AI for Small Business Websites: What's Actually Worth Building

Almost every client conversation I have now includes some version of "should we add AI to the site?" It's a fair question — and a hard one to answer honestly, because right now "AI-powered" gets slapped on features that range from genuinely useful to actively pointless. Here's my attempt at a straight answer, based on what I've actually built and shipped, not what's trending.

The AI features that actually pay off

A support chatbot that's scoped correctly

A chat widget trained specifically on your own FAQs, policies, and product or service pages — not a general-purpose chatbot with your logo on it — can genuinely cut response time on repetitive questions: shipping times, opening hours, pricing tiers, "do you serve my area." The key word is scoped. It should know your business and nothing else, and it should hand off to a human (or at minimum, collect contact details) the moment a question goes outside that scope. A chatbot that confidently makes things up about your business is worse than no chatbot at all.

Content and SEO workflow automation

This is where I've seen the most real, measurable time saved: using AI to draft the first pass of meta descriptions, alt text for large image libraries, or product descriptions across a big catalog — then having a human review and edit before it goes live. It doesn't replace the strategy or the editing, but it removes hours of repetitive first-draft writing.

Personalization, at the right scale

"Customers who viewed this also viewed…" or content that adapts based on a returning visitor's behavior is genuinely useful — once you have enough traffic and enough catalog for the data to mean something. For a five-page brochure site or a catalog of a dozen products, this kind of personalization has nothing meaningful to work with, and it's not worth building yet.

Where AI is still overkill — or a risk — for most small business sites

• Fully autonomous customer service with no human fallback. The moment a customer has an unusual request — a complaint, a refund, something contract-specific — an AI with no escalation path just frustrates people and can create real liability if it makes commitments it shouldn't.
• AI-generated blog content published without review. Beyond the accuracy risk, generic AI-written content tends to read as generic, and search engines are increasingly good at recognizing (and not rewarding) content that adds nothing a dozen other pages don't already say.
• Adding "AI" as a label rather than solving a problem. If a feature would be a good idea whether or not AI was involved, build it. If the only justification is "it has AI in it," that's a sign to slow down.

A simple way to decide if an AI feature is worth building

Before I recommend an AI feature to a client, it needs to clear three questions:

• Does it save real time or money, in a way I could estimate in hours or dollars — not just "it feels modern"?
• Is there a fallback to a human when the AI hits the edge of what it knows?
• Would a visitor trust it — or does it feel like a barrier between them and getting an actual answer?

If a feature can't clear all three, it usually isn't ready to ship yet, no matter how impressive the demo looks.

How I approach AI projects

I start small: one well-scoped chatbot, one automated workflow, measured against a specific, current pain point — not a full "AI transformation" from day one. Once something is live and I can see it's actually saving time or converting visitors, it's easy to justify expanding it. That approach has been far more reliable than trying to guess upfront what will work.

The businesses getting real value from AI right now aren't the ones with the most AI features — they're the ones who added exactly one that solved a real, specific problem.

If you're weighing whether an AI feature makes sense for your site, I'm happy to look at your specific situation rather than sell you a generic package — see my AI solutions work or get in touch to talk it through.