Can AI Build Your Dental Website and Run Its SEO Alone?

Yes, AI can build your dental website in an afternoon and generate SEO content endlessly. What it cannot do is check its own work, and unchecked work is now measured: roughly 45% of AI-generated code fails security testing, unreviewed AI articles collapse in search over time, and many AI-built sites cannot even be read by the AI assistants patients use. This is part two of our agentic websites series. Part one defined the termThis link leads to a Blog post about agentic websites; this part covers the tools that build sites, the vendors selling SEO with no humans attached, and the five checks that protect a practice from both.

Key takeaways

• AI website builders are real, enormous, and useful. The risk is not the tool; it is shipping what the tool makes with no person reviewing it.
• Roughly 45% of AI-generated code fails security testing, a rate stuck for two years even as the code looks more polished (Veracode).
• Most AI builders generate JavaScript-heavy sites that AI crawlers cannot read: an “AI-age” website invisible to AI answers.
• AI-only SEO misses the same ingredient: fully AI articles fell from 28% to 3% top-100 visibility over 16 months, with no lasting recovery (SE Ranking).
• Five checks, about an hour, tell you whether an AI-built site can be trusted with patients. They are at the end of this article.

The AI site builders are real, and so is the price

Start with what is true in the pitch. The tools are not toys. Lovable reached roughly 8 million users and about $500 million in annualized revenue by May 2026; Bolt passed 7 million users in fourteen months; Wix, WordPress.com, and Squarespace have all rebuilt around AI generation. The buyers are who you would guess: Hostinger reports 93-95% of its AI-builder users are building a website for the first time. (Sacra; Hostinger)

The price is real too. In pitches practices have forwarded to us this year, the anchor is a website for around $1,000 and bundled “SEO plus hosting” near $150 a month. That number does the selling by itself.

So this article will not tell you AI cannot build a website. It can, and we use AI in our own production every day. The question that actually protects your practice is narrower: what happens when nobody checks the work?

What ships when nobody checks the work

Veracode has security-tested AI-generated code across 150+ models for years, and the result has barely moved: roughly 45% fails security tests, stuck near that level for two years even as the code got dramatically better at looking correct. The gap between “it runs” and “it is safe” is widening. (Veracode, Mar 2026)

The failures are not hypothetical. A 2025 disclosure found 170+ apps built on Lovable exposing data through 303 endpoints with missing access controls: names, emails, phone numbers, payment records, readable by anyone who knew where to look. Security firm Escape scanned 5,600 live “vibe-coded” apps: 2,000+ vulnerabilities, 400+ exposed credentials, 175 leaks of personal information including medical records. RedAccess found roughly 380,000 public AI-built apps, about 5,000 of them leaking sensitive data, healthcare conversations among it. (CVE-2025-48757; Escape, Oct 2025; RedAccess, May 2026)

Two details make this a dental problem. Your website’s forms collect protected health information; a patient typing “molar pain, upper left” into a leaky form can become a HIPAA breach, and recent OCR cases against individual dental practices have run $50,000, $62,500, and $70,000. And the person prompting the AI usually cannot see the problem: a Stanford study found developers using an AI assistant wrote less secure code and were more confident it was secure.

None of that argues against the tools. It argues about accountability. A generated site that a qualified person audits, hardens, and signs their name to is a fine product. The same site shipped straight from the prompt is a liability with a nice design.

The AI-built website that AI cannot read

Here is the part of the builder story almost nobody selling one mentions. Most AI builders generate your site as a JavaScript application: the page arrives at the browser nearly empty, and code has to run before any words appear. Patients with browsers never notice. The bots that read websites for AI assistants do, because the major AI crawlers fetch the raw page and do not run JavaScript. They open an AI-built site and see close to nothing. (SEJ practitioner audit, Apr 2026)

Lovable’s own documentation concedes the point: only apps created from May 13, 2026 onward default to server-side rendering, the technique that puts real content in the raw page. Everything built on it before that date ships as a client-side app. (Lovable docs)

Sit with the irony, because it is the whole argument in one image: a practice buys an AI-age website, and the AI systems patients use to research and choose care cannot read it. In part one we said the shift toward patients’ AI assistants reading your site for them is real. It is, and this is how a cut-rate build fails that future silently. There is a free test later in this article; it takes thirty seconds and no technical skill.

AI-only SEO is missing the same ingredient

The website is half the pitch. The other half is the ~$150-a-month promise that AI agents will do your SEO, no marketing team needed. Start with Google itself. Nick Fox, Google’s SVP of Knowledge and Information, at Google Marketing Live 2026: “The way to optimize for AI Search is the same way to optimize for Search: Create great content.” uSERP’s CEO went further in Digiday: any GEO service that will not admit AI visibility is about 80% fundamental SEO “is selling you snake oil.”

Then there is what happens to the content itself. SE Ranking ran the controlled experiment: 2,000 fully AI-generated articles on fresh domains, tracked for 16 months. Visibility fell from 28% of articles in the top 100 to 3%, with no lasting recovery. Graphite’s June 2026 simulation study showed why the ceiling is low even when it works: left unsupervised, 89% of AI responses converged into near-paraphrases of each other by the end of the simulations. Run the same playbook as every other practice’s AI, and the output converges on sounding identical. Differentiation is the one thing the tool cannot generate, because it does not know what makes your practice different unless a human feeds it and checks it.

We also see how this gets sold. This month a practice we track forwarded us an AI-SEO pitch whose automated report claimed they had zero AI presence. Our monitoring, which we treat as directional rather than a lab measurement, shows that practice in roughly three of every four AI answers we track for its market. The same report flagged their schema as missing; it was there, in a format the tool did not parse. That report was not an audit. It was a sales document wearing an audit’s clothes.

Cory’s framing for where this settles: “When everybody starts implementing some version of AI SEO… it’s the human factor and software automation that’s going to separate you out from the competition. And that’s the game we’re playing.”

What we would not do

We would not let AI-built code touch patient data without a named person auditing it first. Not a policy aspiration, an operating rule: the Veracode and exposure numbers above are what “we skipped the review” costs.

We would not publish AI-generated SEO content nobody read. SEO consultant Lily Ray demonstrated the failure mode this spring: she published a fictional Google update on her own blog and watched AI answers repeat it as fact. Unreviewed content does not just underperform; it pollutes the answers patients see.

We would not promise you an AI ranking. SparkToro ran the same brand-recommendation prompts 2,961 times: the same list twice came up less than 1-in-100 runs, the same order less than 1-in-1,000. There is no stable “AI rank” to buy; appearance rate across many answers is the honest measurement.

And we would not sneer at the $1,000 site. The tools are legitimate and mostly serve first-time builders. The line we hold is thinner and harder: reviewed or unreviewed. That line, not the logo on the tool, decides whether a practice gets hurt.

Five checks before you trust an AI-built website

Whoever built the site, these five checks take about an hour and no code. Get the answers in writing, the same discipline we suggest for any website and SEO vendor pitching you coldThis link leads to a Blog post about SEO email spam.

1. Test the login wall. If the site has any patient login or portal, sign in as one test user and try to reach another user’s information. If you can, strangers can.
2. Make them show you the secrets audit. No passwords, API keys, or database credentials sitting in the page code. Show, not promise.
3. Turn off JavaScript in your browser and reload the site. What loads is roughly what AI crawlers receive; a blank or near-blank page means your content is invisible where patients increasingly ask. This holds whether the site is custom or templateThis link leads to a Blog post about custom dental websites vs. templates.
4. Trace one form submission end to end. Who receives it, where it is stored, who can read it. Patient form data is health data.
5. Get a name. The specific person who reviews changes before patients see them and fixes the site when it breaks. “The AI handles it” is not a name.

A vendor who welcomes these five is selling you something real. A vendor who bristles just answered a different question.

FAQ

Are AI website builders bad? No. They are legitimate, widely used tools, and most of their users are building for the first time. The risk is publishing what they generate with no qualified human review of security, accuracy, and voice.

Will Google penalize my site because AI built or wrote it? Not for AI involvement itself; Ahrefs found 86.5% of top-20 results contain some AI content. What Google’s 2025-2026 updates punished is unreviewed volume: AI pages published at scale with no editorial oversight, some sites losing their traffic entirely.

Can $150-a-month AI-only SEO work for a dental practice? AI raised the floor of SEO, and cheap tools now cover basics that used to cost more. The measured ceiling is the problem: fully AI content decayed from 28% to 3% top-100 visibility in SE Ranking’s experiment. Practices compete above the floor, in ongoing, human-directed SEOThis link leads to a Blog post about ongoing SEO.

What is an agentic website again? A site built AI-first: it evaluates its own content and SEO against your priorities and your competitors, recommends what to add or change, and, with a human specialist approving, editing, or rejecting every change, implements the work. Part one covers the definition and the three questions to ask anyone selling oneThis link leads to a Blog post about agentic websites.

How do I find out what AI currently says about my practice? Ask the engines directly, or run our free scan. Either way, read the actual answers, not a vendor’s score; automated reports have called a well-cited practice invisible. AI Overviews and AI answersThis link leads to a Blog post about AI overviews for dentists reward being checked, not being scored.

Whether your next website is built by AI, by people, or both, the AI assistants your patients use already have an answer about your practice, written with no input from you. That is the baseline every decision above should start from. Our free AI Visibility ScanThis link opens a new tab to our AEO website shows what ChatGPT, Gemini, Perplexity, and Claude say about your practice, in about 60 seconds.