SEO Pitfalls for AI-Generated Websites
In the past 12-18 months, LLM services went from being a passable assistant to programmers to being on the level of, or even above the level of, junior developers.
More and more developers started using them in their workflows.
More and more companies started integrating AI assistants into their services.
But even today, these AI assistants have a flaw – they are function-oriented and will often miss some ot the substance.
And they are very prompt-dependent.
Let’s look at an experiment – we asked several popular AI services to create a simple website, which will be hosted in a shared hosting account. The site would be about a flower delivery business, with just 3 pages – a Home page, an About Us page, and a Contact Us page – nothing special.
We asked all AI services to list the tech stack they will be using before they generate any code.
Here is how they fared:
- Gemini 3.1 Pro went straight to React
- ChatGPT 5.5 suggested React if we wanted more features down the road.
- Claude Sonnet 5 suggested React if we wanted more features down the road.
- Mistral Fast suggested static HTML5
- Qwen 3.7 Plus suggested static HTML5
The top 3 AI services all recommended React.
Which in itself is not an issue – React is very quick, and with it, you can create beautiful websites. And React code and docs are heavily featured in the training data for most LLMs.
However, by default, React is client-side generated.
What does this mean?
It means that the site is built dynamically in the visitor’s browser; it’s not built beforehand on the server.
And while this is not a problem for any human visitors, it’s a problem for most of the bots that crawl the web.
Like GoogleBot. Like the Bing bot. Like AI crawlers.
GoogleBot will try to load the JavaScript content, but since it’s client-rendered, it will queue the generation for a later time after it has crawled the page. This time may come or may not come at all.
BingBot is not good at handling JavaScript.
The social preview scrapers run no JS at all, and the AI crawlers (GPTBot, ClaudeBot, PerplexityBot) mostly don’t execute JS either.
So, you end up with a site that loads up quickly, that looks amazing, that works perfectly across all devices, but is basically just a blank HTML document in the eyes of all search engines and services.
Here is how to avoid this pitfall:
Explicitly state which tech stack the AI service must use
Depending on the type of website, you can go with just a simple combo of HTML or CSS. You can go the classic route with PHP and MySQL / MariaDB / PgSQL.
You can also choose a JavaScript framework that uses server-side rendering (SSR). React supports SSR, but you will have to ask the AI service to generate the website specifically with SSR in mind.
Whatever you choose, you have to make sure that the end result is a server-side rendered HTML file that has all the content in it.
You can quickly test if GoogleBot sees your code: Google Search Console -> URL inspection, Test Live URL -> rendered HTML tab. If your content isn’t there, Google isn’t seeing it either.
Refine your prompt and better state your goals
When we say that AI services are function-oriented, we mean that when you say “I want a website”, they will give you a website.
But what you mean when you say “I want a website” is more likely: I want a website that loads up quickly, looks amazing on mobile, can be discovered by Google and the rest, is simple to add on to when I want to expand, etc.
Some AI services will try to fit most of what they assume you mean, but most of them will just go right ahead and finish a clearly stated task without too many additional questions.
So, it’s important to figure out exactly what you want before you click the proverbial “let’s go” button.
Spend more time in the planning stage, and ask the AI to create “to-do” lists so that you can monitor your progress.
Ask the AI how to best structure your prompt in order to achieve the stated goals.
Consider using a CMS
Most people rush to AI-generated websites because they can just describe the type of website they want and the machine will do all the work for them, while traditional CMS platforms require more involvement.
However, there is a reason why such CMS platforms (like WordPress, for example) are so popular – they are battle-tested, and they are proven to work.
In addition, in recent years, most popular CMS platforms have fully embraced AI and are compatible with popular AI agents like OpenClaw, Hermes Agent, or Claude.
And the agents themselves know very well how to work with the CMS platforms, so you can use an agent to manage your site for you from your messaging app.
Consider a static website
Not all websites need to have a backend or admin area. Sometimes, having just a few static pages is more than enough.
Some websites don’t need more than a description of the services they offer and a way to get in touch.
These websites will work much faster and will be a lot simpler to maintain.
And that is before any type of caching is involved, which can make a static website load almost instantly.
Originally published Wednesday, August 26th, 2026 at 7:49 am, updated August 26, 2026 and is filed under SEO.
Leave a Reply