Is Lovable AI SEO Broken? A Senior Dev Weighs In…

Description

Is Lovable AI fundamentally broken for SEO?

Update, we've now developed a working prompt/hack for all-pages Lovable SEO - https://lovableagency.co.uk/unlock-seo

There's been a lot of chatter, but what’s actually going on under the hood? In this video, I team up with a real-life senior software engineer to break down the SEO issue, test different fixes (kinda breaking Lovable in the process), and figure out whether any of this even matters.

By the end, we sort of find a fix - but is it worth it? And more importantly, could Lovable 2 solve this once and for all?

💬 What do you think? If you've found a better fix or know something we don’t, drop a comment. We’re testing in real-time and want to hear from you!

🕒 There is a 50min version of this video, with ALLLLL of the trial and error included, I'll upload it later in the week.

🦤 Posting here 👉 https://x.com/benmilsom

🏹 My automation agency 👉 https://steadybowarrow.com/

🎯 My Google Ads and web design agency 👉 https://steadybow.co.uk/

📌 Chapters:
00:00 – Is Lovable AI fundamentally broken for SEO?
00:27 – A senior software engineer joins the investigation.
01:10 – The issue: JavaScript, rendering, and missing HTML.
02:51 – Attempting to fix it: Prompt engineering vs. GitHub install.
05:56 – Did we actually solve it? The final verdict.

Summary

Is Lovable AI SEO Broken? A Senior Dev's Investigation and Solutions

In this informative video, host Ben Milsom teams up with senior software engineer Elliot to investigate a critical issue affecting Lovable AI websites: their fundamental SEO limitations. The video explores whether Lovable AI's approach to website generation is inherently flawed from an SEO perspective and tests potential solutions.

The core problem identified is that Lovable AI generates websites using client-side JavaScript rendering (React-based single-page applications), which means crucial HTML content isn't embedded directly in the page source. When examining the source code of Lovable-generated sites, the duo demonstrates that headings, body text, and other SEO-essential elements aren't visible to search engine crawlers, making these sites effectively invisible for organic search.

Throughout the video, they test multiple approaches to solve this issue. First, they attempt prompt engineering, asking Lovable to convert the project into a progressively enhanced web application with proper HTML structure. While Lovable claims to make the changes, inspection reveals the underlying SEO problems persist.

Taking a more technical approach, they explore implementing server-side rendering using Vite (the build engine Lovable uses). By accessing the GitHub repository and manually implementing SSR capabilities, they achieve a partial solution that shows promise - search engines can now detect headings and content properly, though the implementation breaks some functionality of the preview site.

The video concludes with practical advice for different use cases. For micro-SaaS businesses relying heavily on organic search, the current limitations present significant risks. However, for small business websites where paid advertising drives initial traffic, Lovable AI remains a valuable tool for quickly creating visually appealing, mobile-friendly sites. Ben suggests this issue will likely be addressed in future versions of Lovable or competing AI website builders.

For those seeking immediate solutions, the video mentions they've developed a working prompt/hack for "all-pages Lovable SEO" available on their website, offering a potential workaround while waiting for official fixes from the Lovable team.

Transcript

0:00 is lovable fundamentally is lovable 0:03 fundamentally broken from an SEO 0:05 perspective what exactly is this problem 0:07 that people keep talking about can we 0:09 fix it just with prompting or is there a 0:11 more in-depth problem at the structural 0:13 level within lovable that means it can 0:15 never be fixed will this be solved by 0:17 lovable to and does any of this even 0:19 matter considering the speed at which we 0:21 can spin up prototypes and spin up small 0:22 websites we getting obsessed about SEO 0:24 when it really isn't such a big deal my 0:27 really good buddy Elliot has joined me 0:28 he is a senior Tech developer software 0:31 engineer uh he knows everything I don't 0:33 know from a technical perspective which 0:35 as we know is a lot so the problem as I 0:38 understand it and I might stick a few 0:39 tweets up at this point is that there is 0:41 something wrong with the way that client 0:43 side versus server side rendering is 0:45 handled with all lovable sites what what 0:48 does that actually mean from from a 0:49 search engine optimization perspective 0:51 when these Vibe coders generate the 0:53 websites they often rely purely on 0:55 JavaScript and single page applications 0:57 to do it and what that means is a lot of 0:59 the HML that would usually be embedded 1:01 on the website is absent and therefore 1:03 there's not much for the search engines 1:06 to pick up on right and we saw that you 1:08 told me just by viewing the source code 1:10 of or viewing the source some of the 1:12 sites I've launched you can you can see 1:13 that directly right cool okay so yeah 1:15 for instance we'd take this one which is 1:16 just a bit of a fun idea of of AI agency 1:19 West Sussex if I hit on that and and 1:22 view the page 1:23 Source you will see none of these 1:25 headings are coming across none of this 1:27 body text is actually a visible when we 1:31 get down to a 1:32 purely an SEO like in terms of what the 1:34 bot sees when it lands cuz a Google 1:36 bot's going to be doesn't actually 1:38 visually sense what the site is does it 1:40 it needs to read it at a backend level 1:43 so no good no good for a local service 1:46 business in this with SEO in mind 1:48 directly right how do we fix this so you 1:50 ask me to go away and just start trying 1:51 to fix this directly and just by asking 1:53 lovable so my thought process being that 1:56 lovable is attempting just a purely spit 2:00 out react um where it doesn't need to 2:03 for something like this maybe it would 2:05 be nice in certain context to use a web 2:08 framework but at the end of the day a 2:10 simple website like this uh what's the 2:14 website we're talking the Home 2:15 Maintenance 2:16 website uh it could just be a purely 2:19 HTML static HTML website and then we 2:21 could use something like react just to 2:23 enhance certain aspects of it right okay 2:25 so we so you asked me to do that 2:27 progressively enhanced web app it went 2:29 away and did something then you told me 2:30 to put convert my pwa into a fully 2:32 functional HTML based web page said it 2:34 did it 2:36 mhm but obviously didn't right because 2:38 when we look at the West Sussex 2:41 published page it's not actually 2:42 performed there so so it's not told me 2:45 there's a limitation there it's just not 2:46 gone ahead and added from an SEO 2:48 perspective or from a from an HDML 2:50 perspective did some stuff did a bit 2:53 yeah so what do we do from here you went 2:55 away to the git I I gave you access to 2:57 the to the GitHub yeah so just as a 3:00 proof of concept uh I followed some docs 3:03 on so lovable uses V under the hood um 3:07 what's v what is V that's a good 3:09 question 3:11 it's it's the build engine basically 3:14 that lovable uses to compile all the 3:15 JavaScript or typescript or jsx which is 3:19 what react uses um to generate your HTML 3:23 and and JavaScript and CSS from um so 3:26 yeah on the V documentation uh there's a 3:28 web page all about Serv side rendering 3:31 yes so if we can render some of this 3:33 content server side we can ensure that 3:34 it ends up in the HTML of the final 3:37 website uh and vit the build engine that 3:41 lovable uses has a really good page on 3:43 how to how to implement that yourself um 3:46 and when I say really good I mean you 3:47 can go on there and you can control a 3:49 control C control V into your git repo 3:52 and you can get as a proof of concept 3:54 something serice side 3:57 rendering um the problem I've run into 4:00 in doing that is that yes it worked as a 4:02 proof of concept and I managed to get 4:04 something service i' rendering which was 4:06 a a test string string of text yeah yeah 4:09 that I put in but the rest of the uh 4:12 generated lovable website did did not 4:14 have any textual content that ended up 4:16 on the the page 4:18 [Music] 4:21 Source we've tried to prompt engineer 4:24 our way to get V installed going as far 4:27 as to go and shove it in and then take 4:30 it go in go in so we let we let other 4:33 build a project then we came in tore it 4:36 all out shoved V in and then gave it 4:38 back to loable so here you go and it's 4:41 here trying it's trying to do something 4:42 that's amazing and look what's been 4:44 built while we were doing that and even 4:45 better version of the site and does this 4:48 use ve oh it's so cool this is this is 4:51 off the back of the ve thing I think 4:52 isn't it can you look at the the htl in 4:56 here I'm just enjoying the site right 4:58 now don't mess up the why does preview 5:00 work within here and it doesn't work 5:01 within there that's a question just just 5:03 runs within an if frame right if you 5:04 look at the source for 5:06 this or in frame source yeah view frame 5:10 the frame source and then get rid of 5:11 view Source at the 5:15 beginning what that 5:19 yeah so it works in here it's just an ey 5:23 frame but is there any is there any SSO 5:26 for this search go in there search for 5:28 2003 5:31 oh 5:33 yeah 5:36 SEO he's got 5:38 it 5:45 okay all 5:48 right 5:49 interesting so we 5:53 have effectively done 5:56 it we done it by hook or by crook 6:00 in conclusion Elliot you have 6:02 successfully with my 6:05 help got to a point that moving over to 6:08 the screen recording here we were able 6:10 to develop a website doesn't need to 6:13 look like this but it does because we 6:14 made something fun and a bit stupid 6:17 developed a website that on the back end 6:19 when you look at it H has s server side 6:21 rendering included yes amazing however 6:26 when you go to actually try and view 6:27 that site it's showing not f 6:30 and what were you just saying about that 6:32 yeah I was saying I don't I don't really 6:34 think this is an us problem I think this 6:36 is a lovable problem maybe it's not 6:38 helped by the fact that we shoehorned in 6:41 V into inside the lavables git 6:44 repo um but yeah just going off the back 6:47 of we can we can 6:50 run the eye frame that was okay so 6:55 that's come stra from here right yeah 6:56 this content all rendered in an ey frame 6:58 we've just pulled the ey frame URL out 7:00 and open it up in a browser and this 7:01 runs absolutely fine that um I think I 7:05 don't know but I'm going to I'm going to 7:07 Hazard a guess that something to do with 7:09 the way that The Lovable preview website 7:12 works we've destroyed in the process of 7:17 reverse engineer website yeah yeah but 7:21 it to a certain degree works and it 7:23 works enough that we saw the Google bot 7:25 is picking up the correct headings and 7:27 stuff so yeah interesting Okay cool so 7:29 so what does that mean for us so okay 7:32 cool so part one showed that there is 7:34 potentially something there that can be 7:35 fixed and that hopefully lovable are 7:37 going to realize this is something they 7:38 need to address now how concerning is 7:40 this if I'm using 7:43 lovable as a micro SAS business looking 7:45 to launch and build micras from within 7:48 it initially I think it's fine because 7:52 in the first section of traffic you're 7:54 going to get as it should be the case is 7:56 you go and invest in Google ads you need 7:58 to validate the Google search is the 8:01 method in which you're going to acquire 8:02 users and the quickest way to do that is 8:04 run Google ads you run the ads and build 8:06 a Content plan that will back up your 8:09 your paid acquisition and blend it with 8:11 SEO in this scenario that's not going to 8:14 work right so yeah if I'm building a SAS 8:16 within loveable at the moment the risk 8:19 is high we're hoping they can fix this 8:20 we hoping they will fix this it feels 8:22 like a no-brainer right like they going 8:23 to have to they're going to have to 8:24 address it if I was thinking at this 8:27 approaching this from a web design point 8:29 of view like oh my God now I can make 8:30 five-page websites really fast I'm not 8:33 so bothered I think this actually is a 8:34 really great effective way to launch 8:36 sites much faster and Visually much more 8:39 interesting with lead magnets and 8:41 dynamic elements on those sites that 8:42 look amazing on phones for extremely 8:44 cheaply and extremely fast in my opinion 8:48 a small business could have a five-page 8:50 website that covers Services I expect 8:52 this to be fully solved whether by 8:53 lovable or whoever whatever lovable 2 is 8:56 or whatever the next great big SAS 8:58 Builder V VI coding tool is something 9:01 will be there this will get you by this 9:03 could all change in the next two weeks 9:04 we just don't know we'll keep going and 9:06 looking around we'll keep noodling with 9:07 things see if we can find a hacky 9:09 solution but no doubt there'll be 9:10 something uh that comes up that we can 9:12 use thanks so much share what you're 9:14 building if you found something that 9:15 we've totally missed we want to know 9:16 because I want to work with this stuff 9:18 really Keen to hear from you thanks for 9:19 watching and uh yeah Elliot thanks for 9:21 coming along I know you traveled a long 9:23 way from Five Doors Down thank you mate 9:26 Cheers Cheers 9:33 oh I forgot how good this is Smokey nice 9:34 one isn't it yeah oh it's 8 and a half%