I have kvetched in these pages in the past about the shortcomings of artificial intelligence: how one AI site led me down the primrose path with a totally fabricated story about an important historical figure and how miscreants used AI to try to flatter me into paying for services they had no intention of ever performing. So it seems only fair that I give AI its due when it does me a solid.

That happened this past week when I decided the website I created more than a decade ago to showcase my books had gotten a bit long in the tooth. It didn’t adapt to phone- or tablet-sized screens, and it was getting too cluttered. I knew it was time for a new site and, after a little market research, resigned myself to spending thousands of dollars to engage a pro.

But then I remembered reading somewhere that AI wasn’t half bad at web design. So, without any expectations, I decided to query ChatGPT about the matter. I pointed it to my existing site’s URL, told it what changes I had in mind, and asked if it thought it could do better. It was only too happy to shoulder the task.

“Yes — I can take a crack at it. I’ll first look at the current site structure and content,” the app promised, “then suggest an update-friendly redesign rather than just a prettier version of the same template.” Nothing wrong with that. After it got a sense of the kinds of books and articles I write, it told me, it would propose an approach “tailored to your actual content.”

And it was as good as its word. I expected that writing code would be in its wheelhouse; after all, it lives in a computer. So, I wasn’t surprised when it suggested a button here, a drop-down menu there. But it proved capable of far more than coding and layouts. It turns out, ChatGPT has opinions about pretty much everything.

For example, when I uploaded the two author photos I was considering using, it was unequivocal. The second one was better because I had a “more natural smile” and because it looked “more like a current author portrait and less like a corporate headshot.” It hated the idea of putting my full bio on the homepage, insisting it would be far better to limit it to 200 words and add a “Read More” button after that. Why? Because “most visitors won’t read a long biography immediately, but researchers, journalists, and event organizers often will.”

It didn’t just organize the links to my books and articles. It actually read the articles and, if not the books themselves, then the capsule descriptions of them. It thought about them. It suggested ways to categorize them along substantive lines. It chose which books to feature because “the books become the stars of the homepage, which is exactly what an author’s website should do.”

And it even suggested the wording for a pithy summary of my interests as a writer for the “hero section” of my homepage — the large, eye-catching area below the navigation bar — and it wasn’t half bad. I wound up using it.

Amazing. The damned app actually “got” me.

The process wasn’t without its annoyances, however. Sometimes, ChatGPT overthought an issue, went off on a tangent, or babbled on about things I didn’t quite grasp; I had to refocus it on the task at hand. It would tell me more than once what it was going to do without actually doing it, but it usually got religion after I shot back, “Less talk, more action.”

One evening, it informed me it couldn’t possibly get me the latest set of changes until the following morning. But when they didn’t arrive on schedule and I complained, it told me it did not work on projects unless the user was in active dialogue with it. I accused it of lying to me, and my tantrum somehow managed to produce the update immediately.

We also had our share of disagreements. I didn’t like the white space next to one of the images; it didn’t see it as a problem. It wasn’t in love with a few images I uploaded to accompany my articles and told me in no uncertain terms which ones needed to go. We even argued over the Oxford comma. But in the end, we were able to get to “yes” without unpleasantness. AI, it seems, is not above conceding a point when the situation requires it. It knows when not to die on a particular hill.

There were times when our lively back-and-forth seduced me into believing I was actually having a conversation with another human being. It helped, of course, that it was so damned agreeable and complimentary. “That’s a good catch,” it would remark if I pointed out something it had overlooked. “Of all the suggestions you’ve made recently, these two are among the strongest,” it observed without any prompting from me. Sure, this occasionally crossed the line into obsequiousness, but who doesn’t like being told he’s right, even by a machine? 

In the end, ChatGPT did a masterful job on my new site, and it took only two days. See it for yourself. We’ve had our differences in the past, but I think AI deserves kudos this time. And — no small matter — the price was right.

Scott D. Seligman is the author of a dozen books, most of which fall into the category of historic narrative nonfiction. He has a special interest in the history of hyphenated Americans.