The upgrade we did not buy

Our hosting company put a new feature in front of us this week. It converts your website into a simpler format for AI assistants to read. When something like ChatGPT looks at your site it gets clean text rather than a page full of menus and buttons. It sits on a paid plan.

It is a good feature and the reasoning behind it is sound. We did not buy it, and working out why took about fifteen minutes.

What was actually being sold?

Stripping the furniture off a web page. A page a person reads is mostly not the words: it is the navigation, the footer, the buttons, the styling. A machine reading that page has to work out which part is the actual content. The feature does that for you and hands over the clean version.

If your website is heavy, built on a page builder, or puts its text behind a pile of scripts, that is a genuine problem and this genuinely solves it.

How do you tell whether an upgrade is for you?

Three questions, in order, and the order matters.

What problem does this solve? Then, do I have that problem? Then, do I already own the parts?

Most sales conversations only cover the first one, because the answer is always yes, the feature does something. The second question is where most upgrades quietly fail. We have watched clients pay monthly for tools that solve a problem they have never had, bought during a week when the problem seemed likely.

So why was the answer no for us?

The third question. Our service pages are written as plain text files and turned into web pages when the site is built. The clean version the feature promised to produce was, in a real sense, where the page had started.

Paying to convert the finished page back into roughly what we had written to begin with is a strange trip to take. So we published what we already had. It took an hour, it costs nothing every month, and each page is now available in both forms.

The difference is not small. Our websites page is about twenty kilobytes as a web page and under five as plain text. Same words. The other fifteen kilobytes are menus, styling and the machinery that makes it look like a website.

There was a second reason, and it turned out to matter more. An automatic converter has to guess what counts as content. We do not have to guess. We know the answer paragraph near the top of each service page is the passage worth quoting, and we know the navigation is not. When we built the plain version, we put the useful part first and left the furniture out entirely.

Is the free version always better?

No, and we would be selling you something if we said otherwise.

We pay for our hosting. We pay for email. We pay for the tool that runs our booking calendar, and building any of those ourselves would be a poor use of a week. The test is not price. It is whether the thing sits inside what you already do or outside it.

Converting our own writing sits inside. Running a mail server does not.

Does any of this even work?

Here is the part the sales page will not tell you, and neither will we: nobody knows yet how many AI assistants ask for the simpler format. It is a young convention. We think it is a sensible bet because it costs us nothing to keep, and we would think differently about it at a monthly fee.

That is worth applying to almost everything being sold with AI in the name at the moment. A cheap bet on an unproven thing is fine. A subscription to an unproven thing is a decision you have to keep making, usually without ever finding out whether it worked.

What to do with this

Next time an upgrade appears in your dashboard, ask the three questions before you look at the price. What does it solve, do I have that, and do I own the pieces already.

If you get to the third question and you are not sure what you own, that is worth an hour of somebody’s time to find out. It is a common place for small businesses to be paying twice, and dashboards do not volunteer the information.