Building an SEO-Ready Website With AI Tools
A Semrush tutorial walks through building a search-optimized website using Lovable and AI prompts. Here's what the workflow actually delivers—and where it falls short.
Written by AI. Bob Reynolds
Photo: AI. Mika Sørensen
Hundreds of thousands of websites go live every day. Most of them vanish into the void—no rankings, no traffic, no AI citations. That gap between "site exists" and "site gets found" is the problem a recent Semrush tutorial sets out to close, using a combination of Lovable (an AI web-building tool) and Semrush's MCP integration to build a search-optimized site from scratch, inside roughly an hour of camera battery life.
The demo is honest about its constraints, which makes it worth paying attention to. The presenter builds a fictional personal finance business called Penny & Plan in real time, narrating every prompt and every correction. The result is less a polished product showcase and more a working document of what this workflow actually looks like when things go slightly sideways—which they do, and which is precisely where the useful information lives.
What the workflow is actually doing
The core idea here is sequencing: get SEO considerations into the build process from the first prompt, rather than bolting them on afterward. To that end, the starting prompt fed into Lovable combines three things at once—a business brief, a style guide, and a block of SEO best practices. It's a denser-than-usual input, but the logic is sound. Defaults are easier to bake in than to retrofit.
Before Lovable starts generating anything, the presenter uses the tool's "plan mode"—a pre-build review step that shows you what the AI intends to do before it spends your credits doing it. The tool asks clarifying questions: how much of the site to build in the first pass, whether to add live payment processing, whether to auto-generate blog content. The presenter's answer on blog posts is worth noting: "Don't write any posts and instead I will provide them later."
That restraint runs counter to how most people are inclined to use these tools—ask for everything, get everything, sort it out later. The argument here is that AI-generated content at scale, published immediately, is a liability, not an asset. Whether you agree probably depends on how you weight speed against quality and how seriously you take Google's documented skepticism of bulk AI content.
Once the site is built, the Semrush integration does real work. Pulling keyword suggestions through the chat interface—rather than toggling between tabs in separate tools—is a genuine convenience. The output includes search volume, keyword difficulty, suggested page matches, and a competitor list with estimated traffic figures. It's not replacing a dedicated keyword research session, but it surfaces enough direction to be useful at the build stage, which is earlier than most people think to do it.
Where the seams show
The post-build audit is where the tutorial gets genuinely instructive, partly because it stops performing smoothness.
The Semrush Site Audit returns a reasonable health score, but it flags real issues: a sitemap using relative rather than absolute URLs, thin page content, sparse internal linking. These are predictable outputs from a first-pass AI build. None of them are catastrophic. All of them require a human to triage and prioritize, even if Lovable can be sent back in to fix the flagged items.
Google's PageSpeed Insights returns a 76—passing, but not clean. The culprit is an oversized hero image on the homepage, a classic and entirely avoidable problem. The presenter notes it, doesn't dwell on it, and moves on. Fair enough. But it's a reminder that "SEO-optimized from the start" still leaves room for the kind of basic performance issues that have existed since the first slow-loading website annoyed its first impatient visitor.
The JavaScript test is the most interesting friction point in the demo. Disabling JavaScript in the browser reveals that the site's budget calculator—a functional tool Lovable built from the initial prompt, without being explicitly coded by the presenter—stops working. The structural pages hold up. The interactive element doesn't. The presenter explains why this matters: "AI tools like ChatGPT and Claude often struggle with JavaScript-heavy websites in terms of parsing them." If you're optimizing for AI visibility—for appearing in AI-generated answers, not just traditional search results—that's a real consideration, not a theoretical one.
The FAQ incident is the tutorial's most human moment. Semrush analysis of the demo blog post suggests adding an FAQ block with FAQ schema. The presenter instructs Lovable to implement the recommendation. Lovable adds FAQ sections everywhere—beginning, middle, end—turning a readable post into a cluttered mess. A few correction prompts later, it's fixed. The presenter attributes this to using build mode rather than plan mode, which would have surfaced the misunderstanding before it executed. "Take that as a lesson to use plan mode more often," is the note—practical advice delivered without drama.
The honest accounting
By the presenter's own reckoning, total recording time ran to an hour and thirteen minutes, with active work closer to thirty minutes. That framing matters because it's where the actual claim lives: not that AI builds perfect websites, but that it compresses the time between "I have a business idea" and "I have a functional site to iterate on."
The schema markup comes back clean—zero errors, zero warnings—which is a legitimately good result for a first-pass AI build by someone who has never written schema manually. The site passes agentic browsing audits in PageSpeed Insights, which the presenter flags as relevant to AI citation visibility. These are not trivial outcomes.
What the tutorial doesn't address, and can't within its scope, is what happens next. Publishing a site is an event. Building domain authority, earning backlinks, consistently producing content that answers real questions—those are processes. The presenter acknowledges this directly: "I'm not suggesting that web development itself is as easy as sending a few prompts to an AI tool, but for many people watching this, it will get you started in a way that simply wasn't possible before tools like Lovable came along."
That's a calibrated claim. A plausible one, too. The legitimate question isn't whether the workflow produces a site—it demonstrably does—but whether a site produced this way, by someone without web development or SEO expertise, will hold up under the iterative maintenance that search visibility actually requires. That answer won't come from a tutorial. It'll come from the traffic data six months out.
By Bob Reynolds, Senior Technology Correspondent, BuzzRAG
More Like This
Anthropic's Claude Routines Targets No-Code Automation Market
Claude Routines lets users automate workflows with natural language instead of drag-and-drop builders. Is this the end of traditional no-code platforms?
AgentZero's Sub-Agents: Self-Modifying AI Delegation
AgentZero demonstrates AI agents that create and manage specialized subordinates on demand. The system modifies itself—which raises practical questions.
Not Every Problem Needs AI. Here's How to Tell.
Google engineers explain when to use generative AI, traditional machine learning, or just plain code. The answer matters more than you'd think.
Anthropic's Advisor Strategy: When Cheaper AI Models Work Better
Anthropic's new advisor strategy pairs expensive Opus with budget models, cutting costs by 12% while maintaining quality. But testing reveals surprises.
What Rust Actually Does Better (And What That Means)
Rust's advocates make bold claims about safety, tooling, and career value. Here's a clear-eyed look at what holds up—and what questions remain.
Replit Builds Real Apps From Plain English Prompts
Replit now turns plain-language descriptions into full-stack web apps. A hands-on demo raises real questions about who benefits—and what gets lost.
RAG·vector embedding
2026-08-21This article is indexed as a 1536-dimensional vector for semantic retrieval. Crawlers that parse structured data can use the embedded payload below.