We built a tool that checks websites. It found a problem on our own.

A few nights ago we ran our own Website QA tool on our own site.

Not to show off. Because that's the rule here — we don't ask anyone to trust a check we haven't run on ourselves.

The machine ran. Two findings came back. The first one made sense. The second one made us laugh, then stop laughing.

"You have zero Open Graph tags."

Open Graph tags are the little bits of code that decide how your link looks when someone shares it. LinkedIn, WhatsApp, Facebook — they read these tags and show a nice card: your title, your picture, your words.

The machine said we had none. Zero. Nada.

We know we have those. We wrote them. They're sitting right there in the homepage code like obedient little soldiers.

So we did what any sensible person does when a machine tells them something absurd. We opened the source and counted.

There they were. Eight of them. og:title, og:description, og:image — all present, all correct, all working.

So the machine was wrong. Or was it?

The odd part? Our tags. We write them slightly differently from what most checkers expect: property=og:title without quotes, and the content first. It's valid HTML. Browsers read it perfectly. Humans can't see any difference.

But the tool's detector had one specific pattern in mind, and our tags didn't match it. So it counted nothing, shrugged, and reported a clean miss.

A false positive. The machine saw what it was looking for, not what was actually there.

This is the moment most people would quietly delete the tool and never speak of it again.

We fixed the detector instead. That night. It now reads tags regardless of attribute order. We re-ran it. Eight tags found. The false alarm was gone, and the machine learned something it didn't know an hour earlier.

Why are we telling you this?

One. Tools are not truth. They're opinions with a confidence score. A tool is like a very eager intern — brilliant, fast, and absolutely certain about things it got wrong. The machine doesn't lie. It reports exactly what its code sees. If the code has a blind spot, the machine confidently reports the blind spot as fact. That's not malice. That's software being software.

Two. This is why we say every finding on our site is verifiable. If we tell you your title is too long, you can check it yourself in thirty seconds. If we tell you your site is missing something, you can look. We built the machine to be checked — because a tool you can't verify is just a very confident opinion.

Three. If an agency hands you a shiny automated audit, ask them one question: did you check that the tool's assumptions match my site? Because the difference between a report that helps and a report that misleads is often a single line of code that doesn't recognise how your site was built. The confident-sounding report is the dangerous one.

The honest footnote

Our tool still has blind spots we haven't found yet. Every fix teaches us where the next one might be hiding. That's fine. We'd rather say "we found another one and fixed it" than pretend the machine is perfect. If a tool never admits it was wrong, it's not honest. It's just polished.

Want to see if our machine gets confused by your website? Run the free check. Ninety seconds, no signup, no email walls. And if it flags something that isn't actually wrong — tell us. That's how the machine gets better. That's the whole point.

Want to know how your own website measures up?

Run the free Website QA — ninety seconds, no signup, honest results. Or ask Sage on the bottom-right of any page.