Create Blogs in Seconds Using Perplexity AI

Summary

The video script introduces an AI news fact-checking setup that automates the verification of facts within articles. It uses chat GPT for drafting but addresses inaccuracies by splitting articles into facts, checking them individually via an online service, and updating a database in Airtable. The process includes creating tables for articles and facts, setting up automations for fact-checking, and integrating with an external API for verification. The result is an accurate, verified article, beneficial for bloggers and content creators.

Takeaways

• 😀 The video introduces an AI news fact-checker setup designed to automate the verification of facts within articles generated by AI, such as Chat GPT.
• 🛠️ The creator shares a personal story about a blogger who uses AI to write articles but struggles with fact-checking, which inspired the development of this AI tool.
• 📝 The AI fact-checker works by splitting an article into individual facts, checking each one for accuracy, and then compiling the verified information back into a coherent article.
• 🔍 The tool uses an AI prompt to format the article into logical chunks and then leverages an external service, 'perplexity,' to verify the facts by searching the web.
• 📈 The process involves creating an Airtable database with tables for articles and facts, and setting up various fields and automations to manage the workflow.
• 🔗 The setup includes linking records between the articles and facts tables to associate each fact with its originating article for organized verification.
• 🔧 The video provides a step-by-step guide on building the AI news fact-checker, including the creation of views, fields, and formulas within Airtable.
• 📚 The tutorial covers complex strategies and thought processes behind the automation, teaching viewers how to implement similar systems on their own.
• 🔄 The fact-checking process is automated through a combination of Airtable's automation features and external API requests to verify the accuracy of each fact.
• 📊 The video demonstrates the use of iterators and conditional logic to handle the verification of multiple facts and update the status of each within the database.
• 🎯 The final product is a verified article with corrections and supporting evidence where necessary, ensuring the information is accurate and up-to-date.

Q & A

What is the main purpose of the AI news fact-checker setup described in the script?

The main purpose of the AI news fact-checker is to automate the verification of facts within articles, ensuring that the information posted is accurate, relevant, and up-to-date.

Why was the AI news fact-checker created?

The AI news fact-checker was created to address the issue of inaccuracies in articles generated by chatbots like Chat GPT, which sometimes produced fake or outdated information due to the data it was trained on.

How does the AI news fact-checker handle the verification of facts within an article?

The AI news fact-checker splits the article into different facts, loops through these facts, and uploads them to a fact database. It then uses an external service like Perplexity to verify each fact individually and updates the database with the verification results.

What role does Perplexity play in the AI news fact-checker process?

Perplexity is used to search the web and verify the accuracy of each fact. It provides the system with the ability to research and validate the information, ensuring that the facts are true or false with supporting evidence.

What is the significance of the 'Start Fact Check' button in the AI news fact-checker setup?

The 'Start Fact Check' button triggers the automation process within Airtable, initiating the fact-checking sequence for the article, which includes splitting the article into facts and starting the verification process.

How does the AI news fact-checker ensure that all facts in an article have been verified?

The system tracks the status of each fact check and only marks the article as 'ready' when the total number of facts equals the number of facts verified, indicating that the verification process is complete.

What is the importance of the 'compiled facts' field in the AI news fact-checker database?

The 'compiled facts' field is a rollup field that compiles all the verified facts into a single view, allowing for an easy compilation of the final verified article.

How does the AI news fact-checker differentiate between verified and unverified facts?

The AI news fact-checker uses a 'fact check status' field to indicate whether a fact is 'not started', 'in progress', or 'complete'. Only facts marked as 'complete' are considered verified.

What is the benefit of having a 'supporting evidence' field in the fact database?

The 'supporting evidence' field provides the rationale behind the verification of a fact, offering specific evidence or sources that confirm whether the fact is true or false.

How can users of the AI news fact-checker benefit from the final verified article?

Users can benefit from the final verified article by having a publication-ready content that is fact-checked and updated with accurate information, ensuring credibility and reliability.

Outlines

00:00

🤖 Introduction to AI News Fact-Checker Tool

The speaker introduces an AI news fact-checker setup they've created to address the issue of verifying facts within articles generated by AI, such as Chat GPT. The motivation stems from a blogger's challenge of manually fact-checking due to inaccuracies. The tool automates the process of fact-checking and promises to demonstrate its functionality, output, and the thought process behind its creation. It's mentioned that the tool is complex, involving various strategies and an understanding of Airtable.

05:02

📝 Building the AI News Fact-Checker Database

The tutorial begins on setting up the AI news fact-checker by creating a new database with 'articles' and 'facts' tables in Airtable. Fields are added to the 'articles' table, including an auto-numbered 'article ID', 'article' for text, 'AI prompt' for formatting, and 'fact check status'. The process involves creating views, linking records, and setting up automations for the fact-checking process. The speaker also promotes their school community for further support and resources.

10:03

🔍 Fact-Checking Process and Airtable Setup

This paragraph delves into the specifics of the fact-checking process, detailing the creation of fields within the 'facts' table of the Airtable database. It includes 'fact ID', 'original fact', 'accuracy', 'fact correction', and 'supporting evidence'. The setup involves linking facts to articles, creating views for managing facts, and setting up automations that trigger the fact-checking process upon user action.

15:05

👨‍💻 Automation and Fact-Checking Workflow

The speaker outlines the automations needed for the fact-checking tool within Airtable, starting with the 'Start Fact Check' button that triggers a Make.com scenario. The process involves fetching the article record, updating statuses, splitting the article into facts, and preparing the data for fact-checking. The automation handles the transition from 'not started' to 'in progress' and sets up the structure for fact verification.

20:06

🔄 Iterating Facts and Online Verification

The paragraph explains the iterative process of sending facts to an online verification service, likely Perplexity, to check their accuracy due to rate limits. It involves parsing text to ensure JSON compatibility, making HTTP requests to the verification service, and handling responses. The process is automated to avoid crashes and ensure each fact is verified individually.

25:08

📝 Finalizing Fact-Checked Articles

The final part of the process involves receiving responses from the fact-checking service, formatting them into JSON objects with fields for accuracy, evidence, and corrections. The information is then parsed and used to update the Airtable records, marking the verification status as complete. An additional automation updates the article's status to 'done' once the verified article is compiled, concluding the fact-checking process.

30:09

🎉 Conclusion and Call to Action

The speaker concludes by expressing excitement about sharing the AI news fact-checker tool and encourages viewers to like and subscribe for support. They highlight the value provided by the tool and its real-world application, hinting at future developments and offerings within their community.

Keywords

💡AI News Fact Checker

The 'AI News Fact Checker' is the central theme of the video, referring to a system designed to verify the accuracy of information within news articles. It is an automated tool that uses AI to ensure the facts presented are up-to-date and correct. In the script, the creator discusses building this system to help a blog owner who was facing issues with the authenticity of information generated by Chat GPT.

💡Chat GPT

Chat GPT is mentioned as the AI tool initially used to generate blog content. It is an advanced language model capable of creating human-like text based on prompts. However, the script highlights a limitation where Chat GPT sometimes produces outdated or fictional facts, necessitating the creation of the AI News Fact Checker to verify its output.

💡Fact Checking

Fact checking is the process of verifying the accuracy of statements or information. In the context of the video, it is the primary function of the AI News Fact Checker system. The script describes how the system breaks down an article into individual facts and checks each one for truthfulness, using tools like Perplexity to search the web for supporting evidence.

💡Perplexity

Perplexity, in this script, refers to an AI tool or service used for searching the web and verifying the accuracy of individual facts. It is part of the verification process in the AI News Fact Checker system. The script mentions making an HTTP request to Perplexity to check the validity of each fact presented in the articles.

💡Automation

Automation is a key concept in the video, referring to the process of creating automated workflows that can perform tasks without manual intervention. The AI News Fact Checker system is built with automation in mind, using triggers and scripts to automatically check facts and update the database. The script outlines several automations, such as starting the fact check and updating records when fact checks are complete.

💡Airtable

Airtable is a cloud-based database tool that is used in the video to organize and manage data. The script describes setting up an 'AI News Fact Checker' database in Airtable with tables for articles and facts, and using it to store and update information throughout the fact-checking process.

💡API Key

An API key is a unique code used to authenticate requests to an API (Application Programming Interface). In the context of the video, the API key is used to access Perplexity's services for fact verification. The script mentions obtaining an API key from Perplexity to make HTTP requests for fact checking.

💡JSON

JSON (JavaScript Object Notation) is a lightweight data format used for data interchange. In the script, JSON is used to structure the data for fact checking, ensuring that information is properly formatted for processing by AI tools. The video discusses formatting text for JSON compatibility and parsing JSON results from AI prompts.

💡Rate Limits

Rate limits refer to the restrictions on the number of requests that can be made to an API within a certain period. The script mentions adding a 5-second sleep timer in the automation to comply with Perplexity's rate limits, preventing the system from sending too many requests at once and potentially crashing.

💡Supporting Evidence

Supporting evidence is the information or sources that back up the accuracy of a fact. In the video, the AI News Fact Checker not only checks if a fact is true or false but also provides supporting evidence for its verification. The script describes how the system uses Perplexity to find and include this evidence in the final verified article.

💡Webhook

A webhook is a way for an app to provide other applications with real-time information. In the script, webhooks are used to trigger actions in the AI News Fact Checker system, such as starting the fact-checking process when a new article is added to Airtable.

Highlights

Introduction of an AI news fact-checking setup designed to verify the accuracy of information generated by AI, specifically addressing issues with Chat GPT's factual reliability.

A blog owner's challenge of manually fact-checking articles written by Chat GPT due to inaccuracies, highlighting the need for automation in the verification process.

The demonstration of an automated solution that integrates with Airtable and Make.com to streamline the fact-checking of AI-generated articles.

Explanation of the process where the AI-generated article is broken down into individual facts for verification, showcasing a method to handle complex information.

Utilization of Perplexity AI to search the web and verify each fact, illustrating the integration of external APIs for enhanced fact-checking.

The presentation of a verified article with corrections and supporting evidence, emphasizing the system's capability to provide accurate and updated information.

A step-by-step guide on building the AI news fact-checker, offering educational value and transparency into the construction process.

The creation of an Airtable database with 'Articles' and 'Facts' tables, detailing the initial setup for the fact-checking system.

Description of fields and views setup in Airtable for organizing and managing articles and facts, providing insight into database structuring.

The use of automation within Airtable to trigger fact-checking processes, demonstrating the practical application of Airtable's automation features.

Integration of a webhook to connect Airtable with Make.com, illustrating the technical setup for inter-service communication.

The setup of a Make.com scenario to handle the logic of fact-checking, including the use of GPT for text processing.

Implementation of an iterator to handle the verification of multiple facts, showcasing the use of looping mechanisms in automation.

The parsing of JSON data to manage fact information effectively, highlighting the importance of data formatting in automation workflows.

Use of HTTP requests to interact with the Perplexity API for fact verification, detailing the technical interaction with external services.

The final compilation of verified facts into a final article, emphasizing the system's output and its practical use for content creators.

A call to action for viewers to subscribe and like the video, indicating the creator's engagement with the audience and desire for community growth.

Transcripts

hey I hope you're having an absolutely

incredible day today I wanted to run you

through this new setup that I built

which is an AI news fact Checker now the

thing that's really cool about this and

I'll just pull up the make.com scenario

to show you what it looks like the way I

can explain this the best is through a

little story so I was introduced to

someone recently through mutual friends

and he runs a Blog where he makes a

considerable amount of money and most of

the blogs are written using chat gbt so

he'll use chat GPT to come up with ideas

and also to write the majority of the

article and the issue that he was

encountering is that not all of the

facts from Chad gbt from what it was

creating were actually real and some of

these were fake or some of them were old

based on what chat GPT has been trained

on and so he was spending a lot of his

time manually going and having to search

up this information on Google and so I

told them that there is a solution to

this and it can be fully automated and

so here it is and so firstly I wanted to

show you how it works and show you the

output and explain to you a little bit

of what's going on and then we'll go

ahead and build this out ourselves I

think that you will learn a ton if you

actually follow this to the very end

because it's pretty complex there's a

lot going on and it'll teach you a lot

about make as well as a lot of uh air

table so you you'll be learning some

basic strategies all the way up to some

more complex strategies and you'll also

get my thought process behind how this

was actually achieved so let's say he

takes his article he wants to be able to

fact check the entire article make sure

that what he's posting is relevant as

well as up-to-date information right

with what is currently available now the

issue is is that when you have a 2word

article chat GPT is really good at the

beginning and the end and so all of this

information in between kind of gets lost

and so you need to break up the entire

article into different facts and so what

I did is you paste in your article here

you click fact check it's going to run

it through this so essentially first it

comes up here it splits it all into

different facts it is then going to Loop

through these facts upload them back

into the fact database so here you can

see that it's split through every single

fact this was another one that I had run

right you can see is all complete here

and it's just going to check every

single one so that's when it comes down

through the second part of the

Automation and it is doing a request out

to perplexity to verify this information

and it verifies every single fact

individually through perplexity so it's

actually searching the web and

perplexity is doing that research for

you and then it goes and it re-uploads

it back into air table and you end up

with a final verified article that you

can see here so before we go any further

I'm just going to show you how it works

and we're going to run it right now so

we're going to back out of this

automation here and now we are going to

click Start faction now this is going to

trigger the automation within the air

table now if we come back here in a

second we should see that the execution

has started running it has triggered the

watch responses air table trigger and is

going to start splitting it into facts

if we hop back into our air table and we

click into the fax database we should

start to see all the facts starting to

be populated right here the scenario is

running boom it split the facts and now

it ising through all the facts and you

can see them start to get pushed in and

the accuracy is pending with the fact

check status in progress and with every

single fact it's going to tell you

whether the fact is actually true or

false and then provide you with

supporting evidence as to why based on

the research that it did so now you can

see it's looping through the second

portion right so right down here it's

looping through and it's doing the

research within perplexity and then we

come back into the generate article you

can see all of the different facts that

have been verified you can see that

these first four were all false where

the fact correction is and why it is

false and then lastly it outputs the

final article as a verified article it

marks the fact check status is done and

you can read the entire article and you

can view the changes that it made

yourself to make sure that it is

actually accurate all right so you can

see here you know if we looked at the

first one elvenwood is home to the

Redwood Forest Reserve where towering

ancient trees reached towards the sky

and then we see that it's marked it as

false and if we read the fact correction

the Redwood National Park in California

so it's not actually elvenwood it's the

Redwood National Park it makes the

changes based on the information that it

found and then it also provides you with

supporting evidence stating you know it

mentions elvenwood is not a real

location it seems to be a fictional

artistic reference as seen from this

book real locations for this are Redwood

National Park in California so it makes

the changes it does all of this research

for you so that you don't have to and

then it outputs the final article so in

my opinion that is incredible and now we

are going to dive in and learn how we

can actually build this out ourselves so

I hope you're ready this one's going to

be a big one and this one is super

awesome it has a real world application

there's someone doing this right now

making a ton of money and I'm super

happy to be sharing this with you I've

worked on it for the last couple of

weeks you know really iterating it it

used to have a lot more steps I tried to

simplified as much as possible and this

is where I ended up so let's get started

so the very first thing that we're going

to want to do is create a new AI news

fact Checker database so you're going to

go ahead create a new database from

scratch right so I just called mine AI

news fact Checker and then you're going

to want to have two tables the first

one's going to be articles and the

second one is going to be facts now

within the views I just created one view

so you come down here you click on new

grid and that'll give you the view that

you want and you can just retitle that

all articles and then within this you're

going to want to add a bunch of fields

so we are just going to Loop through all

of these fields now and we are going to

build it out from scratch together

before we go any further and we run

through all of the different columns

that we have and how it's all linked I

just wanted to mention that I do have a

school Community where I provide all of

these make automations for you also have

weekly calls and anytime you have any

questions whatsoever you can reach out

to me personally and I'll be able to

help you walk through and debug your own

issues and really get you from where you

are at to where you want to be so the

community is 60% off right now as we're

new and as we're growing and as I want

to make the community as best as it

possibly can and in order to do so once

we hit 20 community members it'll only

become 40% off and it'll continue like

that until it is full price and you'll

be locked in at a lifetime price of 60%

off and what I'm going to be building

out in the future I haven't seen anyone

else doing you'll be able to research

your niches pull in the best performing

posts of their data and then also be

able to create your own post all managed

within airtable and make.com and I'm

super excited to be sharing this with

you and building this out and sharing my

journey along the way so for this you

are going to need to have the article ID

which is just a simple ID Auto number

field I called it article ID you come

down here you select auto number and

that way every single article will have

its own unique ID you're also going to

want to have an article which is going

to be long text that's where you'll be

pasting in your article to have the

information verified you're also going

to have this AI prompt so it's also

going to be a long text and then the

text that we have here is split the

following article into logical chunks

based on ideas or facts we're going to

format the result as adjacent object

with an array called pairs now each

object within the array should have

three keys so we're going to have the

original which is going to contain the

original chunk fact we're going to have

a fact check which is going to be set to

pending meaning that the fact hasn't

been verified and it's either going to

be true or false and then we are going

to have correction which for the time

being should be said to null and if it

is false then it's going to have the

correction uploaded in there so we also

do not want to include any delimiters

like Json around the Json output right

so we want it in the pure Json format

and lastly here is the article and then

we are going to add a checkbox the

checkbox is going to be start fact check

and that is the style that we are using

we're going to have the fact check

status and there are three different

fields so you're going to come in here

you're going to choose single select so

you're either going to have not started

in progress or done and the default

option is going to be not started so

whenever you create a new article it is

going to automatically be in the not

started status next we are going to have

a record ID the record ID is going to be

the formula and then the formula is

simply going to be record ID with both

parentheses we are going to have the

total fact so this is essentially what

allows us to determine if the uh fact

check status is done or not right so we

have the the total fact and the total

fax is going to be count of the fax link

so I'll show you how to set that up in a

second then we're going to have fax

verified which is a count once again of

the fax link and we're going to turn

this on so only include linked records

from fax where fact check status is

complete once again don't worry about

that for now you can just go ahead and

set them up and choose count but we

won't be able to select the source yet

until we create the fact table and then

we are going to have the ready and this

is the formula which is saying like if

total fax is greater than zero and the

total fax is equal to the facts verified

give it the status of one otherwise keep

it at zero so essentially when these two

are equal put this to one otherwise if

all the facts haven't been verified yet

then the status is not ready because the

article is not done lastly I have two

other hidden Fields here right now which

is the fax linked and the compiled fact

so we'll get to those in a second for

now let's hop into the fax View and one

other thing that's important to note

that I forgot to say is you want to

group by the article so that every

single article that you create is going

to have this nice little dashboard now

if we hop into facts right I have two

grids so you're going to want to create

two grids the first one you're going to

click here you're going to click grid

the first one is going to be all facts

and then the second one is going to be

generate article so if we open up the

all facts it's got a pretty similar

setup to the Articles table so first

you're going to want to have a fact ID

which is going to be an auto number next

thing you want to have is the original

fact right so this is where your facts

are all going to be split so that is

just long text go ahead and save that

then you're going to have accuracy

accuracy is also going to be long text

then you're going to have fat correction

also long text and that's where if there

is a correction so accuracy will

initially be set to pending and then

once it is run through perplexity it's

either going to come back as false or

true and if it is false then you're

going to have a fact correction here and

then you're also going to have

supporting evidence which is going to be

long text as well so fact check status

is a single select and there's only two

options which is either it's in progress

or it's complete you are also going to

have a record ID so once again this is

going to be a formula with the record ID

you are going to have an article link so

now we're going to make our first link

right so you're going to go link to

articles so you go link to another

record and then you're going to select

the Articles table and that'll create

the article link and then I also have a

few other fields here that are hidden

I'll unhide them so that that you can

add them as well but you have the

article which is from the article so

that's a lookup field and we are going

to be looking up from the article link

that we just created and then we are

going to be looking up the article so

that's important because that'll allow

everything within our make to be

processed at once so what that means is

let's say you have 10 different facts

they're all going to be under the same

article which is the article ID so let's

say article ID is one and within article

ID 1 we have 10 different facts that are

created well this is going to link them

all together here and then we also have

ready which is another lookup field and

we're looking up in the article link

that we created and we're looking up the

ready table and then lastly we have the

fact display which is a Formula which is

essentially saying that if the accuracy

is equal to True provide the original

fact otherwise provide the fact

correction so this is is going to be

important when it comes to compiling

everything together at the end because

for every single fact we're going to

have the fact display and then we're

going to roll up the fact display so

essentially what's happening is if the

accuracy is equal to true meaning that

the the fact is actually true provide us

with the original fact otherwise if the

fact is false provide us with the

correction all right so I'm just going

to go ahead and hide those fields and

then I also have it filtered by the fact

check status so where fact check status

is not complete so as soon as it becomes

complete it comes off of all facts and

it goes into generate article and then

they are grouped by the article link so

that you get a nice view in this

article's all facts table now if we go

to generate article right so you just go

create a new grid generate article

that's what you can name it and then I

filtered by you know I want to show

records where fact check status is

complete and ready is equal to one and

then they are grouped by the article

Link in this table as well so that you

get this nice little dashboard now we're

going to hop back into articles and

within articles now we can create these

last three right here so if we go to

Total facts and we do a count we're

going to select the the count link the

count source which is going to be equal

to the fax link now in order to connect

it to that we are going to have to

unhide two more Fields so this is where

we're going to create our last two

fields that we couldn't before because

we had to create the fax table so I'm

just going we're going to go ahead and

unhide those now and so here we can see

the fax link so this is going to be a

link we're going to allow linking to

multiple records and we're going to call

it the fax link so essentially this

article is linked to all of these

different facts and if we hop into the

facts right 9394 we can see that all of

these are linked under this specific

article ID where these are the fact IDs

so we have that there and then lastly we

have the compiled fax which is a rollup

of the fax link so a roll up of all of

these fact and we are just joining those

values and we are separating them by a

space that's all that's happening there

now that we've created the fax link and

the compiled facts we can hop into total

facts so once again we going to select

count and we are going to select the fax

link as the count source so right we see

that there are six facts so it's going

to show us that there are six total

facts and then facts verified is also

another count we're linking it to facts

link again but we are adding in this

condition where the fact check status is

equal to complete so it's only counting

complete facts or facts that have been

verified and lastly if we look at ready

it is just saying if the total fact is

greater than zero and total fact is

equal to the facts verified then give it

a status one meaning that it's complete

all right and so now we have both of our

tables setup and if we go into our

automations there's a few automations

that we need to have as well so the

first one is going to be the start the

fact check right when I click the button

to start the fact check we need to send

that trigger into make in order to

create the make scenario so we are going

to have it so that when a record matches

a specific condition we are going to

connect it to the article's table and is

going to be when start fat check is

clicked and when that happens we're

going to run a script

the record ID is going to be equal to

the air table record ID and then we just

have this simple code go ahead and open

this up you can see here the name and

the value we we need to create an input

variable so we call the the variable

record ID capital I and the value is the

air table record ID and we are just

saying you know fetch this web hook so

we're going to go ahead we're going to

create a web hook and make now and the

action is going to be equal to fact

check so if we hop into the fact Checker

we click on watch responses right that's

the first thing we need to do so you're

going to create a new trigger it's going

to be the watch responses in air table

and right you would just go ahead You'

create a new web hook you'll see here

right that that web hook has a specific

address we're going to copy that address

to clipboard and we are going to paste

it right here so await fetch so you're

awaiting for that URL of that web hook

to get this trigger going and then we're

going to pass the record ID as well as

the action fact check so that's the

first thing you can go ahead you can

choose your record right so you would

just choose one of the Articles or

choose an article load in some sample

data there what I did to test it out is

I said create an article about a travel

destination with both real and fake

information please do not specify which

is real and fake just write the article

and then I copied all of that

information and I pasted it into the

article's table so I came in here I hit

plus I create created a new article and

I just dropped it under the article it's

going to start and then we want to

create a router it's going to have two

routes the first route is going to be

the fact Checker which is going to split

it into individual facts so for that you

want to set the label to fact check the

condition is going to be the action that

is passed from Air table so the action

in in this case is fact check and we're

going to set the text equal to fact

check so that when we hit the start fact

check button it is going to trigger this

automation of starting the fact check

and it's going to run this script and go

past this information and know to go

down this path now to test that once you

could just choose the record like we

just did and then hit Test action and

then come into to make and hit run once

and it is going to trigger that now next

thing we want to do is get that actual

record that is just triggered right so

that we can split it into facts so

you're going to go ahead you're going to

create your connection within air table

if you don't know how to do that and it

is your your first time it's really

simple you click on your account within

developer Hub you can then create your

own personal access token name it

whatever you want and then you just want

to make sure that you add all of the

possible Scopes you click here add scope

click all of them and then add the right

base right the right database so I just

have it set to all databases and then

you're going to go ahead hit save and

then that'll get created and then it is

very simple to just set it up afterwards

like that then you are going to want to

choose your base so we named ours the AI

news fact Checkers so you want to make

sure you select the the right table in

our case it's articles and then the

record ID is going to be the record ID

that was passed when we tested out this

trigger for the first time so now let's

grab that information the next thing we

want to do is update the record so once

again create your connection select the

Articles table grab that same record ID

from the first step here and then we are

going to update the fact check status to

input progress right so you can just

turn off mapping and then it is now gone

from not started to in progress so that

when we click Start fact check it gets

flipped into in progress status and we

will leave the rest blank ahead and hit

okay now the next thing we want is to

split this into Json so we're going to

create a chat completion we're using GPT

40 the role is set to user and then the

message content is you are an assistant

that formats text for Jason

compatibility your task is to process

the following article text by replacing

new lines with n and escaping double

quotes with one two three four five

slashes right return the formatted text

only here is the article text where I

paste the article and the article is

going to be from the record that we just

grabbed right the article and then we

going to ensure that the Json is

properly formatted without any line

breaks with within string values and we

don't want to include any delimiters

like Json or around the Json output set

the max tokens equal to zero after that

is done we're going to split the article

interax we want to create a completion

so we're going to choose the GPT 40 the

r set to user and then message content

is going to be the AI prompt right so so

we're going to split it into different

facts and different chunks with the

original the status and the correction

and then we are going to pass that

article here's the article Max token is

going to be set to zero hit okay and

then once that is done we'll get a bunch

of different pairs and we now want to

parse that Json right and so to do this

you would add these two steps we're

going to head back into the AI news fact

Checker we're going to test the action

again we are going to run this once and

then we'll get that information passed

down through here and then to parse the

Json we'll grab the result from the

splitter so you can see here's my result

right you can see this Pairs and then

you're going to have it split properly

into the different facts now once that's

done we need to iterate through all of

those facts and we need to pass that

back to air table so for that I just put

in an iterator and that is linked to the

pairs so you can see here it is now

split nicely into original fact check

and correction it was parse properly and

we're going to grab that iterator we're

going to grab those pairs we're also

going to sleep so I found if you put in

a 1 second sleep timer it prevents it

from all the records being updated at

once which sometimes causes it to crash

so having this sleep timer of 1 second

allows each individual fact to come in

without any errors and then lastly we're

going to create a record so now we're

creating a record in the fax table so

instead this time connect it to the fax

table and then from there from that

record we are going to pass the original

fact so that's going to be the original

from the iterator then the accuracy is

going to be set to pending right for the

article to be verified or for that fact

to be verified and then the fact

correction is going to be the correction

which is blank because it has not had

any chance to be verified yet we're

going to set the fact check status to in

progress and the article of the record

ID right is going to be the record ID

from the very beginning this allows us

to have all the facts contained within

that article based on the record ID so

once all of that is done and go ahead

hop back into here you're going to test

the action again you're going to hop

back into make you're going to run it

once and you should now have if you hop

into your facts table if you go into all

facts you should have all of your facts

here since mine have already been

verified when I did this example at the

very start you'll see that the generate

article portion is full but in your case

under all facts you should see all these

set to pending all of this should be set

to blank and you're going to have the

original fact the accuracy set to

pending and the fact check status to

impr progress now let's build out the

second portion of the fact Checker so

now that it's split everything into

facts and it can verify all facts

individually next thing that we want to

do is we want to get a record so same

thing we're going to come into air table

we're going to create a new Automation

and that's going to be fact check online

and so what's going to happen is when a

record enters the allfax view right so

when a record enters a view we're going

to select the fax view is going to be

set to alax we're going to run a script

so it's the same script that we just had

the previous step once again we're going

to create an input variable which is

going to be the record ID and it's going

to be the air table record ID and then

we are going to fetch the same web hook

we're going to pass that record ID and

this time the action is going to be set

to online verify so when we come back

here in the router when we have this

filter it's going to be set to online

verify where the action that is passed

from the watch responses is going to be

equal to online verify so now it knows

to come down this path and not the fact

check path once again we are going to

get the record we are going into the

fact Checker database we are going into

the Articles table and we are grabbing

the record ID from the very start so to

get this kicked off you'll want to come

into here choose a record choose any one

of the records that are going to be in

your table and then you're going to hop

into run a script and click Test action

and then you're going to run this once

and then your response will come through

here and you should be able to get that

information and the record ID now the

next thing that we want to do is parse

the text right so I was getting this

error because of the blank spaces and

how it was formatted coming from a table

into make it wasn't always properly

formatted and a lot of that had to do

with titles and then there would be a

space and then there would be the body

section of that article and when that

happens it doesn't allow you to have

proper Json formatting so what we are

going to do is put in this text parser

which is going to remove that and allow

it to pass through uh in proper Json

formatting so I just did sln as the

pattern and the new value is going to be

dou sln in instead so essentially what's

happening is this is the search pattern

that it's looking for and then it's

replacing that pattern with this new

value and then we are going to want to

check Global match check that it's case

sensitive because it's always going to

be lowercase n and check the multi-line

as well single line you can set to know

and then the text is going to be the

original fact is created within your

table hit okay next thing because of

rate limits with perplex we don't want

to send all the facts at once so we add

a 5sec sleep timer and this is just due

to perplexities rate limits with their

API so next thing we want to do is make

an HTTP request to perplexity so we are

going to have this as the URL https API

perplexity completions we're going to do

a post request so we're going to be

sending information to them for that to

verify we are going to set the header to

accept application Json as the value the

second header is the name is going to be

authorization and then it's going to as

the value is going to be ver Bearer and

then your API key that you're going to

get from perplexity lastly the third

item that you want is the content type

as the name and the value as the

application Json for your third header

then we are going to have the body type

set to Raw the content type set to Jason

and then this is the request content so

here is the model that we are using for

this specific case the role of the

system is really to ensure that the

information is up to date accurate and

relevant right because we are verifying

all these facts so that's what we're

asking the system to do and then the

prompt that we're passing it so we're

going to pass the role as the user the

content is going to be you are

factchecking assistant your task is to

verify the accuracy of the following

statement and provide a detailed

response it is going to be the text that

has been parsed properly so the fact

that has been parsed properly and this

is is why it's so important to have that

in there because if you don't have the

text pared properly then this Json

request will not work because there's

going to be spaces in it so your

response should include whether the

statement is true or false supporting

evidence or sources and a corrected

version if needed and then format it in

the following response in the following

format accuracy true or false supporting

evidence provide specific evidence of

sources here and then a corrected

statement so if the original statement

is false provide a corrected version of

the entire original statement with

accurate information if the statement is

true leave this field know so once that

is done we are going to have another

completion created within chat

GPT I would recommend that you hop in

your fact check and you test the action

you hop back in here you run a once and

you make sure that you get a status code

of 200 for perplexity because otherwise

the rest of the steps won't work so once

you get a status code of 200 then you

can proceed onto the chat GPT step

and before going any further I'm going

to be sharing this request content with

you so that you have it and you can just

copy paste rather than having to do it

all yourself and then we're going to

have the create a completion within that

GPT so we're going to set the role to

user and then we're going to have the

following prompt so format the following

fact checking result as adjacent object

with these fields now accuracy

supporting evidence correction remove

any prefixes right so we want to get it

into the proper Json format so that we

can split it again using par Json and

then pass all those values back into our

air table and then the factchecking

result is going to be the data

choices message content so it's going to

put that into proper Json formatting for

us and then lastly we are going to parse

the results from chat GPT so you just

come in here you select the result it

should be parsed properly based on the

prompt that we gave it and then lastly

we are going to update our air table

record So within fax you're going to

grab the record ID now the record ID is

not going to be the original record ID

within air table we're going to want to

grab the air table record ID make sure

you choose this one down at the bottom

from the following step right here and

then the record that we're going to

update is the accuracy so the Json has

now been parsed properly it's going to

be the accuracy under the accuracy the

fact correction so whether the fact is

true or false is going to be the

correction and then the supporting

evidence is going to be supporting

evidence which is going to be pass right

here we're going to update the fact

check status to complete so make sure

that you turn this off and select

complete when this is switched into

complete it'll also trigger the number

of facts that are verified and that is

it for that and you want to go ahead and

save and you can go ahead and test it

again test the action and run it once

and have it run through this entire

prompt and if it's done properly you

should end up with one of your facts

under the generate article with the

accuracy either set the true or false

and if it's false it should have a fact

correction with supporting evidence now

I do have one other automation that's

important and that is when the fact

check is equal to complete so lastly

we're going to to set one last

automation within air table which is

when a record matches the condition

where the article's table as the

verified article is not empty so

essentially what's happening is when the

verified article is not empty right

we're going to update the record of the

fact check status of the articles with

the record ID and the fact check status

is going to be equal to done so now what

that'll look like on your main page

within the AR articles it's going to

trigger and say hey this article is now

complete and your verified article

should pop up here now you should be

able to turn on all of your automations

as well as turn on your make scenario

and back in here you can paste the

article that I'll have shared with you

and you should just be able to click

Start fact check it should run

completely in the background and output

a verified article that has been

searched has has found the latest

information updated it where it could

and output a completed and verified

article that is correct now this took me

a ton of time to build out and I'm super

psyched to be sharing it with you I hope

you got immense value out of it and if

you did I would really appreciate if you

dropped a like And subscribe it's free

and it really helps my channel grow and

allows me to continue making videos like

this so I hope you have an incredible

day and I'll catch you on the next one

peace PE