I’ll start with a confession, because that’s how I start most things on here.

I have opinions about websites. Lots of them. Speed matters, clarity matters, trust matters, and I’ve said all of it in meetings with a very straight face. Every agency does this. Every agency knows what makes a website work. Me included.

The problem with opinions is that they’re free. Most of mine came from the last few projects I worked on and whatever I read that morning. So this autumn we did something boring. We stopped guessing and started counting.

We took 100 B2B software companies, 50 in the US and 50 in the UK, drawn at random from public growth lists. None of them are our clients. We ran the tools a marketing team would run. Then we checked what those tools said against three things they don’t look at: what real visitors get, what the browser sends before anyone clicks the cookie banner, and what ChatGPT, Gemini and Perplexity say when a buyer asks for a recommendation.

I want to tell you what we found. Mostly I want to tell you what it did to my head.

Start with speed. Google’s lab test, the one behind PageSpeed Insights, the red number your developer sends you with a sad face, called 88 of 99 homepages slow. Then we looked at real Chrome users on the same sites. On 54 of them, real people got a fast page.

The median homepage took 12.2 seconds in the lab. Real visitors waited 2.2.

I don’t want to know how many budgets went into chasing that red number. Don’t get me wrong, the lab test has a job. It finds the heavy stuff that will hurt someone on an old phone. But it tests a simulated slow phone, and your customer is a person.

Then the uncomfortable one. We opened every homepage from Europe and recorded everything it sent before anybody touched the cookie banner. 39 of 100 were already sending data to LinkedIn or Meta. And 31 of those 39 had a consent platform installed.

The banner was there. The pixel just didn’t wait for it.

I’m not a lawyer and this isn’t a legal opinion. We report what the browser sent, nothing more. But here’s what bothers me. No speed score, no SEO tool, no shiny audit PDF will ever flag this, because none of them looks. Somebody paid for that consent banner. Somebody ticked the box. And everybody went home thinking it was done.

That’s the part that gets me. The feeling of done.

Now the question everybody asks me lately. Will ChatGPT recommend us?

So we asked. For every company, two buyer questions, each one twice, to all three assistants. When the question named the broad category, something like “best security software for mid-size companies”, 31 of 100 got recommended. When the question named their exact niche, in the words from their own homepage, 93 of 100.

Same companies. Same assistants. Different question.

And the assistants knew these companies. Asked directly what a company does, they described it correctly in most answers and never mixed it up with somebody else. So the AI wasn’t the problem. The problem was the way these companies describe themselves, wich is rarely the way a buyer types.

Which brings me to the headlines. We read the big line at the top of every homepage. On 33 of 98, it said what the product is. On the other 65 it was a slogan or a benefit. One of them, word for word, was “AI for what’s next.”

I laughed at that number and then I stopped laughing, because I know exactly how those headlines get written. Somebody in a meeting says we don’t want to sound like everyone else. Everybody nods. And the line that would have told a stranger what you sell gets swapped for a line about how you feel.

The expensive problems are the ones nobody scores.

Read that twice.

A scanner gives you a number, and a number feels like control. You fix the number, the number goes green, you sleep well. Meanwhile the pixel is firing before consent, the AI can’t place you in your own category, and the headline is a mood.

One more, because it made me smile. We asked ChatGPT the same niche question twice. On 30 of 100 companies it changed its verdict. So if somebody on your team asked once and came back happy, congratulations, you have a sample of one.

Every website tool gives you a score. None of them tells you what the score misses, and that’s the stuff that costs money.

And before you jump at my neck: yes, I run an agency, and yes, we sell website audits. I know how this sounds. That’s exactly why the study page has the method, the limits and the numbers that didn’t flatter anybody. Check our work. Please. That’s the whole point of counting.

If you want to check your own site, you don’t need us and you don’t need to buy a tool. Give it an afternoon:

• open Search Console and look at what real visitors get, before you touch a lab score
• open your homepage in a private window, open the network tab and don’t click the banner. Look for LinkedIn and Meta
• read your headline out of context. Would a stranger know what you sell from that line alone?
• ask three AI assistants about your niche in your buyers’ words, twice each. Then ask about your category. Write down the gap
• look at your profiles on G2 and Capterra. The assistants read them, so they’re part of your website now, like it or not

The full study is here: every number, the method, the limits. It’s the first edition of a series we call The Website Gap. Next one is private clinics, in November. Same rules, 100 websites at a time, untill we run out of industries or patience.

Stop polishing the number. Start checking what the number can’t see. And if you find something ugly on your own site, good. It means you looked, and most people don’t.

Keep hustling, stay humble, follow your dreams.

— Nicu