This is the most tactical episode I think we've ever done. It's all about how to use AI in content, whether you can, how you can, where it actually works, where it doesn't work, some tools that you could actually try at home. And it's with Victor, the VP of marketing at Strapi, a headless CMS company. So this is the kind of thing he thinks about all the time.

So there's a good flywheel of making sure the quality gets higher in terms of output, but also it helps us identify the gap so that we can keep providing better answers to the users.

There's a lot of skepticism with content and AI and developers. I'd just be curious to hear, like, how you've been using AI to make content for developers, I guess.

Yeah. Yeah. So we're definitely being very careful. I think that's one of the things I I I keep saying internally is, like, you know, developer trust is really hard to gain and really easy to lose. So we have to be really careful.

And so the first we first started with AI powered content workflow about a year ago. I met a guy called Marcel Santini who created a a company called GrossX, And I was really impressed with, like, how much effort they went into put putting together workflow that were, like, pretty pretty intelligent. So, like, they they had multiple steps to ensure, like, the the output meets, like, the expectation. And the quality is really a a factor of, like, what goes in. Right?

Like, you garbage in, garbage out. So if you put a lot of effort into giving it as much context, doing a lot of the work not leaving to the AI a lot of the research and a lot of the, you know, the little nugget of of information that you can't really outsource to AI, I find. So so when I say, like, you know, AI powered content is still very much human in the loop at every step.

Mhmm. Yeah. It's more augmented. Yeah. Like, how what what does that actually look like?

You give, like, kinda specifics of, like, what what it looks like?

Yeah. So today, like, we have three content workflow. Three, we call them work stream. And so there's one for net new content generation. There's one for content refresh, and then there's one for what we called programmatic SEO.

So the for the first one, the net new content, here, we we look mostly at, you know, the information you can get from SEMrush, like, okay, like, what's the what people are asking about, so, like, the keywords, search volume, things like that. And so try to identify, okay, what are the right topics to cover? And that's mostly for, like, non branded search. So just to to be top of mind on some specific topic or the top of search. So that's the the first one.

The second one is for, like, content refresh. So here, we're looking at a lot of the we we have about 2,000 blog posts today, and so a lot of them tend to be quickly out outdated because, like, they run on older version of different software. A lot of them are tutorial based. So it's not just older version of Strapi, but older version of the other frameworks. And so here, we'll look at for yeah.

Like, are they do they require a complete, like, redo, or, like, is it minor updates? In which case, the AI can perform, like, the they can do a diff and perform some small small update. Or if it's a it requires a complete rewrite, then it's just adding a disclaimer. So that's that in and of itself is already, like, a a big value add for developers because, like, there's nothing worse. We used to get developers complaining, hey.

Like, I I started, like, following this tutorial, but it's outdated, and, like, there's no disclaimer. But just adding the disclaimer in in itself is it's been super helpful. So that's the second one. And then for the third one is I think that's where developers' expectation for programmatic SEO, sync, like, integration pages, some of, like, the feature pages. So they they're not developer, like, are not necessarily going to go, like, follow, you know, from step one or a to z.

It's more like, okay. Like, do that do those two technologies work together? So, like, I think the expectation for that type of content is lower. And for that reason, I think, like, that's a good use case for AI powered content workflows because, like, that's just making sure, like, the developer can quickly understand, okay. They work together, but they're not gonna follow.

And so, like, yeah, they're not gonna read too much through it or care if if some of it is is AI generated. But that's giving you a bit of the high level. I'm happy to to to dive in deeper on one of them.

Yeah. It'd be kind of interesting to know, like, what the actual, like, kind of I'd be curious even just, like, on the system that you have set up, like, in, like, how you've built this, I guess.

Yeah. So initially, so we work a lot with that company that I mentioned called Gross Egg. It was started by by Marcel, and it's growing really fast. And I think they have a lot of

companies How do you spell how do you spell it? Sorry. Just so that people Gross

gross like gross and then x.net.

Okay. Okay. Cool.

And and so I think today they work with with pretty large companies like Hagen or, like, Webflow and and others. And so yeah. So they used to GrossX was working with AirOps. I I don't think that's the case Mhmm. Anymore, but AirOps is another interesting company.

It is similar to n a ten for, you know, creating those, like, workflow builders. And so the the first part is to really give it again as much context, so we call that the the artifacts. So there's three core artifacts that that we we give the the system is some information about your audience personas, some information about the company context, and then information about the writing guidelines, like tone of voice and best practices for interlinking and and so on. And so, like, that starts there. A lot of it's product marketing work, just making sure, like, you have a really deep base of you know, in terms of messaging, in terms of positioning, in terms of capabilities.

So just that's just there. And then from there, you can build the workflow that pull from that memory so that it won't have to keep training that that those workflow again and again. So for, like, a general workflow, we have a a seven step process. Like, it starts with assignment creation. So we we're looking at yeah.

Like, what is the what is the assignment, then doing some research.

Because the assignment is like you human puts us like, a human puts the assignment, like you, I guess. So Correct. Okay.

Correct. But you you just still give it, like, some, you know, like, some guidelines on do you have different kinds of articles? So, like, here you would say, for instance, like, the are we writing a tutorial? What, you know, what what's part of the that tutorial? What technology is being used?

And and so that's part of, like, the brief.

So I guess if it's for, like, Strapi I mean, how to set up headless CMS for a car come I don't know. Like, what something like some vertical. Yeah.

We pick like, I'll give you an example, like, to build, like, a learning management system with Strapi and Next. Js and, like, Mux for, like, the video, you know, like, suggest you usually, like, have, like, three or four tools, like a front end, maybe like a search engine, maybe a media library, and then you just package it up. So that's that's an example. And and for those tutorial, quite frankly, we're not fully there yet in in terms of AI writing those contents because Yeah.

I was gonna last last

That that's the use case where we we we work with influencers. So we work with usually, we work with influencers on getting those project. So so we have the code. We first get the code created, and we have a repo public repo. Everybody can check the code, and then we write like a we create like a a tutorial based on that piece of code.

Scaling DevTools is sponsored by WorkOS. WorkOS helps you get enterprise ready. That means they give you all the things that you need to start working with enterprises. Things like audit trails, skin provisioning, role based access control, and single sign on. Let's hear from Utpal from digger.dev, a dev tool using WorkOS.

We haven't had to think about OTH at all. I think support is great. We have a Slack connect channel with them. Issues, if any, there haven't been many. But anytime there have been issues, it's been addressed super quickly.

So Odd Trail, SSO, stuff like that, we don't think about that anymore. Generally, we don't think about that anymore. And you don't necessarily think about these enterprise features, but they still lead revenue. And it kinda is a no brainer in that sense.

WorkOS helps your dev tools start selling to enterprises much faster. And they're trusted by dev tools like Cursor, Val, and Vercel. If you use their user management off, you can get your first million monthly active users completely free. So let's say like this specific example, you got like a piece of code, a project that has like Marks and Strapi and Next. Js and like all this stuff.

And that that is working. And then you create the kind of assignment and then you go for you're gonna tell me the next part, but that's till this part.

Yeah. So just for context, like a tutorial like that is not a use case for our AI workflow.

Okay. Okay.

That's no. I'm I'm saying just

that's the part where we

yeah. That's the part where we we do internally, and we work with, like, influencers and and developers on creating.

Okay. So you would work with an influencer the whole way on that. Okay.

Yeah. So we we work with, either we do like, we have developer advocates internally that

build Yeah.

Right, build the code, build the project, and then build the tutorial around it, or we we find people, you know, that are doing that for a living that, you know Mhmm. That work with us, and then we we write, like, a piece of they publish it on their channel, like, a a YouTube tutorial on their on their channel, and then we write the text version on our on our blog. But this is not, like, the good use case for AI powered. Don't think it I've seen anybody do that well.

What would be an assignment that would make sense? Like, would it be, like, something quite simple, I guess?

Yeah. So here we're looking at a lot of, like, the, like, listicles in some so, like, okay, like, what let's look at the top editors. It's still, like, a valuable piece of content. Like, the the top editor in 2025 is if you give, like, a nice, like, comparison and, like, yeah, try to add some some good visuals. That's a good piece of content that AI can do a really good job at.

And so that's an example. Another example would be to try to do, for us, like, a comparison of we have a lot of plug ins in the Stripe ecosystem, and so, like, trying to see, okay, all the plug ins for ecommerce. Mhmm. And then you just kinda, like, showcase, like, the top plugins that cover different aspect of an ecommerce use case. Those yeah.

Those are examples of where we have a content workflow for net new content.

Yeah. That's really cool. Okay. And so you have the assignment. So you say like, okay.

So the top Yeah. Editors in 2025 is an assignment. Yeah. Correct. You put the visuals, you create visuals upfront or like you this goes in system.

The you know, I'll view the seven steps and then we can Yeah.

Guess, sorry. Sorry. Sorry. No no worries. No worries.

So, yeah, first is, like, assignment creation, then it's about the research. So that has, like, human in the loop, but, like, also, like, know, the AI can do some research looking at top ranking URL for that specific keyword that we're trying to rank for. Yep. Then it looks at, like, an outline generation. Then it goes into actually writing the article.

Then we have a step for fact checking and code review. Oh, cool. Then we do internal linking because that's a big part of, you know, SEO and what where LLM shine. It's just really, like, having a good inventory of all your content and creating what is called, like, topic cluster. Super important.

This is like making changes, potential changes to other blog posts and stuff, for instance, linking to this one. Is that

It's it's mostly referencing other blog posts.

Oh, within this article.

Yeah. It's creating a bunch of link Okay. Like, that that point to other places.

Yeah. And so you're not making changes to other ones linking here. It's like linking to other pieces within this piece.

The that that's part of, like, the the work stream we have for content refresh.

So Okay.

Yeah. Updating when we, like, update like, do a mass update of our existing, like, content inventory, we can do that for existing post. In in this context, I was describing, like, we do internal linking for net new content.

Mhmm.

Yeah. And and the final step is, like, all the metadata. It's, like, making sure, like, that is done well for, like, Optimus, like, you know, SEO and and GEO best practices. That's the thing nobody wants to do, so it's perfect use case for AI. Yeah.

Yeah. Yeah. Yeah. That's amazing actually. Yeah.

This is incredible. And I feel like and just just like this is a side point, but right now, it feels like tutorials and stuff is it's not great art. Do you think that's gonna change? Do you think it's do you think you'll one day, like, in two years or three years or even less, you'll be able to do the same process but like tutorials?

I think so. For sure. I I am pretty convinced. Like, I mean, you've let's see how far we've gone in terms of, like, not only to create like, the you can vibe code an app. So you the that part in itself is, like, also kinda taken care of.

Like, you don't need to, you know, stitch everything together. Like, it'll give you, like, a good skeleton. You still need again, you'll still need human in the loop. So for the the app creation and then, like, for for the tutorial, I think, like, yeah. I mean, we use I'll give you an example.

We use a tool called Kappa, kappa.ai, which is really good at it's, like, essentially your your custom LLM. It's trained on our own, like, documentation and all the the resources that we give it and say, hey. Like, basically, if you don't know, don't hallucinate. Don't say, like, you know, something you shouldn't. Just basically, this is what you know.

This is what you what you should answer to. And and we use that as the fact checking step of the workflow. And so that's I think using those kinds of tools, I think, is gonna make sure, like, there's no hallucination and the quality of the of the tutorial is is is getting better and better. So so, yes. I'm not sure exactly how how that's gonna be done, but I definitely see a world where that's happening.

Yeah. I yeah. I met one of the Kappa guys, really nice guy, how to use it for, like, fact checking. It's like a as a it has, like, an API that you can hit or, like, how does

Yeah. It has an API. So, like, we exactly like, we we call the API from the workflow, and then that just makes sure, like, you know, it's Kappa doesn't flag anything as that does not that's in the in the the output of the workflow. There's nothing yeah.

Wow. So Kappa can, like, flag stuff that is not like, it's like, no. This is wrong.

Well, so we we we it's like, you know, you can use Kappa. You copy paste the blog post in Kappa, and you say, hey. Like, is that does that sound so you can still do that, but we're just automating it by calling the the API. So could we for the other we use other models for the first step, and so we wanna make sure we do the fact checking with the model that we have a 100% trust in.

Because you've you've been using that on your site, so you know that, like Correct.

That's trained, know, it's You

know, at the edge. Yeah.

Our engineers and, you know, everyone in the team has been checking, and and so that's an an interesting part about Kappa is, like, you have that feedback loop. So now we're also looking at we're using that for content discovery. He had introduced a new feature called content gap because you can see what where Kappa, like, sometimes say, I don't know. And so, like, you're like, oh, they should know. And so, like, that's just pointing, like, what the content where the content gaps are, and then he tells our team, okay.

Like, we need to write about this.

Yeah.

And so there's a good flywheel of you know, Kappa is not only making sure the quality gets higher in terms of output, but also it helps us identify the gap in the in the initial step so that we can keep, you know, making the Kappa smarter and then providing better answers to the to the users.

This is really cool. I didn't know about this. This is that's amazing. It's like tests for, like feels like they're kind of, you know, like people's it was like, if you get if you've got in theory, if you've got good human written tests, you could like vibe code the hell out of it and like this feels like they're kind of the equivalent in content, I guess, of like

you Exactly.

You could get it to create lots of stuff and then know it's

true, hopefully. And so one thing we're we're, like, experimenting with is also for, like, social listening. We we use another tool called Octolens. Mhmm. That's just like listening to what people are saying about your brand, and it's really about so trying to also infer some topics that the community is asking about.

So that's another in terms of content gap detection, it's not just just Kappa, but also, like, okay. Let's listen to what the whole world is talking about in terms of, okay. Like, what about using vibe coding tool for the front end and then a headless CMS for the back end? A lot of people is is asking about that. You know, we should write about this.

So hopefully that helps as well to give you the full picture of the stack.

Yeah. This is so cool. I feel like I I feel like this is very useful. I wanna try it. Yeah.

I I've I've started to use a lot of stuff for like docs and things where like it's like you can take like a if someone messages like my my boss would be like, oh, we need to link this in here and like mention that this is a limitation or that people should be aware of this. And then you can use like, ours is linked, like, Codex is linked up to the docs. You just it's like, you literally just paste that into Codex. Can you change this in the docs? And it will Yeah.

Like 99% of the time do it really well. Yeah. And it's and you've got like poor request and just boom. It's updated. So Yeah.

This is this feels like and that that to me feels like a very useful use of AI that's like quite it's it's useful. It means we can update docs faster. Keep and this feels like similar in a sense that like okay. Right now, it's like list calls, but I can see like so it seems like this is gonna just allow you to create such like incredible scale of content that

Yeah. Yeah. Another example is around like f h if FAQ. So, like, for the feature pages, we're trying to anticipate what are the the question that users have or might have, and then and then creating really smart FAQ with AI. That's something that that works really well for us, and it's providing value.

And it's really good for GEO, so generative engine optimization Mhmm. Or whatever people call it these days. That yeah. That's that's another, I think, good use case for AI.

Yeah. Okay. First question. Is GEO is this a real thing, or is it like just do SEO?

I think I think it's quite different. But there's obviously an overlap. I mean, it's the the same principles, know, backlink, importance in SEO, importance in GEO, but they are, like, things that are different. And so it's, you know, a different game. And one one thing that's really important is the freshness of your content.

So that's why I think AI plays a big role in keeping your content fresh because, you know, we we see a lot of the content that used to perform well, it is slowly decreasing because there's probably a fresher content that's coming and and taking our spots. And so that's why, like, we're able to monitor, like, those downward trends for some of the content that used to to rank really well and try to improve. It's telling us, okay. Like, that content is no longer fresh. Like, we need to spend a little bit of time to to to make it better and update it with the the latest.

So that's another good use case, I think, for for AI spotting like the

yeah. Yeah. And you're kind of doing that like man like, are you doing that by eye? You'll see this is degrading in your dashboard and then you will go, like, what what does that look that's how you're doing it.

Yeah. Yeah. No. That's that's step is to your premium. It's manual, but, yeah, you can track it.

Do you have, like, a flow for updating, like, another workflow for

Correct.

Refreshing? Okay.

Yeah. Exactly. That's what we were mentioning earlier earlier.

Okay.

The e for interlinking, for topic cluster creation. Okay. That's another thing that's changing with with SEO versus GEO is that in SEO, you know, there used to be, like, one keyword or two keywords, and, you know, that used to make or break whether you you're you're getting, like, traffic. And with with Neo now, people type, like, 20 words queue like, prompt. Right?

So so all of a sudden, like, the the long tail of content is is is more important. And so you can go after very niche topics with AI by automating some of, like, the, yeah, the content coverage for for that niche, for that long tail, you can do that, like, better with AI and create content cluster. So you can have like, let's say, you you have in a content cluster, you have one piece of content that's, like that's your authority. You say you you point every single blog post on that topic, point to that that core piece. That piece should be, you know, human human return and, like, a very good tutorial.

Like, somebody, like, spent hours, like, making it super good. And then all the other piece of content that are more, you know, informative but less value, but they all point to that that main that main post. And that's how you create a content cluster, and that's how, like, you know, g that's important for GEO.

Is that different because of GEO? Like, it's just more like, cards on table, I'm not good. Like, bit of a noob in SEO. I I get the premise, but, like, is that much more important in geo, the clustering?

Yeah. I I think so. And, again, because of, like, the the prompt are getting longer, so that's an opportunity to surface, you know, content that cover like a more specific Yeah.

So you need they're like your net kind of the slightly lower quality surrounding pieces are like the the net that catches people and then they get to there and then they get linked to the one that's really good. And that's the one that really Correct. Helps them. But they wouldn't have got there without the the one that addresses their specific words that they put in.

Correct.

Yeah. That makes sense. Because I guess also people just don't know what search sometimes probably. Right? Like it's just they don't know what they're looking for.

They don't maybe don't know what like a headless CMS is. Just like they want to be able to add Yeah. A blog and they don't wanna use WordPress or something and like the

Yeah. So and that's another thing. So freshness is super important. Then you have like the social mentions. That's why I think like the Mhmm.

You know, the Reddit is becoming even more important. Just making sure, like, you see user generated content, making sure, like, you have that third party endorsement from people that's more like the the share of voice of your brand and and how that's surfacing in in those key channels where, in our case, where developers are spending their time. Used to be Stack Overflow. Now it seems to be a lot more Reddit. Yeah.

So having that presence and having that community of fans that recommend your your brand, I think, is is super important.

Mhmm. How what's your do you have, like with Octolence, do you do you just have, like, a because I think we've got it set up, But not right now, because we're so small, we don't get many mentions. But we just get lots of like competitors and like keywords and stuff. To be honest, like it's something that I have not been checking very much. Because it's just like just got these like always white Slack channels with like tons of stuff.

Do you how how do you like make sure you use it well, I guess?

Yeah. How do you break through the noise? I think it's a good a good question. Like so we have a a weekly recap. I think it it does a really good job at sending you a weekly summary, and then that's actually something that we share straight with the with the PMs.

It's not just for, like, a a go to market team, but also for, like, you know, product and engineering so they they can get something back from from the community. And so it gives you, like, the highlight and then with, like, the different mentions as, like, a as a citation so you can refer back to the but it it gives you a summary. Okay. Highlight what are things people love, and then lowlight what are things people don't like. And so that's giving the team a lot of the feedback and say, like, confirming, you know, what what they know from interviews and research, but that's coming on a on a weekly basis.

So it's it's good feedback for the team and then for, you know, for what a Strapi developer advocate also. The goal is to not go every time somebody mentions your brand or, like, a competitor, you come in and say, hey. Check out, like, your own brand. That's not that's what you should not do. Okay.

What you should do is, like, yeah. Don't do that at all. I think, like, people will see, like, it's spam. First of all, you'll be you'll be banned. Like, they have moderators.

It's not gonna it's not gonna go through, but it's really about, you know, helping people. If people have questions, they are stuck. We mostly do that for our own subreddit, and then in some cases, for other, you know, other subreddit, it's just, you know, providing value. So it's so we use it in just to recap, feedback for the product engineering, content gap analysis, like, what what people are asking about, what content we should create, and then for, like, kinda support and making sure that support is you know, they have answers. Because the chain though those Reddit questions, they they show up in in JajiPT or in Google.

So if you if they come non not answered, then, you know, some somebody's the chances are if somebody's asking, a lot of people are gonna be wondering. And so it's important that everything gets gets answered when there is a is a question. And that's another use case for for Kappa. We used to on the forum or on Discord, like, people ask question, we used to tag our Kappa so that, like, they all have answers. And because those things get indexed, and then that that's like the another way where you you have a flywheel going on is to to use AI to sue for support coverage.

So wait. Someone someone asked the question and then Kappa answers or or Kappa gets tagged in the answer with so that they build up their knowledge base. Is that what?

So we check, like, if it's not answered by a human. Mhmm. We have instead of letting it go unanswered, you just have come in and give, like, an AI generated answer,

which is

often pretty good. Yeah. So that way, you try to keep the best of both worlds. Like, know, trying to get that community collaboration fast. Alive.

But, you know, in case people don't get answer, you loop in the AI.

Yeah. And she's sorry. Just going back. I just have one quick question that was, like, probably very specific, but, like, how do you get because I I think, like, especially, like, a bigger startup. How do you get, like, one per get someone to answer, like, to Freda?

Is it, like, yeah is it, like, someone has the responsibility? Or, like, yeah, do you share it? Or how do you?

Yeah. We we share it. So we we have a Slack channel where, like, mentions get posted, and then Yeah. People, they can just check, like, let's say, okay. I answered it.

Just put, like, the check emoji in. And so, like, it it where whenever it's not, like we don't do a 100% answer. We try to do you know, so that that it's a team sport. Yeah.

Yeah. Okay. And then within people do try yeah. That makes sense.

Mhmm.

That's cool. Actually, yeah. One other point going back to like the SEO, geo stuff. Seems like you're using tools for that. You mentioned SEMrush.

Is that what you're using for like for geo as well?

Yeah. So we use Scrunch for Scrunch? Yeah. For GEO, it's like a lot of people know about Profound. It's another one.

So it's giving you, like, citation. It it essentially does the the the same thing as I think, actually, SEMRush now, they have also, like, geo capabilities. Yep. But it's like, yeah, they are querying you you define your prompts that matter to you, and then you monitor, like, your your presence in those key prompts. But those tools are moving fast.

Yeah. That's it's still new.

Okay. Is it useful?

I think it probably does provide some value for sure. Like, you you can really understand how you're doing against your competition, what are the the sources. So building authority, like, domain authority is something that was important with SEO. It's still important in GEO because you see some of that in the answers to the prompt. And so, like, it also, you know, tells you what other good outlets to to try to get coverage.

So so, yeah, it is useful for sure. Okay.

Yeah. This is really helpful, by the way. Just generally, like, thank you for sharing so much. Because I think like this is the sort of thing that I mean, for myself, this is useful. Sure people listening finding this super helpful as well.

My pleasure.

Yeah. Are there any other ways that you're using AI in marketing that people may not be realizing that you can use it for?

That's a good question. So we we mostly use it for for content, and we're trying to see and what's interesting was with GrossX and some of the tools I mentioned is, like, we're trying to see what's the what should be from a tool outside of Strapi versus, like, a tool that we develop internally? That there's there's a build versus buy question. And because as a content management system, there is obviously a lot of things that, you know, we feel should be the role of your of your CMS. So it it's more like building some of those agents internally.

So we just announced, like, you know, in Strapi AI, you have a you get a AI architect. So you you give it, like, a Figma file, and then it it creates the the content model, the the content schema for that file. That's one example. Another one is, like, content translation, so you can dramatically translate your content. Another one is, like, for the metadata.

You you when you upload your images, just making sure it fills out the the meta tags. So those are, like, some of the things that, you know, we do outside of Strapi, like, we've been doing for, like, you know, a couple months, but we also that informs a lot of the strategy for for us internally. Okay. Like, that's if we're doing it, the other people are wanna are gonna wanna do it as well, and, like, it should be done in the in the CMS directly because not everybody can afford to have like, to work with an agency and build some advanced content workflows. So I I think that's, you know, that's important to say it is an investment for us.

I speak a lot about, like, channel market fit. So for us, like, you know, content is one of our main acquisition channel because developer you know, it's important. It works well with developers. We have a, you

know, a

low contract value, low ACV. So, like, you have to also look for a channel that is low CAC because, you know, either when you look at CAC versus a ACV, you need to also make sense of of the channel. And so for us, content makes sense, especially because we're a content management system. But Yeah. You also we also recognize that some people, it might not make sense.

Even though they are, you know, they're in-depth too, maybe it depending on the size of their contract value, they they should think about maybe content a bit differently.

One thing you said there just on like languages. Do you think like okay. So let's say we started like, we're like six people, And if we started using Strapi for for our blog and stuff, at what point do you think it makes sense to start like using language? Like putting other stuff, like, to start translating?

That's a good question. It is it is it kinda depends on what you're seeing in your community. I remember talking to Sam at at Mastro.

It's it's so funny, like, because I've spoken you're about to say that they're big in Japan. Right?

Yeah. Exactly.

It's so funny. I've spoken about this, like, five times on the podcast. I think Mastro is, like, so known for being big in Japan. It's so funny. Yeah.

He's totally nice. Everyone

does. Yeah.

I know some others then. So CodeRabbit. One is a CodeRabbit is another company that is big in Japan. Yeah. So I think it also depends on, like, your user base.

Like, it's it's maybe a chicken and egg problem. Right? But Yeah. But it I think before investing into well, I guess it is it's changing. The unit economics of doing internationalization is getting better and better.

It used to be costly because you you have, like, translators and build, like and and pay, like, a translation tool. But that with AI, like, all of a sudden, the cost of internationalization is is shrinking. And so I think it's gonna be it's gonna be a no brainer.

For tutorials and stuff, is it better? Because you don't need to translate the code. And so that's kind of the anchor, like even if you mess up the stuff in between, they're still gonna have to code. So is it like gonna be maybe better than just a I don't know. Or have you have you have you

yeah. I think so. I think it's gonna be it's gonna be it's gonna be better because, yeah, you have that common denominator so that a lot of the, you know, tutorial, like, a lot of the text is code anyway, so that don't have to translate as much. So yes. Yeah.

Yeah. Do you

do you guys translate into different languages?

Not so much today,

but

for the reason I mentioned. Because you you know, it's a matter of economics. And but I think when we we, you know, we just are shipping that AI translation, think it's definitely on the road map to

Okay.

To .food and and then start translating.

Yeah. Okay. Okay. So if we start seeing your blog in Japanese, then we know what's coming down the line.

Exactly.

The release is coming soon. Yeah.

Exactly.

Okay. Great. Cool. And then do you see, like, among, like, your customers any interesting stuff that people are doing with LLMs and stuff that's, like, paying off?

Yeah. Programmatic SEO is a is a big one. So I think a lot of a lot of people are able to create, like, a very large number of pages, thanks to headless CMS and and AI. So for for our customers, you know, that that's the you give, like, the the the structure is always gonna be the same for all those pages, and that's where, like, you know, the the headless CMS approach to content modeling makes a lot of sense. And then you you just have AI in the loop to to fill in the the the the valuable piece of content for each specific technology.

So, yeah, we have a lot of companies doing that really well. You know, Just Eat Takeaways is is one of them that is about to localize the you know, for, like, groceries, like, food deliveries in x cities. So that's that's an example.

Oh, okay.

So so that's, yeah, covering the the long tail of of of pages and keywords.

Okay. Nice. And is there, like because I know, like, whenever this stuff is talked about, there's also, like, some people that say, like, oh, you need to be careful about this sort of thing because then, like, Google could see there's a lot of LLM.

Yeah. I think it provides value. As to me to me, as long as it provides value and it's it's not like you you're, you know, doing the hacky technique of filling the page with a bunch of keywords multiple time. Like, that that's not what we're doing. It's it's literally you providing genuine it's providing structured data in a way that is and you display it in a way that is helpful, not only for humans, but also for robots, like crawling your site for dexing.

So that's my rule of thumb. As long as it provides value, the fact that it's automated or AI generated, I don't think it it's gonna be a red flag. And we we have yet to see someone getting penalized by creating a very large number of pages in in that way. Because, again, if you do it well, it it provides value, and and and, yeah, there's no reason to to to flag it.

Yeah. That makes sense. Yep. Yep. So if it's useful, it's fine.

I guess that's the same with generate code as well. If it's good, then it's it's it's fine. Exactly.

But, yes, to kind of wrap it up, like, human in the loop is is still very important, even more so for, I think, dev developer tools company because, you know, developers have that AI detection radar is is activated. And and so you just you just have to provide value even more so with developers than than others personas. So I think that's why we if you're gonna invest in AI powered content, just make sure you have also, like, you know, human in the loop and at every step. I think that that's the the condition to make sure, like, it it pays it pays out.

Yeah. Yeah. That that makes total sense. And that's definitely been how I found this just how helping you if you just get like the draft or whatever. It's like so good.

And then I hadn't even considered all the ways that Yeah. The processes that you've set up and stuff like that. It's, like, really, really interesting.

One thing that I'll I'll mention as well is, like, we talked a lot about content generation, and I think AI can do great work there. But that's only half the story. Right? The other half of the story is content syndication and distribution. And so I think AI can play a big role there as well in terms of repackaging your content for specific channels.

Like, you you know, like, car sales do good on LinkedIn, and then you have threads on x, and then you have, like, shorts on, you know Yeah. TikTok. I think thinking about that as well is is very important, and I think AI is playing a big role in terms of you you give it it's at the intersection of headless CMS because, again, like, that's the point of a headless system is, like, the head is not predefined. It's API, and so you can adapt it to to whatever format, whatever channel, and and also where, you know, AI can can bring a ton of a ton of value. So that's where, like, we see like, our work is well, we're not done.

Right? It's like I giving you, like, areas where we're where we're investing is the content distribution piece and also, like, around image generation. I think, you know, LLMs are okay at the moment, but they still take a lot of training to provide useful content. You know, they can provide a lot of good text, but if, you know, as as as a reader of articles or content, you you need to have that content breeze and have, like, you know, visuals. And Yeah.

And so that's what we're trying to do to try to to mix and match, you know, text, visual like photos, images, videos, GIFs, all kinds of of, yeah, visuals as well on top of it.

Are you getting any early successes with that? Or

We're I I can't report quite yet on that. Like, I I I I'm Still in flight. Yeah. It's still in flight, but we we're we're investing heavily on that topic.

Okay. Yeah. Yeah. That makes sense because I guess when we start seeing less Unsplash images across the web, we'll know that it's starting to work.

Yeah. Exactly. Yeah. I think it's all we're almost there. I mean, you just look at what's happening with, like, the the the models are just getting much better.

Right? So

Yeah. My sister's been posting that like the Gemini like homeless prank where you put like a told my mom that there's like a homeless guy and that she invited him in for lunch and stuff like, it's it's I don't know if you've seen that but it's like taken over. It's like, so my sister's using Gemini now, which is like, feels like and she's not techy at all. It feels like it's it's definitely breaking and I don't know. Oh, Yeah.

Yeah. We saw a few and I mean, it's gonna be Yeah. It's gonna be wild, I think. And that that's another important, you know, fact you're you're bringing up is is the quality of content. Like, the volume is getting exponentially, like, bigger, and and the quality is going lower.

So I think it it's really you know, can can we have the guardrail so that, like, you can really, you know, lean in on people that have, like, op opinionated or subject matter expertise to counterbalance the all the crap that's out there in terms of, you know, content. You see what I'm saying? Like, the Yeah. To me, that's gonna be important. I think we'll we'll see a big shift as well in terms of and that's why influencers influencer marketing are taking off is because people trust, you know, their favorite you do YouTubers to to deliver really good quality content.

But, yeah, I think I think that's an interesting topic as well.

So so I feel is it what you're saying is like, there is all this opportunity to create lots of content and to like keep things up to date and have this like much bigger, like surface area of content. But also, you can have like having something really exceptional is even more valuable now.

Yes. I think so. That's why that's why I'm getting to. And and Yeah. And to to us, it's really working with doing some curation of content.

And that's so you you curate, You break through the noise, but you you're doing the work of, like, you know, a newsletter editor. You just, like, reads everything, finds like, cherry picks, like, the top. Yeah. I think there is value in doing that as humans, and then and and then feeding that to a LLM workflow to structure your thought. You know?

You don't wanna you wanna give credit to I'm not saying you steal their work and and, you know, you you make it yours. No. Not what I'm saying, but you give credit.

Hard to hear from Victor. No.

But you know what I'm saying? It's like, don't

take some

YouTube video then, like, translate it and then write a tutorial about it. Like, you wanna give them credit, and you wanna reference, like, the people that have put the work in. I I think that's really where I think the human in the loop makes the most sense in in the the content curation step. And and that goes with, like, the personal brand of, like, some, like, public trainers that, you know Yeah. They invest the time, and and it's a craft.

Right? And so you wanna make sure you you also, like again, what content comes in, like, it's the quality of the output at the end.

Mhmm. Yeah. There's not shortcuts. There's some shortcuts, but it's still Yeah. It's still quality in, quality out.

Yeah. Yeah. Okay. Amazing. Well, Victor, I think that's all we've got time for.

But this was incredibly useful. I think this might be the most tactical episode we've ever done. And I loved it. Hope everyone listening enjoyed it as well. Thank you so much.

You're welcome. Thank you, Jack.

And where can people go to learn more about Strapi?

Strapi.io is the main website and, yeah, we have a new project coming coming soon that's gonna be in the in the vibe coding space. We're trying to bridge the best of both worlds in terms of our content management expertise with, you know, the wow effect of AI pipe coating.