Thank you everyone for joining. I have been very excited about this webinar and we have more webinars coming focusing on AI. I think this is a hot topic right now and we're looking into making it more practical, how to make AI more practical. And today we are speaking about localization of websites and blogs using automation and an AI.
So... Before we go into introduction, I want to kick us and put us in the right path. Magnus, I know you had some amazing results. Can you just share that with us in just a couple of seconds to get everyone excited to kick off this webinar?
I'll be really happy to. This has been a fantastic collaboration. And, you know, the team.
over at Spacio Crypto is really good. There's only six of them and they deliver some amazing stuff. And also the Make platform has really delivered on this one. When we started, it was two languages.
And today we're running eight translated languages, total of nine. And one thing really did, the primary driver was speed. You know, we wanted the user experience was the key in the beginning.
And then at the end, we realized that, wait a minute, this is... This is really cost efficient. And the number was suddenly we realized that we were, you know, really compared to other solutions for translation, we're saving 80% of the monthly cost. But that was like the added bonus. But the primary thing was really to have the user experience and the site speed.
So it's been it's really fantastic. Wow, that's significant. So that's what we're going to be looking into today.
This is going to be a very practical webinar, we're going to be going into the specific details how Magnus managed to achieve that to automatically localize our websites into eight different languages so let's look at the agenda uh and actually before going to the agenda a little bit of housekeeping so we have we have the chat down there within zoom you can uh chat with everyone there but if you do have a question please do not add it to the chat just add it to the Q&A. There's a button there called Q&A. Add your questions there. We're going to be answering them at the end of the webinar.
Also, rest assured, the webinar will be recorded. The webinar will be shared with you afterwards and also going to be available on our website. So let's kick off this webinar with a quick introduction about Make. Hopefully, if you...
are attending the webinar, you already know about Make. I'm not going to spend hours introducing Make, but Make is an automation platform that enables you to automate any workflow, whether it's simple or complicated. We have more than 1800 connectors and apps that you can connect together. Thousands of people have been using Make for years and they are achieving amazing results.
That's it. I'm not going to go more into details. We're going to be discovering Make today when we're speaking about the implementation of the automation that Magnus has been doing. So now Magnus, to you, maybe introduce us about yourself and Ghostflow. Yes, I'd be happy to.
Basically media technology in any form and before, you know, we started with media and technology, but everything has merged now. So it's media technology. Online publishing in any format, a lot of video streaming, integration, automation and basic problem solving.
The more harder the problem, the more fun. And we also, a lot of people want to migrate from existing platforms. So we assist in bringing people off Substack and other big platforms to independent publishing on Ghost to really be able to speak freely. I like to describe us as a small team of experts. That's a quote from Hitchhiker's Guide to the Galaxy with Zapport people blocks, you know, so that's basically Ghostflow, a small team of experts.
That's what we do here. And for this project, again, I'm really, really happy to be able to share this project with you and everyone here today, because it has really been a fantastic collaboration and just amazing results. For the Spacio Crypto site.
When Salvatore and I started working together, he had two languages, and he was using these dynamic translators available to plug in web-based stuff. And the site had two languages loaded in more than 10 seconds. it just wasn't that that's not going to work uh translations were so so and you couldn't really control them either very difficult to to to get the context right because this is it was crypto information site which is what spazio crypt is uh just unbiased uh crypto information and it was really difficult to get the context for the translator right and the seo was not impressive you know because because the metadata site maps and everything it just was difficult to get that in place.
And in order for Google, and they want the site to be useful in less than 3.8 seconds. And these are the original benchmarks we had when we tested it before we started to work. And I mean, page was usable in eight seconds.
Largest content paint was 6.5. And, you know, it just wasn't really working. So that was the original request from from that. from my client and i think these these are these are like it's challenges that a lot of people are are facing especially if you are focusing on expanding your audience by going to local local markets and yeah it it comes not only with the challenges of translating the upside but as you see the performance of the upside the seo of the upside i think this is a very common it challenges for anyone going into localization so that was the challenges that you were facing uh what What are the situation right now? The situation right now, we're on eight languages.
We're doing Turkish. No, we're not doing Turkish. We're doing Russian, Chinese, Italian, French, German, English, of course, Korean, and Japanese. And one thing we realized that this solution actually scales linearly.
So there is no... tier like oh you can have five languages for this price so you have to switch to another package or you have 500 000 words or something you can just keep adding languages and it will actually scale linearly that's also a really nice thing they are all the separate instances running on their own ghost instance so we have full control over the UX all the menus all the buttons and everything is localized so it turned out to be to be really really nice yes Brilliant. So I think right now that's what we will be looking at. How did you manage to achieve that?
And I think you were describing that to me, you separated that into, let's say, three different automations. So the first one, I think, was like the translation part. Yeah.
So what did you do? How did you start this? Well, the translations have to be static.
So we use any CMS that has an API, we can use those. could be Ghost, WordPress, anything where we can interact with Make Platform. So whenever a post is published in Italian, we publish in eight different languages. All the subscribers, if you're a subscriber to this blog, you get the post by email immediately. We submit it to site engines and we push it to social media also.
And all this happens in less than 30 seconds. So that's pretty impressive. And it's so fast. It's just we actually throttle the thing to just keep not to overload stuff.
But it's really fast. So that's the first part, the translation and CMS integration. We're going to show how to build this with WordPress later today.
And then we have to get them into the SEO and indexing. And one important thing is that these AIs, they actually read data from the web clearly. And I think one thing that is new and kind of important is Bing, because previously no one was like, oh, what's Bing?
But Copilot is reading data using Bing as a source for information. So by submitting to Bing using IndexNow, we also get the stuff into Copilot. So that's nice.
And we also submit it to Google. And we translate everything, the title, the URL, all tags, excerpts. tags, categories, everything is localized in the different languages. We have unique site maps for each one. So that turned out really, really nice.
Social media, of course, prompting a summary where we quickly, we have flows and we push it to any kind of destinations, Medium, LinkedIn, Twitter, and it also happens almost instantly. It's really nice. And then we had one problem that we ran into, or problem or situation was, how do we switch language, right? All the, everything is localized. So let's say someone is reading a post in English and then we say, oh, it's available in French.
I want to switch to the French version. We're wondering, what's the URL of that one? So that was one thing we had to solve. But again, the Make platform has just an amazing set of tools to make that happen.
So we're going to have a look at also how we did to sort out what is the URL for the different language versions. Awesome. That's brilliant. So thank you very much for the introduction.
Now we're going to let's move into the actual demo and see how we build that. And I saw someone who's adding a question there on the chat. Just again, as a reminder, if you have a question, please add it in the Q&A tab.
But just a quick not answer to the question, but actually we have something related to what you're asking about. You were asking about using either a subdomain or a sub directory. And I think I know Magnus has something to say about that. So it's gonna come. uh uh during the webinar but again if you have a question please make sure to add it in the q a so yeah let's go to the the first the first demo and actually yeah doing doing the translation all right so we're gonna build this flow from scratch uh and the first thing uh we want to do is to to connect wordpress with the make platform and to do that we're going to use a web hook a custom web hook for today's demo uh we have built a small, simple WordPress site that has some demo data.
And today we're going to work on this post about cats. So we're going to learn about cats. This is from Wikipedia. So we're going to learn a little bit about cats. I assume you're a cat person, Magnus, right?
Cat person, definitely cat person. Yes. In order to do that, so we want, whenever we update, publish a post in WordPress, we want to notify Make that, okay, a new post is available. And in order to do that, we have to fire a webhook.
And to do that, we have a plugin here, which is called WP Webhooks. And these are all the events that will fire this webhook and communicate with the Make platform that, okay, we have a comment, here's a post created, user created. And today we are using, to simplify everything, we are firing webhooks on post-updated.
Maybe before we, sorry to interrupt, before we go into these details, can we give a very quick introduction to what actually is a webhook? I think like some people, this world might be a little bit new to them. So what is a webhook?
Of course, a webhook is any kind of notification with associated data. It can be anything, a new order. We have, you know, the sun is shining, order has been shipped, you have a new member, post is updated. Basically, any event that happens in any system, we can then trigger a webhook.
We usually say that a webhook fires. And when the webhook fires, it also attaches some data with the event. So this is the outgoing data that goes with the hook.
And we can see that it will send post content. It will send the title. the status of the post and everything so let's let's have a look and see what how this plays of out in make so if we do this and we run this once we see that make is now waiting for for the data so we go here and we're gonna hit our posts and we're gonna go and edit our cat our cat text let's add another something headline and now i save So the post is updated.
But go back to make now, I see that I have received data from WordPress. And we see that this post ID is 198. And under post, I see post content. And here is the actual content.
This is what I need. So we now move data. The post has been brought over to the make platform.
The first thing that happens is that this is supposed to be HTML. It is not. It includes some other stuff here that we need to filter out when we talk. So, you know, we're going to make this very, very simple today. So we're going to have a text parser and we're just going to take HTML to text.
So we're just going to strip all HTML tags because we're taking a little shortcut today for the demo. So as you can see here, HTML, and this is where Make has taken the data. that WordPress has sent and turned it into building blocks. So all this data is now available to us for processing and changing and modifying. And I'm interested in the post content block.
As you can see here. So we do that. Perfect.
So let's try this again and see what happens. We do. OK, fine.
We do here and we remove this exclamation mark. We save. It's updated. And now we see that output here from this module is just the pure, the raw text, which is perfect. Great.
So we now have the control of the text on the make platform. So let's translate this. And we're using DeepL Translate for this. It's very impressive AI context aware, really, really impressive, which is what we use also for Spacio Crypto.
I think one question probably some people will ask here, can we use just Google Translate? Oh, yes, of course you can use any. There are a number of different translators available here.
If we do. You can pull in Translate, you can have DeepL, you have Google Translate, you have Localize, you can even have OpenAI, I think, Translate. You can plug in any translator.
But we really like DeepL. So here's my connection, previously established that I'm paying money to DeepL. And the text I want to translate is this text parses text.
Thank you. And the target language. is all the languages that are available on the DeepL platform. Today, let's try for Spanish.
And the formality is less formal. We're not so strict. Source language, it is English. It's not necessary to... It can actually identify this themselves.
And then some other stuff, if we want to do some more advanced stuff here and things. But overall, this is good. So that's the body.
but we also have to translate the header. So we do another translate module. And this time we want to translate the title, post title. Thank you. And again, the Spanish target language and we're good.
So that's it pretty much. So we can play this now if we do run once. we're going to edit this post we go back to wordpress and we save it and then we're going to go to make and we see that uh for example the headline has been translated to spanish and it's now el gato is this but the really fun thing then is that we can add a wordpress module here we go wordpress and we want to create a post so we put this in the end of our flow I'm going to move this a little bit here.
And this is the connection to my temporary blog here. The title is, let's see, 11 is the title. So the title is from 11, translation and text, El Gato. And the content is the other one.
That's number 10. And here's the text content. the type for this it's a post excerpt we don't have date will be automatic slug status format not but uh i want to do two things i want to flag this post as translated so i insert a tag it's translated uh i am the author of this post and also we want the cover image and it's 199 that can be sorted out later great so this is our post i'm going to save this here and let's see if we can we're on this once uh i go to the post and remove it there i save it and we go here and i go home and let's see if we can refresh this there's always something wrong whenever you do a demo something else it's always more exciting yes uh let's see create the post oh sorry i forgot to publish it of course It's in draft mode. And here it is.
And everything I forgot to set it to published. So what we can do is just to publish this and it's published. And now if I go here and there it is.
El Catos. Yes. So basically the only thing missing here was the status of the post. And the status should be publish.
All right. Great. So actually that's very useful.
So like it's going to be up to you if you want to have the status as only a draft, if you want to check it before it gets published. Or you know what, I trust DeepL, I know my automation, everything's working fine. I just want the post to go be published right away. So maybe to give like a recap of this very simple scenario, you are creating a webhook, which basically till make every time there is a new post, it's going to notify Nukeric. Here is a new post and what this will do, it will just clean up the content of the post and then use DeepL or any other translation platform you would like to use.
You're going to translate the content, you're going to translate the title, and then you're going to also publish it in WordPress. Correct? Yes, correct.
So let's run this again. As you noticed, I changed now to target language Japanese, which is fun. So we go run once, wait for new data because it fires very... This is... We're not gonna do this one.
We're gonna go back into our cat, the English version, and we add another space and we do save. Host updated. And let's see if this will auto publish. It will. So now we have it automatically located into...
What do you say? Japanese, right? I think so. Awesome. I think this is Japanese.
So we should keep adding languages here at this stage. You know, any language that DeepL support or Google Translate or anything. That's fantastic. So we have English, we have Spanish, El Cato. You notice also that DeepL notices that this is Latin and doesn't touch it.
There's a lot of that stuff going on with DeepL that is very, very smart. Yeah, I think this is going to be very important in the kind of translation tool that you use, like how much advanced they are, how much can they understand the context. And as you're also showing that if you want to make it more formal, less formal, I think this is definitely depends on how advanced the tool that you are using. Yes, yes.
So then that's the easy part, I was going to say. So this is why it's a bit more advanced. Yes, because now we have a few problems here. And this is the question regarding how do we put the URLs here?
If you use same site, you can just put them on subdirectories, give them categories E and FRJ. And they will be French and we have Japanese. Or if you have separate instances, you can put them under different subdomains, which is what we're using for Spatio Crypto.
And then we're going to have a problem because they are separate. And how do we? how do we switch language?
So we need to figure out the language switching thingy. And there's another version also, we have to tell Google that there are a number of language versions available. And they use a meta tag called hreflang.
So we also need to say that this post is available in English, French, Spanish, German, Russian, Chinese, and Japanese and provide the URL to that one for the search engine. And the way we solve that, is that we go back to make here is that we provide we build a url lookup table so whenever a post is translated it checks in and say hello i was just translated to japanese and this is my japanese url and when we translate to french it will actually say okay this is my french url and then we have a post id that goes across all language versions so knowing the post id we can then quickly figure out what are the URLs. And then we need to get hold of this data somehow.
All right. And in order to do that, we query this database. This is this data store that is within Make. And we have an API. We built a small API on Cloudflare workers.
And basically you send the post ID. and you get the all the language versions available italian english french german russian and since with this information uh we can do this let's see here is the spot your crypto site it is in japanese and of course we can switch language here but then that takes us to the home page we don't want to do that instead we want to switch to english german french and spanish so if we switch this one to read this post in English. We actually know that this is the same post but translated to english um so it's it's all very you know straightforward that we're able to take the data from the url lookup table uh we provide we build a really simple api here that gives us the urls and this one we can then render into the sites I know this is a little bit of an advanced part. That's why I would not include the automation here.
But if you actually want to build that, I know like Magnus, you can reach out to him afterwards and can share all the blueprints for this automation. I can actually copy-based his automation basically to implement this specific part. But we will still continue. I think we are now moving into the social media, the social media part.
So like how can we automate social media? posting for the different languages that we published. Yes.
So let's see. I'm going to need some help for this module. Yes.
Because I'm going to need three animals. All right. Yeah, I think that's going to be exciting.
See, Akne, we can be using the audience here and we're going to be building this like, oh, we already have some suggestions, which actually I definitely agree with. So we have Panda, Zebra. zebra i like zebra and bear also okay living planet i'm gonna send an email with uh they had panda zebra i like zebra and bear yes any kind of bear specific bear i can spell that we're not going to do hippopotamus because i can't spell it we also have we have we have another someone uh who once suggested Ornithurinco.
I don't know if this is an actual animal or not, but I think, yeah, for the sake of simplicity, let's stick with these three animals, panda, zebra, and bear. Panda, zebra, and bear. Okay, good. So this email address, you see, it's a little bit of a strange email, but let's just fire off this one. So I'm writing an email.
I have three animals here. So we'll send this and we're going to wait for... uh the uh there it is so what we're gonna do is we're gonna go back to our uh wordpress site here which previously and let's see what happens if i refresh this oh here's a story about animals on planet earth that's interesting oh the panda and the zebra and the bear pandas beloved for the distinctive black and white fur All right, so we have some interesting facts about pandas here. They feed on bamboo. We have zebras.
What do they know? Herbivores, tight-knit family groups, really nice. And bears, powerful, fascinating mammals. How did you do that, Magnus? Yes, how did I do that?
And oh my god, it's in French and in Japanese as well. That was fast. How did that happen?
So the same thing is now available. We have it in French. It's Le Panda, Le Zebra, and Le... I'm sorry, my French is really bad. Un Urs?
I don't understand French, but it sounds perfect to me. But even this one is pretty cool. So we have it with headlines and everything. Okay, you made everyone excited.
Now you have to... I have to show you what happened here. Okay, let's have a look. Let's see if we can find something called Planet Earth.
So this was the flow that just happened. So basically, the email is catch. We catch the email with using a mail hook.
That means that Make will actually read the email and make this information available to you. Panda, zebra and bear. So that information is now pulled from the email.
And then we send this to an AI, Anthropic Cloud in this place. and says create a message and the message that we're using is introduce yourself as Claude the helpful AI and write one paragraph of interesting facts on each of the animals provided and you can have a user prompt which is the animal and if we run this with Claude it will run on dog elephant and oh my goodness rate exceeded um ah there you are and here we are dogs elephants and ants. So Claw just generated dogs, elephants and ants here, which is pretty cool.
So this is the prompt, this is the system prompt we're using, and this is the user prompt. So in this translation flow planet earth, there we are, we tell Claude that the content, the user, is the incoming text from the email. And then we also have a system prompt which is introduce yourself as Claude and write three interesting things.
And once we have that... Maybe, maybe, like, so to interrupt here, because a lot of people, when we speak about AI, The first thing that comes to their mind is just chatting with AI, whether it's Google Gemini or ChatGPT. And like this is, it's the limitation where people are thinking, but actually in order to make AI more practical for your business, it's about connecting AI to the other systems that you are working on. And it's exactly what you are doing here.
It's okay, now I'm going to take the data from email as an example, right? And then I'm going to give the prompt to AI here using Make. and then send this information to a different system that I have. And I think this is the power of AI. How can you connect it to all of your system rather than just going and chatting with it?
It is, and it's a lot of fun. And it's just really impressive, especially what they can do here. And I requested the answer in HTML and H2 for headings. So this is what we get. So we have the HTML ready text response.
Here it is with headlines, pandas, zebras, and bear. And then we create an English post using this information, the text response and everything. And then I also translated it to French.
I created a French post, translated to Japanese and trying to create post. And then it seems that I stripped the tags again. And what's this?
Create a post. Do you think this could already be on Facebook? Do you think it's possible? Let's see.
I'm curious. Oh my god, that's already available on Facebook. So we're already posted on social.
Pandas, zebras, and bear with a nice image also. It's pretty cool. And then the last step here is something, another AI module, and then sends an email. So I just want to check if I have an email. Oh, here's an email, the amazing planet.
And this seems more aligned to teenagers or kids. pandas are adorable. They're smileys and stuff. Do you ever wonder what zebras have striking black and white stripes?
It changed the tones of the post, right? And it's basically because when I sent it to, sorry, Planet Earth, there it is. When I sent it to Claude again, the output from previous modules, I used a different prompt. And the prompt I used for this one is a kid's newsletter. You're a children's book teacher.
rewrite these descriptions of animals into an exciting weekly newsletter titled did you know children 10 to 15 age and again answering html with headings so basically you can have your weekly email done in by writing an email um so this will um it's um you know you get your your weekly email done just by sending off one email so it's kind of it's kind of fun that's amazing Yeah, so that was what we did with this flow. And I mean, it's all very straightforward, right? Pick up the email, generate the three basic facts, post it out, post it to Facebook.
And of course, we can add any social media here. And today, I only sent it to myself. But this last step could be MailChimp or Mailgun, hit your mailing list.
And suddenly you have sent your email also. yeah brilliant yeah and I see like you just again just a reminder uh because I still see people adding some questions in the chat please just make sure to add them into the Q&A tab so we can answer them uh at the end yes so um so that's it pretty much it's uh fairly straight um forward um the uh the whole thing um okay so um so that's it pretty much um and build these flows. It's, I mean, it's just, this is the beauty of the Make platform that this some very, very advanced, advanced stuff becomes almost trivial. Yeah. We just add the modules, move the data along the pipeline, change it, modify it, push it here, take the output, do this.
And it just happens. I think maybe that's the challenge here, especially in the use case that you're describing, which is not translating just to one language, but you are translating to like eight languages right now. Does this mean that you need to create eight different scenarios or eight different automations?
You have to update every single one of them every time you make a change? Or how are you managing that? Yes, we did actually initially, which was a little bit rough. But because they're all separate instances, so we had to use different connections to each language.
And that means that we had to duplicate the scenarios for each language. So whenever we had to change, if we had to make a modification to one scenario, we had to do it in eight different places. And I was really happy to read about dynamic connections. So now we can skip all that, basically.
And we just send the language, the connection as a variable. So we can build this with one scenario. And then we just send, okay, this is for the French side.
This is for the Japanese side. This is for the Korean side. This is for the Russian side.
And that will make life much, much easier for us. But this was brand new, called Dynamic Connections in Make. And it makes everything so much easier. And it's a really, really good feature. Awesome.
Yeah, so it's a new feature at Make. Please go and discover it. I'm not going to jump too much into the details now.
That was impressive, Magnus. Thank you. Thank you very much. So what did you achieve at the end? oh at the end uh let's see some takeaways here um you know we published once in italian or team spatio crypto they are able to to do some amazing stuff with a very small team uh they do they do really really they have an enormous reach and they have an enormous uh presence um french german spanish english russian chinese japanese um they they push it to twitter facebook linkedin telegram medium um and the AI translation is static and editable, full UX control.
It looks really, really good. And then of course the surprise was the cost saving. And the components that we use are a make of course, and there's a CDN, bonus CDN for the media delivery. It runs on a private server, DeepL translation, and chat GPT. And that's it basically.
That's all the running costs that Spazio Crypto has today. And as you can see, we did really well here. The speed index is loading below two seconds.
We're on 81 on desktop. SEO is 100. Desk practice is 96. So this is lighthouse value. So we're very happy about how Ghost, in this case, as a CMS delivered.
I'm sure this can be done with WordPress also. But with Ghost, it's extremely fast. platform which is why which is why we like it and i think if i can also quickly show because this been the index or two seconds um to put this in context because what we have here is uh let's see if we have um we should have um this uh there it is um this post i mean there's a see it loads instantly and there's a lot of stuff going on here uh there's a slider here um all the uh categories are translated these are the tags um newsletter loads of video uh and there's even you know a ticker at the bottom and all this where we're able to load in you know if i do a hard reload here boom i mean it takes it takes basically no time And the only way to do this is to have static translation. There's no way to achieve this with the dynamic translations. So that's what we have to do.
This is the beauty of the solution. Brilliant. Brilliant.
No, thank you. Thank you. Thank you very much.
That was definitely impressive. What I really enjoy, like how you took us through that although like at the beginning you look at the scenario and you have a lot of branches and like it's so complicated but actually once you go you go into the details it's actually quite simple like you just understand uh the the few basics it's similar you can uh you can you can do it yourself and i know some people have been asking that question uh can you share this uh this uh scenarios uh starting with them and i know that you are happy to do that so you can reach out to uh to magnus go to uh to go to his website, reach out through LinkedIn, and he will be sharing that with you. And is there anything else or can we move to the questions now, Magnus? Yes, I'd be happy to answer any questions or anything that you want to.
So again, just a reminder, we're going to have the webinar is recorded. We're going to be sending to your email within a couple of days. And it will also be available at Make website.
If you are not already a Make user, you can scan this QR code and you can have two months of the pro plan for free. All right. Shall we move to the questions?
So I think, yeah, someone was still asking about the URL and if we can utilize that on the top level domain, basically instead of subdomain, we are using a subdirectory. And I think you already answered the question that, yes, that is possible, right? Yes, correct. There's two ways of doing it. You know, you can just assign a category to the translated posts and then route them, D-E-F-R.
Or if you're very, very talented with the server side, NGINX server, web server hacking, you can proxy them and just pull them all in under one URL, for example. There are a number of different ways of doing it. SEO-wise, I'm not sure that one is better than the other. Some people would say that it's more nice to have it in the slash subdomain thing.
But we haven't noticed any of that. You know, we're doing really well on SEO or actually Spots of Crypto is doing extremely well on SEO. Awesome. Great. Thank you.
And so there was like two or three questions about the specifics of the translation itself. Can we exclude the specific words? Can we like direct the translation to something or the other? And I think this is more into the specific translation tool. that you use and what what capabilities do you provide right yes yeah they all provide dictionaries where you can say this word translates to this so don't touch this like spots your crypt it translated to space crypto initially they're like no no no it's a brand don't touch it so we're able to tell it to don't uh you know spots your crypto stays that's the brand name it does not don't translate to space crypto because you know so and then all then make actually DeepL will not touch that word.
So you build dictionaries of all the specifics. If you have domain-specific words, you build your own dictionary and tell it to don't touch these. Awesome. And then we have a question about the associated costs.
So yeah, what are the tools that you are paying for within? And I think you kind of answered the question already. Yeah, we can go back one slide.
And make, of course. For credits here, we're running CDN for the asset delivery, the images to get the speed, the global delivery, the VPS private server. It's running on a dedicated VPS somewhere. And then, of course, the translation costs and the AI parsing for the social media.
I think for the Cloudflare workers, that's free up to like, I think, a million requests or something. or if it's like 100,000 requests then. So if you're running below 100,000 page views per month, you're good. But if you go above 100,000 page views, then I'm sure Cloudflare is going to start charging some money. But currently we're on the free tier there because we're not running 100,000 page views yet.
Get closer. Cool. And then we have someone asking about how to control the output quality here. I think this is also related to...
that translation model that you are using right um yeah uh you know really the content side it's very much with the teams but your crypto uh my understanding is that they are very very happy about the translation um the english version is is is that's the one i can read uh my italian is really bad but i'm reading the english and it's flawless and i think the other ones are also um The only one that we noticed that DeepL was struggling with was Korean because it's a mix of two languages. It's how, you know, the written language is part. They have some Chinese characters in there as well. And then they have their new language, Hangul or something. And it was difficult, you know, for the Korean, for DeepL.
But I'm sure they're working on that one. But the other ones are flawless. And I'm sure all the other translators do an equally good job.
It's just that we're very happy with DeepL. for this project. And I think also like as you showed by mistake, that actually you can set up the automation that you are not going to publish the post right away, it's going to stay as a draft and then your team can be reviewing any posts before hitting the publish.
It's going to be up to you if you want to control the quality before publishing it or not. So this is in your hands. And someone was like, how can we insert the same image from the original post?
And I think you're already doing that, right, with what you shot. Yes, yeah. We send the WordPress media ID.
We submit that because in WordPress, you can, in the media library, we have our cat image and the item is 199, so we can send that one. But I think it's also provided with a webhook. I'm pretty sure that the webhook sends the media so it will tell Make that I'm using this image as a cover image.
We can just go on the plugin and have a quick look at Webhooks and tell it that when it sends a post updated outgoing data it will actually send a media id or something. I think, um, post status, paying modified or, you know, post type thumbnail. And I'm, I'm sure media ID should be in there or we can just pull the post.
I think it is probably a part of the thumbnail as well. So, yeah. So, so you get this, the, the media ID, but I entered it, I entered it manually this time, but it can be automated also. Yeah. Yeah.
Understood. Um, and, um. What do we have here?
So yeah, like some people are asking about the quality and like how some people call it the hallucinations of AI, like what is the percentage of the hallucination? Like based on your experience, how much do you see that? It's really good to have a verification stage. Yeah. I've noticed, but that to have one step before the AI goes out initially.
it behaves sometimes, strange things happen. I'd say one out of ten times you get something very strange from the AI. So basically what we do is for other, we fire off an email and say this is the suggested copy for the AI, would you like to publish this? And then we just put a link in the email that fires a hook and then pushes it to social. So it doesn't go straight to social.
it does it and then we send an email and say okay are we good to publish this and you go yes and then it goes out so i think that's a very healthy thing to do still yeah um yeah but so far yeah i don't know but this i think this turned out um really nice uh i mean uh and uh and and also it's early days these ais is going to get better every every day it seems right so they are constantly improving and still So I think, you know, I think it's, it's what you said, and others, others were mentioning as well, I think it's important to still do the quality check before publishing anything with her, like, in this specific example, or any other examples, when using AI, it's a tool that enable you and support support you in your work, but you still need to be checking the quality, the quality of that. And I remember, like listening to, to some of Google official speakers and they were being asked if the content that is produced by AI, will websites be penalized for that? And their answer was no. At the end of the day, are you providing value to the user or not?
Whether the content is written by an AI or written by a human being, the important thing is, are you providing value to the user or not? this is what you need to do just okay you can do and use ai to support you with creating the content but then you're gonna need to go check that content make sure it's valuable for the user then you can go and uh and publish it um all right we have uh maybe one one last question uh but maybe we can share the qr code for the people again uh before asking the question just uh for everyone anyone who did not sign up to make yet you can uh use the qr code to have two months for free for the pro plan and someone was asking about implementing that for previous posts so you already have upside it's already live you already have a long list of uh posts how can we implement a translation for all these previous uh previous posts oh you yeah the the initial translation where we run everything we had that also you know we the the posts were available we had about 200 italian posts 250 and we just read men ran a batch translate, pull all posts and push them through the pipeline and build them. And of course, it's a little bit of a cost, the initial one. to take uh take a lot of 250 250 posts and move them to eight languages but it's a one-time cost but but basically we just set up we just changed the the uh the flow from instead of to wait for a page published we just say give me all the posts and we just iterated through them yeah um which is nice because you know everything it's a it's simple you just uh ran it and then uh yeah but that's it that's an adventure because you want to make you know maybe run 10 posts first see everything works out looks good and then you just run all of them makes sense makes sense great great thank you very much magnus i think that that was awesome alec and i can already see from the number that no one dropped like from the beginning of of the webinar till the end during all the questions no one dropped it was very very interesting very insightful thank you very much Thank you everyone for attending. The webinar has been recorded.
We're going to send you the recording within a couple of days to your emails. It's going to be available at Make as well. You can reach out to Magnus.
He can share with you the blueprints for all the automation that he built so you can copy-paste it yourself. Thank you very much. Have a great day.
Thank you very much. Have a great week. We're going to have more webinars coming soon.
Thanks, Magnus. Thank you very much. Thank you.