# How to Get Free PR Using Make.com & AI

## Метаданные

- **Канал:** Nick Saraev
- **YouTube:** https://www.youtube.com/watch?v=dMcIhQ2v8lE
- **Дата:** 26.02.2024
- **Длительность:** 43:11
- **Просмотры:** 5,889

## Описание

Note: HARO has since been discontinued. You can do the same thing with their new service, HERO. 

In this video I'll show you a simple AI flow you can use to automatically reach out to reporters & journalists on HARO (Help A Reporter Out).

WATCH ME BUILD MY $300K/mo BUSINESS LIVE WITH DAILY VIDEOS ⤵️
https://www.youtube.com/@nicksaraevdaily

JOIN MY AUTOMATION COMMUNITY & GET YOUR FIRST CUSTOMER, GUARANTEED 👑
https://www.skool.com/makerschool/about?ref=e525fc95e7c346999dcec8e0e870e55d

WHAT TO WATCH NEXT 🍿
How I Hit $25K/Mo Selling Automation: https://youtube.com/watch?v=T7qAiuWDwLw
My $21K/Mo Make.com Proposal System: https://youtube.com/watch?v=UVLeX600irk
Generate Content Automatically With AI: https://youtube.com/watch?v=P2Y_DVW1TSQ

MY TOOLS, SOFTWARE DEALS & GEAR (some of these links give me kickbacks—thank you!)
🚀 INSTANTLY: https://instantly.ai/?via=nick-saraev
🧠 SMARTLEAD.AI: https://smartlead.ai/?via=nick-saraev
📧 ANYMAIL FINDER: https://anymailfinder.com/?via=nick
🚀 APOLLO.IO: https://get.apollo.io/bisgh2z5mxc1
👻 PHANTOMBUSTER: https://phantombuster.com/?deal=noah60
📄 PANDADOC: https://pandadoc.partnerlinks.io/ar44yghojibe
📝 TYPEFORM: https://typeform.cello.so/rM8vRjChpbp
✅ CLICKUP: https://clickup.pxf.io/4PQo61
📅 MONDAY.COM: https://try.monday.com/1ty9wtpsara2
📓 NOTION: https://affiliate.notion.so/3viwitl53eg7
🤖 APIFY: https://www.apify.com/?fpr=98rff
🛠️ MAKE: https://www.make.com/en/register?pc=nicksaraev
🚀 GOHIGHLEVEL: https://www.gohighlevel.com/30-day-trial?fp_ref=nicksaraev
📈 RIZE: https://rize.io/?via=LEFTCLICKAI (use promo code NICK)
🌐 WEBFLOW: https://try.webflow.com/e31xtgbyscm8
🃏 CARRD: https://try.carrd.co/myjz1yxp
💬 REPLY: https://get.reply.io/yszpkkqzkb8f
📨 MISSIVE: https://missiveapp.com/?ref_id=E3BEE459EB71
📄 PDF.CO: https://pdf.ai/?via=nick
🔥 FIREFLIES.AI: https://fireflies.ai/?fpr=nick33
🔍 DATAFORSEO: https://dataforseo.com/?aff=178012
🖼️ BANNERBEAR: https://www.bannerbear.com/?via=nick
🗣️ VAPI.AI: https://vapi.ai/?aff=nicksaraev
🤖 BOTPRESS: https://try.botpress.com/ygwdv3dcwetq
🤝 CLOSE: https://refer.close.com/r3ec5kps99cs
💬 MANYCHAT: https://manychat.partnerlinks.io/sxbxj12s1hcz
🛠️ SOFTR: https://softrplatformsgmbh.partnerlinks.io/gf1xliozt7tm
🌐 SITEGROUND: https://www.siteground.com/index.htm?afcode=ac0191f0a28399bc5ae396903640aea1
⏱️ TOGGL: https://toggl.com/?via=nick
📝 JOTFORM: https://link.jotform.com/nicksaraev-Dsl1CkHo1C
📊 FATHOM: https://usefathom.com/ref/YOHMXL
🛒 AMAZON: https://kit.co/nicksaraev/longform-automation-content-youtube-kit
📇 DROPCONTACT: https://www.dropcontact.com/?kfl_ln=leftclick
📸 GEAR KIT: https://link.nicksaraev.com/kit
🟩 UPWORK https://link.nicksaraev.com/upwork
🛑 TODOIST: https://get.todoist.io/62mhvgid6gh3
🧑💼 CONVERTKIT: https://partners.convertkit.com/lhq98iqntgjh

FOLLOW ME
✍🏻 My content writing agency: https://1secondcopy.com
🦾 My automation agency: https://leftclick.ai
🕊️ My Twitter/X: https://twitter.com/nicksaraev
🤙 My blog (followed by the founder of HubSpot!): https://nicksaraev.com

WHY ME?
If this is your first watch—hi, I’m Nick! TLDR: I spent five years building automated businesses with Make.com (most notably 1SecondCopy, a content company that hit 7 figures). Today a lot of people talk about automation, but I’ve noticed that very few have practical, real world success making money with it. So this channel is me chiming in and showing you what *real* systems that make *real* revenue look like! 

Hopefully I can help you improve your business, and in doing so, the rest of your life :-) 

Please like, subscribe, and leave me a comment if you have a specific request! Thanks.

## Содержание

### [0:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE) Segment 1 (00:00 - 05:00)

what's going on everyone welcome to another video in our cours make. com for people who want to make real money and in this video I'm going to be diving deep into how to build an automated backlink generation machine using a publicly available service called Haro which stands for help a reporter out now Haro shoots you emails once a day well actually three or four times a day they can be uh pretty overbearing with a big list of questions that journalists have developed and journalists are basically just like shopping around for quotes and so um I've built the exact same system that I'm about to show you and then sold it to a variety of clients anywhere from like an 800 us price point all the way up to maybe a 2 and a half to $3,000 price point uh as you'll see the build is extremely simple although the outsized impact of having something like this constantly running 247 uh is difficult to Fathom I mean you can get backlinks in fors Business Insider entrepreneur. com that sort of thing uh and I'm sure that vulnerabilities like this will eventually be patched but as of my latest knowledge you could still use this sort of system to you know just take yourself to backlink Heaven that sounds like something you're interested in and you want to learn how to build a business or SEO system that generates actual leads uh backlinks or money then this video is for you stay tuned and I'll see you in a sec okay so if you guys are new to Haro or help reporter out a brief rundown this is basically like a source website that allows journalists to get sources from business thought leaders and doctors and lawyers and all these authoritative people that you usually see quoted in like a magazine or whatever uh and we're just going to build a system that takes in these requests and then if it's relevant to us we're going to use AI to process it and then write an email response to the person that's requesting the source so very simple very straightforward from a bird's eye view uh but Haro is just like such a great system um this probably isn't going to be around for a while I received some notification I think like a year ago saying that they're going to uh slowly start changing the way that the system works maybe they're going to put it like in a portal or something like that uh you can imagine how the systems that we're going to be building for that are going to be slightly different from the email based system that I'm going to show you today but when they do that you know somebody leave me a comment and then I'll build out a way to do that as well everything in life is automatable um yeah and so this is an example of one of the flows that I built up before you'll see that a bunch of modules are everywhere that's just because I was testing it right before this video to make sure that it worked uh and this is an example the flow that we're going to be building out we're going to start with uh previously I used a web hook now I'm just going to use an email module we're going to set a bunch of variables then we're going to use AI to do things like collect titles and then write emails so that is what is in store we're also going to get to use the iterator module which is a module that I don't think I've explicitly shown or talked about until now uh but it's very simple and very straightforward you don't need to be too concerned about what we do with that so here is the new scenario first things first why don't we take a look at one of these har emails just so we know what we're working with let's do Har then do Har Friday evening Edition so these are the emails that they're sending out what looks like three times a day Friday morning afternoon and evening um as you can see it's just a massive list of queries uh people are in biotech and Healthcare asking for pros that you know they're seeking Pest Control professionals and Travel Health experts uh they want people that can talk on virtual reality and medical practice right there are all sorts of things and obviously there are a ton of requests here it looks like 93 in total not all of them are relevant so we're going to want to develop a system that can automatically find the ones that are relevant to us and then uh read through the descriptions and then respond and as you can see this is a really long email right like we're not even a quarter or a fifth of the way down we can scroll all the way to the very bottom and uh yeah it looks like the 93 oh actually the message was clipped so it goes on even further than that so I'm not going to you know open it up and read through this probably take until uh the end of time another asteroid would probably come and extinct us all before the time that I was done reading this email but there are a couple of things that we can pick up just pattern wise um that'll make it really easy for us to do this so as you see first they start with like an index and then they just have a bunch of oneline queries like hey seeking Pest Control professionals and Travel Health experts Anonymous or Brands what are your biggest marketing challenges in 23 blah blah if we scroll all the way down though you'll see that at a certain point that changes uh and then there's this big list of stars which I guess is just how they formatted the emails way back when Haro was first made uh and then you'll see that they actually like go longer on the specific requests so now there's a whole what looks like three or four paragraph section just about that first request which is seeking P control professionals and Travel Health experts and so what we can do is we can basically parse all the titles and then extract the relevant queries just from the titles like let's say if I'm Nick at left click and I work in operations and business Consulting you know I'm going to tell the AI hey I just want you to extract all of the stuff about business Consulting and then we can use those titles to basically split the email into the sections that we want so maybe starting from here down to here and then we can just feed this section into AI we can do some fun email parsing stuff to get the specific uh

### [5:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE&t=300s) Segment 2 (05:00 - 10:00)

email record that we want or email address that we want to send to and then maybe we can just write something like hey you know I'm a professional at bed bugs and I'm more than happy to help here's a quick quote right in that way it's like an automated quote generation system um obviously not everything that we send out is going to be 100% true there are going to be some issues with factualness and like veracity and that sort of thing so you do have to be careful when you're using this U but the system has generated tons of links for a variety of people uh and I find that you know if you don't really have the expertise that the AI thinks that you do people just aren't really going to respond or put the link up so very low impact in that way I have a bunch of prompt set up here which we're going to be uh drawing from hence why I have a Google doc window open but why don't we start with the very first problem which is just getting the email from our inbox now you can imagine how if you're building the system for the future like I am um we want to look at emails that are going to be coming in from now until Infinity but I can test using the watch emails module by right clicking and then press choosing where to start so oh I guess I have to set it up first so uh I'm just going to click and let's use this account I just know that is going to work my folder that I'm going to pick here um in your Gmail inbox you have a variety of folders this is something that most people don't realize but there's a difference between the inbox folder like the sent mail folder the all mail folder your unread and all that stuff this is just a quirk of how Gmail works so if you're using a different email um it may be a slightly different approach for you but I know for a fact that I just want to look at all mail that has ever entered my inbox or ever will enter my box um this is basically like setting up that mail hook web hook that we were doing earlier I'm just doing it using the email module because it's a little bit easier I would prefer to use web hooks though because you can typically respond a lot faster any who um okay criteria we're going to choose all emails send your email address this is where you basically Define the filter that you're going to use to search through emails so um if you notice if I just type Haro in my mailbox I got tons of these right har har so odds are um the email that or the subject line that you're curious about is just going to or the subject line that we're going to use to filter it's just going to be h o and odds are I'll probably do pretty well so I'm just going to filter by subject lanaro um I'll have no phrase and then I will not Mark messages as red maximum number results one click okay and that it'll ask you where to start um now we want to test this on historical emails so I've received a bunch of hor emails obviously so I'm going to select the first email down here and it's actually going to go through my whole email inbox do the filter that I showed you and then allow us to select where we want to begin so why don't we do this one here um this Friday morning one looks like that'll give us three emails to test then I'm going to run this module so now we're selecting that and you can see we have a bunch of fields here we have the date the email ID the size just a bunch of that we don't really give a crap about sender recipient right text content uh and text content is what we're going to use so text content looks just like the body of our email but it starts off with look what we found when we analyzed over 100,000 press releases so this looks to be just like some advertising that they're doing and then the query start and you see we have the formatting that we had before with index then we have what looks like a hashtag and then the name of the industry and then a bunch of titles right and so these titles are probably sufficient for us for our AI system to like automatically figure out whether or not it's relevant to us so that's what we're going to use uh and then we're going to do some fancy text parsing one thing that you realize in make. com is you get better and better is that like a ton of the really high value um operations that you can do or the workflows that you can build really just depend on your ability to use text parsing um and obviously we can use AI to do a lot of that text parsing for us especially in like less structured data sets you know when we're scraping a website or whatever where the data isn't going to be the same every time but you still need to have an understanding of how to do text parsing um and so you know the modules and the functions that are available to us like the split and the replace and the get I find myself using them basically in every workflow now so getting familiar with them uh if you want to build higher systems is pretty essential okay anyway uh what I'm going to do here is I'm actually going to take my own advice and I'm going to set a testing variable so I'm going to this is the only time I will ever use the set variable module on its own and I'll just call it test text the text that I'm a sticking here is I'm actually just going to go through this whole email and then just copy and paste all of this in just because I just want a way that we can test our flow without me having to like explicitly grab an email every time so click okay all right so we have the text and then now that I verified that this email module works now I'm just going to start here and then I'm going to rename this to testing text just for my own sanity uh and now every time we run this scenario we're just going to get this as an output which is the exact same thing as we got the first time now we're just not querying our own email over and over and over again so assuming that that's representative of the data set what do we do next well the way that I see this working and I know because I've developed the system so I'm kind of cheating in that way but the way that I see this working best is first to

### [10:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE&t=600s) Segment 3 (10:00 - 15:00)

save our on tokens we pass in all of the titles and then we have ai tell us which titles are relevant to us and then in order again to save on tokens we pass in all the descriptions or we pass in the relevant descriptions and then we only have ai generate responses to the descriptions that are of course relevant to us and so the way that I would do this in practice is I would set multiple variables here and again I always use set multiple variables over set variables because there's just no reason why you would ever just want to have one for maintainability uh you always just want to be able to jump in and add more and then why don't I set a variable called um uh let's just call this like titles I guess right because we basically want to extract everything up until here so we'll just call this uh titles and then the variable value okay so let me just go through this and see this yeah this text is probably how we're going to separate the email honestly um let's just do that we could obviously separate it from here all the way down to here but it doesn't really matter I'm going to have like an extra 20 tokens or 30 tokens or something if I include the title so I'm just going to grab the very uh like the first half of this using split and then I'm going to feed in all of the testing text here and then I'm going to copy this long string of asterisks and you know what let's just make sure that this man how many asterisks is this oh I don't want to count this but let's just make sure this is the same in every Emil that's five that's 10 that's 15 that's 20 that's 25 so it looks like they're 28 asterisks I don't know man um I wonder if I can just test this and make sure it's the same over here let's paste this in okay yeah so it looks like it's finding the exact same number of asterisks good sanity check that means it'll probably work on all of the emails that we want to test it on um but I'm just going to split all the titles this way um and then let me just do this as like demonstration and then I'll show you how to get the first item so what happens if we split the um two things then we have an array now with the first half being all of the text up until that line delimiter or whatever and then the second half basically starts uh immediately after that so the first half is everything from here on up and then the second down basically so that's good that's our intended Behavior but what we want is we only want to grabb the first one and so you can do that and make just by using get which I've probably shown you guys 10 times by now but I'm just going to get the first and then I'm going to run once again sanity check make sure this works titles we now instead of having an array we just have a long string and looks like it goes down to 33 and then it stops I don't know if that's just like make um like having uh wanting to like cut it early or whatever as opposed to keeping going so let me just make sure that this is yeah okay so it looks like the titles that we fed in were pretty light so there were only 33 in the testing text so that makes sense so it goes all the way down to 33 uh looks like that other email that I got where we got 90 queries or whatever that was sort of weird but uh yeah that looks good to me okay now we're going to set descriptions we're going to do the same thing so we're just going to say descriptions and then I know for a fact this is going to work because we just worked uh set it up on the first uh so now I'm going to separate this and then just press run once and then get it all the way down starting at the top working my way down you can obviously just like um use one variable and then have an array to index it into two but I don't really care about that um and you know I'd rather have like variable names like title description it would just be way easier okay great so yeah looks like we got what we want we can obviously like separate it so that it goes all the way up to here I don't really think it's super relevant or important um when we separate the text it'll probably be okay but I am noticing that every description is separated with this line so that's like a good uh this is a good instruct instructional video sorry on text parsing just because it looks like help a reporter out is extremely structured in the way that they do things all right so now it's time to introduce a into the process we have a big list of titles what are we going to do next I'm going to go down here to open AI then I'm going to use the completions endpoint and I have my own setup so I don't have to go through the connection rigal on my own I'm just going to use gp4 for now this is probably in or suboptimal from a token perspective because gp4 is going to charge you like three cents per a thousand tokens something like that uh but you can imagine how the positive impact of getting one of these probably vastly outweighs the money that you'd spend on the tokens now I have my prompts set up elsewhere so I'm just going to copy these in and the reason I did that in this video is because uh these prompts took a little bit of trial and error and I just want to make sure that the video isn't three hours of me just coming up with

### [15:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE&t=900s) Segment 4 (15:00 - 20:00)

promps but I'll read through all of these after I'm done copying and pasting them in just so that we all know where these prompts are coming from and then some of my thought processes in developing them so let's paste in this one here for user that looks pretty good and then let's do another one for assistant which will be an example of the correct query that I want to Output all right last thing I'm going to do is because I'm sort of training it here now I'll just have the user prompt I'll feed in the titles for advanced settings uh Max tokens of 300 I think that should be okay but let me just test and see how long this is I'd always just use like an online word counting Tool uh 58 words yeah 300 characters is probably going to be more than enough maybe we can even make it shorter let's just do 150 temperature I'm going to choose a low temperature for this because I want it to be very um as less creative as possible all right so what exactly do these prompts do so the first is a system prompt that says you're a helpful intelligent Administrative Assistant this is just sort of a catchall and it makes it identify as something that is smart uh next up is user prompt and this is where you would put in information about you because of course we're getting AI to filter these titles based off of something so they needs to be that something that we're filtering by so in this case I'm saying I'm a business owner specializing in AI marting software development and design below is an email with a list of titles all prepended by a number and the right bracket with a right bracket symbol filter them based on relevance and if the title involves the sorts of things I work with on a daily basis if it's in my area of expertise add it to a Json object that returns only the list of associated titles don't include the company names or the number then I say use the following Json format looks like this where I have a right curly bracket with an additional space it looks like and then I have a variable called relevant queries colon and this is just like an adjacent format um with a nested array so then I have an array with element A B C D where a BCD are queries IR relevance now when you provide it with con with instructions this complicated it's usually good to give it at least one example the number of examples that you give have to be weighed against the amount of money that you're spending per request so I re rece three emails like this a day I know mathematically that means I'm probably going to be spending 10 cents or something like that per run maybe 30 cents a day in total if you add that up over the course of 20 business days that's like $6 for me that makes sense I'm more than happy to spend $6 to have a system that's constantly responding uh and generating me backlinks but for you that might be a little bit different and so you sort of have to you know weigh this by what the perceived cost is and then what the perceived benefit is to you the next prompt is a user prompt and this is where I'm just going to give it a user assistant pair example so I start off with just a giant list of titles and I went very extensively here because I wanted it to do a good job so I added 130 titles I got this just from a couple of other um horo campaign so I copied that in and then I tried to have it model the response format that I wanted and so that's what we did here relevant queries here are a list of the titles which all look good and then after that we have another user prompt with the actual titles from the actual data and so this is where we're going to get the list of titles that are actually relevant to us now we have everything that we need to actually test this thing so why don't I try and click run once that looks good to me it's running it's gp4 which historically takes a little while longer than GPT 3. 5 or gp4 Turbo okay and if I go down to choices one message you see that we have a list of titles it looks like there's some weird formatting going on over here not really sure where that's coming from but that might just be the person that wrote that title any who buy build or fix what's best for first-time home buyers nearly half of Americans have considered starting a business what's stopping them the importance of financial planning for a secure future return to office and corporate sustainability goals careers now handle anxiety on new job AI to job hunt cell phone repair experts need the One charging s that's draining your battery I'm probably not the right fit for this one um but regardless you see how it's pulling in a bunch of things that are significantly more relevant than you know seeking somebody to help me figure out where the prostate gland is right like obviously I'm not going to be able to provide you guys information on that um okay so I'm going to have this be extract relevant titles I'm going to call this set title and description and let me just set this up how to reach

### [20:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE&t=1200s) Segment 5 (20:00 - 25:00)

out to Haro queries automatically and build a backlink Empire feeling kind of clickbaity today so now that we have a list of relevant titles what do we do next well what we have to do next is we have to turn that list of relevant titles basically that we have over here into um or we have to use that to split the text that we're interested in the description side so that we can then feed that into um our Ai and the way that I did that here was I had that parse Json module of course but then I had an iterator and the way that the iterator was set up was I just dumped through all of the relevant queries and then I used that to um do a bunch of text formatting which you're going to get into where I split the descriptions and do all that fun stuff so I'm going to do exactly that here the first thing I'm going to do is I'm going to add ajacent module on the end the reason why I'm doing that is because I just want to parse the output of the AI module so I'm going to add this as my Json string I'm going to go message and then content and this is just going to make it accessible in make. com you'll see that I get a warning here because anytime a Transformer is the last module in an input uh it'll just say hey this shouldn't really be happening this isn't really doing anything of value to you and then what I want to do is I'm going to Let's actually just run this once you guys can see what the data structure looks like it's doing its thing I am calling the GPT 4 API every time that I run this test so it's costing me like a scent or two but you'll see that um after we process this we actually end up with a relevant queries variable and then all of the titles are basically in just an array here which is nice so what we would do with this information is we would iterate over this now if you've never used an iterator before use an iterator when you basically have an array of stuff and then you want to run a new scenario a mini scenario for every single one of them so in my case if I didn't use an iterator here then I would have to work with this whole array but if I use an iterator I can actually do a bunch of stuff specifically for every title in this array and so you can think of it as if I use an iterator here um everything up until the iterator is going to run once and then nine more times it's going to go through every title and then do stuff with that title so I'll give you a quick example here I'm going to go to the tools down here and then go to iterators and click iterator and you'll see it's asking me for an array um the array that I'm going to use is this relevant uncore queries array and you'll know that it's an array in make. com when it has two square brackets one to the left and then one to the right the reason why we click the array is because the array has just a giant list of the things we want to iterate over and then the format in make. com is you always do like the highest level array and then you stick that in like we wouldn't press this one or whatever the formatting here confused me for quite a while but you always just want to use the highest level U variable in the array so anything with that left and right square bracket should be good now the iterator I believe is a Transformer as well that's why we got that little excuse me one to the top right and you'll see now what's happening is we fed it as an input a an array with a list of values and then what happened afterwards was we turn that into a list of bundles instead and so a bundle is just one whole scenario execution basically and so everything after this point is going to run per bundle so they'll be you know if I added like a Google Sheets add a row module here then it would run once for this another time for that a third fourth time for that it would just keep on going until all the bundles were exhausted essentially so what do we want to do now that we're iterating over every title well now for every title I want to extract the specific summary that's relevant to us and the way that I'm going to do that is I'm going to use a another set variables module and I'm just going to call this um specific request let's do that and here's where we're going to get into like some text parsing magic so I have the title over here in uh the value parameter where it says nearly half of Americans have considered whatever and then I'm going to go back to this testing text just see where that is nearly half of America okay so this is in the title section so we don't want the title we want the description t section okay great so you see here that the way that we need to do this is we basically need to parse this and then we want to grab all the text from here down to here and so then we'll have this text available to feed into Ai and then create like a customized response to you'll see it has a bunch of information we probably don't need like the deadline probably going to respond to this within 7 Seconds of it being posted so that's probably not relevant the Yahoo finance media Outlet part might be relevant this email is relevant because this is where we're actually going to like write the email back to but basically we need to extract this and then the specific part or the specific

### [25:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE&t=1500s) Segment 6 (25:00 - 30:00)

um uh text from here to here and so the way that we're going to do this if you think about it logically kind of like in make and you know eventually that's going to be this so we're going to work up to this right now what we have to do is first we have to feed in the description that we isolated earlier the whole long description with like 93 or whatever um sections of text like this and then we have to separate it using that title that we've extracted previously and once we've uh split it using this title we'll then have all of the summaries before and then after and so what we need to do then is we need to get all of the summaries after right because the after part is going to include all of this it's basically going to be from here all the way down to the bottom email and then we have to take that and we have to split it again based off of this and then what we want to do is we want to get the first result because the first result will be from here to here so if that doesn't make sense I'm going to run this a bunch of times just so you can see how we iteratively get to the variable that we want to get to um but that's how we're going to extract this text specifically and I sort of hope that I have at least one or two errors here I mean I'm not going to do anything purposefully but if I do then that'll hopefully allow me to walk you guys through this in a little bit more detail especially for those of you that aren't really familiar with how this works okay anyway so uh I just got a little bit ahead of myself what am I going to do here so first thing I'm going to do is I'm going to split and then the varable that I want is descriptions and the what I'm going to split it by is the value that came out okay so that's the very first thing that we're going to do oh and then uh let me just actually no let me do this just so you guys can see everything at every step and it's going to run you know presumably seven eight or nine times however many titles that it extracts and then passes through the iterator so we're going to get to see that happen so there you go ran one time in the iterator and then seven times down here so we actually split the text in seven different places and you'll see that just like we had before with the titles and descriptions we now split it into two sections we split it into the section before that title which goes all the way down up into the summary we split it after and it includes all of the information that we want the name the category the BL blah all the way down so that's great that's exactly what we want now what we want to do is we want to grab the second value here and then we want to split that second value by this you can of course do this in separate modules if you wanted to um but in practice you know you get good enough at working with this that you can usually just stick it all into one so now I'm going to get the second response or the second result of that so let me run that one more time and just waste all of my value BL Ops we just need to make sure we don't overwhelm the gp4 API now in my case I have like a million requests that I can make per minute cuz I'm part of their highest plan but in your case you know make sure that you're just not sending you're not re rerunning this every 5 Seconds let's say okay now what do we get after well because we got the second result uh we have only the section that we're interested in and then um it goes all the way down from there to the bottom of that email so now what we're going to want to do is we're going to separate it all we're going to split it again and we're going to do it by this string and then we're just going to get the first result so I'm going to go back here and I'm going to split that again and what I'm going to do is split it by that then I'm going to get and the first result now we're going to run this one final time and this should work and get us the list of descriptions that we can then use to feed into another AI module to have us automatically write so let's check the output and it worked like a charm we got the name of the person that requested it the category the email blah blah blah now the last thing that I want to do is and this may seem unintuitive but I want to feedback the title and the reason why is because in my head I have a feeling that if I that if uh the text doesn't include the title it'll just be a slightly lower quality so what I'm actually going to do back here is I'm just going to write value and then I'm going to do space and now it's just going to take everything that I just did all the text processing and then it's just going to add or concatenate uh the title itself to the beginning of that which should be every that I need and you know what I'm also going to go in and then I'm going to add a new line as well so I'm going to add value then new line and then I'm going to add all the other information okay let me just save this as a sanity checks my computer's been acting up recently and let's see how it works got some French hazelnut coffee that stuff is the bomb okay operation s specific request elevate your e Commerce unleashing the power of magenta development services and then category so this is the title

### [30:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE&t=1800s) Segment 7 (30:00 - 35:00)

uh of the request that we have now concatenated and I think this is just going to be a little bit better now you'll see a lot of these queries ask for stuff like a headshot a bio and a high Authority website or social media link for attribution purposes so we're going to have to find a way to put that into our uh into our AI module but now we have everything that we need to do to actually like move forward with this so um now I'm going to stick in I'm just going to copy this then stick it back in going to rename this and I'll say write customized email and then I have to edit all of these prompts of course because these are the prompts from the extract relevant title section but I have them right down here in Google Docs so that's good I think this is the same system prompts I'm not going to worry about that I am going to do the user prompt though this is where the money is at um when you paste in it has the nasty habit of just pasting all of your text in with no new lines so if you want to add the new lines which I find very important for forign formatting and tone and professionalism you have to hold down uh shift when you do your command V or control V and that'll paste without formatting you also right click and there should be like a paste without formatting or maybe paste and match style in Mac I don't know but the user prompt that I'm going to use now is hey I own one second copy it's a successful AI marketing company that does over 90k a month my name is Nick surf here is some of my information don't use it unless asked I then have a bunch of links to my socials um if I were you I would go down every single one of these and just follow me on every platform immediately uh if you don't the whole world is a stake um you'll see how one of the things that I provided is a headshot because I knew that a bunch of these were requesting headshots there's my own business website my Twitter my Instagram LinkedIn if somebody asks for these directly I just want the AI to add them in then below is a request by a journalist for outreach write a succinct Spartan email respond to each query if needed be concise and then use the following format it's important that you give it a following format and then you just like train it to use that exact format because if you don't then it'll just sort of wing it every time and sometimes the quality will be good but most times so it's better for you to walk its hand a little bit and this is worth the additional token cost I'm actually giving it an email template saying hey I own this company is successful marketing company that is over $90,000 a month your customized responses go here we'll be in the office for the rest of the day if you need anything just shout and I'll take care of it the idea being that I'm implying that I'm around and I can help uh very quickly now the next thing we need to do is we need to like train this in context and so I'm just going to paste in a huge query here copy this from here and then paste this in and this is basically like a like an unstructured um request and you'll see it doesn't have new lines or anything like that but that's because new lines take some tokens and I don't really I don't want to worry too much about um adding a bunch of those uh just in the training example because I just wanted to focus on the content not the way that it's formatted sometimes the formatting will be a little bit different anyway I paste this in and then I give it an example of the email that I wanted to write and this is really the most important one this is the longest one as well because I'm just showing it hey here's the relationship between the input that I'll give you and the output and what does the output say it says hi and then it grabs their first name which is somewhere in here like Puja I believe I own the company sucessfully marke with 9k a month and then here are a bunch of answers to their specific questions that they're asking right specifically we are seeking your input on the following questions 1 2 3 blah blah so I'm actually going in that I'm answering these questions which uh is just going to train it to basically you know if there are any questions it's going to know that it needs to respond to them in a customized manner then I add a bunch of information most of this information was explicitly requested so I stuck that in and then I add in my final uh response so after that what we do is we feed in the specific requests from here and then this is longer so my Max tokens will not be 150 why don't we go to word counter and see 256 words one token is equal to about 0 0. 7 words so 1 over 0. 7 is 1. 4 so 1. 42 * 250 is 355 this is probably in the longer end of emails that I'm going to generate but just to be safe I'm going to go down here and I'm let's just do like 700 and then I'm going to add slightly more to the temperature because I want it to be a little bit more creative okay great so now we have the right customized email and then just to skip ahead a little bit I'm also going to um yeah I mean I don't want to run this a million in one times okay let's work on the email module as well and then yeah okay great so now that I have this I also obviously need to get the email address that I'm sending it to um if you pay close attention you'll see that the email address is actually nested right over here in um a phrase that's email colon space something at help reporter. net and so I can actually like extract that really

### [35:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE&t=2100s) Segment 8 (35:00 - 40:00)

easily um by just uh using creating a reg x one of my viewers made fun of me the other day for mispronouncing uh red reg gex it's really tough to do that off the top of my head gez uh looking for that specific query so I'm going to go back here and then I'll go to Reg x101 and then I'm just going to paste in all of this text here and just try and have it extract the email address using the regex okay so I'm looking for this email here query blah blah so because I know how to use regx I should be able to extract this pretty easily uh email space looks like there's going to be uh anything here at help a reporter. net and so this extracts this email I think that's going to work but just to be safe why don't I go and put in all of the testing text and see if it extracts all of the emails that I'm looking to extract okay you see that it's extracted the email address here fiq whatever this is their own internal system that they use to like route emails and stuff so that actually looks good what a fast regex uh we're going to parse this text and I'm just going to call this extract email and then once I get the output of that email which I think will work but who knows I'm then going to create a draft just selecting the wrong account here and then I'll write it in drafts the two email will be the email that I just extracted and I don't know if you actually need to put a subject line let me see what I did before looks like I have a couple of email modules here um I also attached this to strip myself okay so it looks like I just said rearo and then I added in some social proof saying that I'm loaded which I find works extremely well uh I'm just going to paste this in as plain text because that's how it's coming out of the model and we should also be actually no we don't need to do a Json U module here because I'm just the output is just plain text so I'll go to choices and then I'll go to messages and I'll stick the content in there and now this should produce uh a bunch of drafts so let's see how this works in the completed system I noticed that some of the emails as they were coming in were Mis formatted because Haro doesn't always stick to the same formatting whatever term internal system they're using to manage these sorts of things is good for the most part but there are some issues and so I put in a few blocks and those blocks were you know if the request doesn't exist basically which happened I think one in every 15 or 20 times don't proceed with the flow for instance um if another thing doesn't exist then don't do that right okay let's see if we could extract the email no it doesn't look like for some reason so that's unfortunate let me go down to choices and just see and make sure that the actual message was good hi Nicole I own this company here are my thoughts on the reasons people hold back from starting a business this looks pretty good so I think the actual like text generation parts are fine the reason why this isn't proceeding is because I added a block to the module that said if you don't find anything don't keep going and the reason why I did that for this specific case if you remember in the previous regx videos I didn't is because I uh I wanted it to proed even if I couldn't find the email in this case it's not proceeding uh because if it continues to go then the email module error out and I don't really want to deal with that so let me see why it was unable to find the specific email and question this to me looks like it should be good yeah so it should be able to find query at helper reporter. net so maybe there's something that I did here or maybe I didn't uh maybe it wasn't Global or something like that oh right my dumbass just didn't put the text in okay well that's pretty stupid I am going to globally match this and then I'm just going to return the first result uh and then for text I'm going to stick the email address in and then if the module finds no matches don't continue and then for safety to solve the problem that I was mentioning earlier I'm also going to add a filter in that says request found and I'm going to say this exists so the only way that this can proceed and actually call my precious tokens is

### [40:00](https://www.youtube.com/watch?v=dMcIhQ2v8lE&t=2400s) Segment 9 (40:00 - 43:00)

if a request exists if it doesn't exist then I just don't want to run this because odds are um you know it's just sort of pointless okay let's run this whole puppy from start to finish and I think that this should work we got our robot Minions on full swing extracting our relevant titles and then we're going to parse the Json from that iterate over it and we fed in what looks like nine responses and so this should function or should run nine times after the iterator so it ran the first time it found the request and now it's generating our email it's then going to hopefully extract looks good then it's going to go in and write that draft so two we stuck the right email in there the content is here and the best part of all is if I go back to my own personal email and then I check the draft I have it right over here Hi Megan I own this company I'm not a certified licensed real estate bank or lending or financial expert I can tell you from my experience in business owner that BL blah and then I also add in a bunch of my experience so yeah that system looks like it's trugging along nicely uh I'm surprised I didn't run into more issues while building that to be completely honest especially these higher level like text parsing flows where you need to put like a split and then a get and then but I think because we did everything iteratively like I actually when I'm building these out alone rather than just trying to do it all at once um I was able to sort of error correct along the way the last thing that we would need to do in order to make this module uh to turn this module from the scenario sorry from Theory and turn it into like a reliable practical business generate uh Revenue generating machine is we would have to sign up to harl and then we'd have to start receiving emails from connect this email module to the rest of the flow and then let me just stop this because I think it's clear that this works and then instead of testing text what we do is we just go down and then feed in the actual text content here and then you can rename this from testing text to set email plain text or something so I'm going to save this I'm going to export this as a blueprint for all of y'all that looks pretty good and yeah that's how you build a helper reporter out automatic Outreach system from start to finish in make. com don't think video took more than maybe 45 minutes again I've probably made five or six maybe $7,000 selling this system now and it's just one that's so simple for me to set up all I need to do now is just drag and drop the email address of the client into a couple of these folders I need to sign up for Haro obviously um but aside from that yeah that's how you do it so you can use this to automatically generate backlinks and build a whole empire if you want to you can also use this in a slightly different way from what I've showed you could write a much longer prompt spend a lot more money C iing into your specific experience and use case maybe put some case studies in do all that sort of thing but I'll leave all that up to you thanks so much for watching if you guys have any questions comments or anything like that leave them down below otherwise like And subscribe and I'll see you guys in the next video

---
*Источник: https://ekstraktznaniy.ru/video/12848*