Gold In, Gold Out

If you were around computers before they got friendly, garbage in, garbage out was almost a mantra. Bad data, bad code, bad logic equals bad results. It was the first lesson anyone learned about computing.

We Used to Know How to Ask Machines Things

Before Google made it easy, you had search engines like AltaVista and Netscape, and you had to talk to them almost like you were writing code. Using "AND" and "OR" and other pseudo-search-query programming to get the results you were looking for. 

Then search engines started reading page content instead of just scraping keywords and meta tags. You could type a real question in real English and get a real answer. The query-crafting skill quietly died.

AI Brought the Query Skill Back

AI feels conversational. You type like you're talking to a person, and it talks back like one. That conversational feel hides query pitfalls and traps, and makes people think that you just say what you want and you get it.

The problem is AI is confident and capable enough to hand you a complete, polished answer whether or not you asked the right question. It's not about syntax anymore, it's about context — telling the machine not just what you want, but what world that thing has to live in.

The Frog Page Problem

Say you ask an AI to build a web page about frogs. It comes back beautiful. Clean layout, easy to read, nice hierarchy. Absolutely nothing wrong with it as a standalone page about frogs.

Now drop that page into a real CMS. If you didn't tell the AI about the CSS file already running your site - and you better know about your CSS framework if you're dropping AI content into you site - there's a decent chance it wrote all its styles inline. It looks great sitting there by itself. The next page or post — tadpoles, let's say — also looks great and comes out with a completely different look and feel under the hood, because the AI has no idea a "frog page" and a "tadpole page" are supposed to share a design system.

That frog page might not be responsive either. Four columns of tiny text that look fine on your 27-inch monitor is unreadable on a phone. Correct is not the same as appropriate and AI doesn't know the difference unless you tell it.

Frog Pages in Miva Merchant

The same thing applies to Miva Merchant. Asking AI to create a page or a content section or a category header without giving AI a good prompt and the CSS framework - Shadows for example - leads to beautiful but useless content. Be very careful about this in your store.

Gold In, Gold Out

Garbage in, garbage out was about quality. Gold in, gold out is about fit. You can feed an AI a perfectly reasonable request and get back something technically correct and completely wrong for your situation — wrong framework, wrong responsiveness, wrong voice, wrong everything except the part you actually asked about.

The fix is a habit, not a skill:

• Give it the CSS file, and say it needs to be responsive — don't assume either one is implied.
• Tell it what has to stay consistent with other pages, not just this one.
• Show it an example of the voice or style you want matched.
• Tell it who's going to read the answer — a picky boss, a five-year-old, a total beginner so the answer is appropriate for the reader. 
• Tell it what you're going to do with the answer. A recipe you're cooking tonight needs different detail than one you're saving for later.

It's the same instinct AltaVista forced on people thirty years ago: say exactly what you mean, don't make the machine guess. AI is a better guesser than AltaVista ever was and a bad guess used to come back obviously wrong. Now it comes back looking finished while it could be more wrong than ever.

Feed It Gold

AI will always give you an answer. Whether that answer belongs in your world depends entirely on what you handed it going in. Garbage in, garbage out is the same as it always was.

Want a second set of eyes on how AI-generated code is fitting into your site? Get in touch.

Related Articles

These articles might be interesting for you as well: