to do to get that template as go on my resource Hub of course there'll be a link to sign up to make.com if you haven't already and there'll also be a link to download download the template and then you can come back to the screen ready to create your first scenario now on the top right we can click create a new scenario and then with that file that you downloaded you can click the more button on the bottom and click import blueprint choose the file and choose the blueprint here and then save it and Bam you get all of this work that I've done today which is going to completely automate the whole process now I've tested this pretty heavily it is working very well for me right now to get this working on your end it's a very easy setup we've just got to do a little bit with API keys and then we can get it running first thing you need to do to get this started is to click on the search records node here and you need to add a new connection mine's already set up but it won't be for you on the connection type you can click on air table token or key and then we need to give it a name this I'll call tutorial 2 and then we need to go back over to air table and create an API key if you haven't already so if you've already set this up and you know what to do you'll probably already have an air table API key that you use within repet and then so you can take that same one over and use it within make as well but for those you who haven't got one yet we can click on the developer Hub up here go to the personal access tokens on the left click create new token tutorial 2 now with the Scopes here I've had issues with these so I recommend just go all out here it seems be one of them that caused the issue and I think it is this weo manage one make sure you've added all of these Scopes because if you miss one of them then it's not going to allow you to make the connection over on make so once you click confirm you'll be able to copy your API key hit back over to make and then you can scroll down and paste it into here and then click save and it's going to have created a new connection for you here in make that you need to set up on all of the other ones as well so all of this should have pre-filled but if it hasn't for you then you need to copy the exact settings here of the base being Smith solar CRM the threads table the view of being ready to process and the only output field we're interested in is the thread ID the limit is 100 also then we need to do the same things on the rest of the air table nodes so we can click here make sure you're using your connection and these are the settings here if you need to pause it and take a look and then finally we have the update record here and the these are the settings you need to copy here so Smith solo CRM threads we're pulling the record ID which should have already populated we have the all messages the topics these map toggles are turned on for these and we also have Smart links as true so make sure that you have that on as well before I run it I should probably give you a quick breakdown of what's actually going on here so you're not completely blind but what we're doing is searching for records and air table that are on that ready to be processed uh view the ready to process view is Handy like that because we can keep it as a a storage or a holding area for all of these transcripts that have come in over theast last day or so as our people have interacted with our with our gpts and they all land here waiting to be processed and then we can just fire at once fire the automation once on a scheduling which I'll show you in a second and then when we fire that it's going to run through all of these and run the processes needed to pull that in analyze the transcript and fill out these fields once we have pulled all of those records in from the ready to process table then we need to aggregate them all into an array which you don't need to worry about too much then we're basically going to Loop through all of the different records so we might get 50 different records that have come in in one batch then we need to Loop through each of them which is done with this iterator here and everything that follows after this it needs to go through all here and then loop back over for each one so one record loop back over two record so it goes over and over and over again on every different record by pulling the thread ID and and doing these next steps with the thread ID that we've pulled in from that specific row we then call the open AI API that allows us to get the information from these threads and we inserting the thread ID from that row into this get URL here and actually I've just noticed one other thing you do need to change uh you need to add in your own API key here so this is your open AI API key so you can go get that from your open AI account and come down here and just replace this keeping the bearer and the space that's going to allow you to call and ask for the information back and it's going to allow you to be authenticated to do that what this does is pulls back all of the messages out of the thread then we process that data or bit if there's no messages in the thread then we go to this one which is basically going to say okay this is an empty thread and label it as empty but if there are messages then we're going to Loop through all of the different messages that came in and smos all of them together into one one big long piece of text saying user said this assistant said this back and forth and then we send it off to a chat GPT node in which I've written this prompt that says it is a transcript analyzer and I've given it a couple different labels you guys can have a look through this in your own time basically in my use case where this GPT has been an assistant helping people learn more about my AAA accelerator I wanted to be able to pick out a couple different labels or or things that might have happened within the conversation so I said they might have asked about pricing of the accelerator they might have asked about the features or benefits they might have asked about me and about the business and about who runs it and they might have asked about something outside of that which have C other then all I'm telling it to do is to Output well identify if they are those labels in the in the transcript and then output a comma separated set of labels for this particular transcript and then all I'm doing is passing in that big smooshed up transcript that we've created I'm passing that in as the input and telling it to okay pick out the labels for this and return it as comma separated labels we then smush those labels together into an array then we're able to send it off to air table and update the record that we've been working on so that's when we pass in okay now we have the whole transcript pulled from the API into that big smos one we can send that back we have also the topics or the tags or the labels that we've got out of the conversation we're going to send those off as well and then we can update the status of it to being a processed record so with all they're talking out of the way we can now show the magic happening live in front of you I've switched over to the process tab which is empty right now but as they go through and are labeled as processed and we've done all the stuff that we wanted to they're going to appear here with their transcript and with their tags as well so if I click the Run once button here we should start to see them populating in here bam transcript topics you can just sit back and watch as it continues to go through all of these different transcripts you'll notice that there's a little bit of a difference between the number in the processed and the number that we're in there ready to process so we can see that it's finished over here it's run through like 40 different messages 14 different rows and in the empty one you'll see there's a couple that have shown up as well with the thread this allows us to get our process data nice and easy put into one table and here you can see the transcripts have been saved all this information here and we have the labels as well as I mentioned there is one more part of this that I've just thrown in there to show you kind of where you can go with this information once you've been able to process it and that is this overview tab up here so what the overview tab does is it's sorted by date so we have each single day of the month you sort of add these records as you need to but what we have is Count columns here so this is the total number of messages that happened on that day this is the number of pricing related questions the questions that have been tagged with sping that have shown up the feature related questions the about questions the other questions now this is a linked record or a linked table to the threads at this point it's set up manually so you can see I can remove this and I have to go in there and add the date as the 17th so you have to go through manually at this point it's a little bit more advanced to set this up automatically but this allows you to show how you can label all of these to a certain date and then start to see some overview data of okay how many messages are we getting as it trending up as it trending down and then you can go into air table interfaces which allows you to make a dashboard out of of this which is an entirely different video but it starts to show you where you can go once you have this kind of quantitative data we have numerical data based off the information and and the things that are happening out in your in your chat Bots and that about wraps up for the video guys I hope this has been helpful I hope this has opened your eyes to how you can be pulling information from the assistant API to get that thread data to get the messages that are happening in all these different channels and aggregate them into one place and then more importantly how you can actually process and analyze that information so that you can start to get some valuable numerical and types of data that you can actually pull insights out of for your business or for the clients that you are working with if you have enjoyed the video please hit down below and leave a like think something like 95% of people who watch these don't leave like so i' really appreciate if you can go down there it supports my channel supports me making videos like this and giving away these templates and everything that I do my you down that you can subscribe to the channel for more videos like this where I teach you how to build AI Solutions and sell them to other businesses or Implement them in your own and if you're not already in my AI business accelerator program where I teach you how to make money with Solutions like this there'll be a link in there as well and of course resources mentioned in this video shown in this video are going to be available on my resource Hub so you can get access to them all there for free if you want to learn how you can create Advanced custom gpts with lead generation capabilities you can check out my video here where I go into depth on that but aside from that guys that's all for the video thank you so much for watching and I'll see you in the next one