By now you know where I land on this. AI is not the enemy of a good website. Treating it as a replacement for strategy is.
So let’s end this series somewhere practical: where AI actually helps, where it quietly does damage, and how I use it in my own process without letting it make the decisions that should stay human.
Where AI earns its place
AI is genuinely useful for speed on tasks that do not require judgment about your specific client. Drafting a first-pass outline. Generating variations of a headline you already know the strategy behind. Summarizing research. Cleaning up a rough paragraph you already wrote in your own voice. Used this way, AI is a fast assistant working from direction a human already set. It saves hours. I would not want to work without it for those pieces.
Where it quietly costs you
The damage happens when AI is handed the parts of the process that require actually knowing your client: what to say, in what order, to move a specific skeptical stranger toward booking a call. That is not a drafting task. That is a research and strategy task, and it depends on information AI does not have access to: the objections you hear on sales calls, the words your best clients use to describe their problem, the moment in a conversation where someone decides to trust you.
Hand that part to AI and you get a website that sounds plausible and converts nobody in particular, because it was not built for anyone in particular.
Even the people building AI tools agree
The Baymard Institute built its own AI-powered auditing tool, UX-Ray, and made a deliberate choice: restrict it to only the roughly 20 to 25 percent of their usability guidelines where the AI can hit 95 percent accuracy against human evaluation. Everything else, the actual judgment calls about whether a design choice helps or hurts, stays rule-based and human-reviewed. Cofounder Jamie Holst put it this way to the Nielsen Norman Group in January 2026: asking a language model to judge whether something is good or bad, rather than just identify what it is, is “where things go off the rails.”
If the people who build AI evaluation tools for a living do not trust AI with judgment calls, that tells you something about where the line belongs on your own website.
The order that actually works
In the SEEN method, this is why research always comes first, before a single word of copy gets written. We find out exactly who you are trying to reach and what they need to hear. Then messaging gets built around that person specifically, in language they would use themselves. Only after that does design come in, sequencing the page so a visitor is walked from “interesting” to “I need to book this” in the right order.
AI can help execute pieces of that process faster. It cannot replace the research, and it cannot replace a human deciding what actually belongs on the page and what does not. That judgment is the whole job. Everything else is typing speed.
The question worth asking before your next update
Not “should I use AI on my website.” Use it where it saves you time on execution. The real question is narrower: is a human who understands my specific client making the decisions about what this page says and why, or is a tool making the average kind of decision it makes for every business that typed a similar prompt.
One of those builds a site that looks like everyone else’s. The other builds the one that gets you booked.
If you rebuilt your website tomorrow, which parts would you trust AI to speed up, and which parts would you insist stayed entirely in human hands?
Sources: Kate Moran, “Demand Accuracy from AI: Lessons from Baymard Institute,” Nielsen Norman Group, January 30, 2026, featuring commentary from Baymard Institute cofounders Christian and Jamie Holst.