ParsonsHosting

AI for websites and search

How is AI used to build and maintain websites, and where do people still need to check the work?

By Chris Parsons, founder of ParsonsHosting. Published .

AI tools can now draft page layouts, write and explain code, suggest fixes and generate text and images for a website. They save time on first drafts and routine tasks, but they can produce insecure code, wrong facts, inaccessible pages and content with unclear ownership. A person who understands the site has to review anything AI produces before it goes live.

Smiling diners share food on a busy restaurant patio under string lights while the owner greets guests, with pines and a snowy Cascade peak at sunset.

Where AI helps

  • First drafts of pages and layouts. Describe the business and get a starting structure to edit.
  • Writing and explaining code. Coding assistants suggest HTML, CSS, PHP or JavaScript and explain code you inherited, such as an old theme's functions file.
  • Troubleshooting. Paste an error message (with passwords and personal data removed) and ask what it means.
  • Routine content work. Alt text, meta descriptions and translation first drafts.
  • Chatbots for visitors. Answering common questions after hours, with a handoff to a person.

The risks, and what to check

Security

The OWASP Top 10 for LLM Applications (2025) lists risks such as prompt injection, sensitive information disclosure and improper output handling, which is when AI output is trusted and passed into a page or system without validation. GitHub's own Copilot documentation tells users to review suggestions before accepting them and to check they are "free of errors or security concerns."

Check: never paste AI-written code straight into a live site. Look for unescaped user input, SQL built from strings, hard-coded passwords or API keys, and file uploads without type checks. Test on a staging copy first; staging is included with every ParsonsHosting website.

Accuracy

Google's guidance notes that generative models "don't retrieve facts, but predict a likely sequence of words," so output can contain errors, and it calls manual fact-checking "critical." OWASP lists misinformation as its own risk.

Check: every price, phone number, address, opening hour, license number and legal statement against your own records. AI tools sometimes produce plugin names, settings or features that don't exist; confirm them in the official docs.

Accessibility

AI-generated layouts can look fine while failing WCAG 2.2, the W3C standard for accessible web content: missing form labels, low color contrast, headings out of order, or alt text that describes the wrong thing.

Check: tab through the page with a keyboard only, run an accessibility checker, and read the alt text yourself. Only you know what a photo of your job site actually shows.

Licensing and ownership

The U.S. Copyright Office's January 2025 report on copyrightability says material generated wholly by AI is not protected by copyright, while human creative contributions (such as selection, arrangement and modification) can be. Code assistants can also suggest code that matches public code with its own license terms.

Check: keep your human edits meaningful, save notes on what you changed, and check the license of any code or image you didn't write. This is general information, not legal advice.

Chatbots and disclosure

The EU AI Act (Article 50(1)) requires AI systems that interact directly with people to be designed so people know they are talking to an AI, unless that is obvious. Telling visitors up front is a sensible habit wherever you are.

Check: label the chatbot, give it a way to reach a human, and don't let it quote prices or policies you haven't approved.

Myth vs fact

Myth Fact
"AI-written code is ready for production." GitHub's own docs say to review Copilot suggestions and check they are free of errors or security concerns. OWASP lists trusting unvalidated AI output as a top risk.
"AI will replace web developers." The U.S. Bureau of Labor Statistics projects employment of web developers and digital designers to grow 5% from 2025 to 2035. It also says AI tools "may soften" that growth and let non-developers do basic tasks. The work is shifting toward review and judgment, not disappearing.
"If AI made it, it's accessible." Nothing in WCAG 2.2 is met automatically; it has to be tested.
"I own whatever the AI generates." The U.S. Copyright Office says purely AI-generated material is not copyrightable; your human contributions can be.

Where people must stay in charge

Task AI can draft A person must
Code and plugins Suggestions, explanations Review, test on staging, approve
Business facts Wording Supply and verify every fact
Images and alt text Drafts Confirm accuracy and rights
Security settings Explanations Decide and apply
Customer replies Templates Answer anything about money, health, safety or legal matters

A simple workflow for small sites

  1. Make a backup or staging copy before any AI-assisted change.
  2. Give the tool context: platform, version, what you want, and what must not change.
  3. Ask it to explain what its code does, line by line.
  4. Review and test on staging, including on a phone and with a keyboard.
  5. Fact-check all text against your own records.
  6. Push live, then check the page again.
  7. Keep a short change log, so you or our team can undo it.

AI doesn't know your server

An AI tool doesn't know your PHP version, which caching is in front of your site or how your email is set up unless you tell it. Check Tools > Site Health in WordPress (or your cPanel) for your real versions and include them in your prompt.

Get help

If an AI-suggested change caused a problem, open a ticket at parsonshosting.com/submitticket.php with what you changed and when. Support is ticket-first, 8 AM to 5 PM Pacific, 7 days a week. See WordPress hosting and the FAQ.

Sources

See all articles