What a machine sees when it reads a website.

If you work for yourself or run a small firm, your website is probably the most carefully considered thing you have made. You chose every word. You checked it on a phone. Somebody told you it looked professional, and they were right.

Then you search for yourself, and there is nothing there.

It is natural to put that down to competition, or budget, or simply needing more time. Often that is all it is.

But there is a plainer possibility, and almost nobody checks it, because it does not occur to you to. The machines that decide whether anyone finds you may not be able to read your site at all.

That sounds unlikely. A page that works perfectly in your browser is obviously readable. It is not. Your browser does a great deal of work on your behalf that a machine does not do.

I checked mine properly this month, for the first time. Three things were wrong with the site, and a fourth thing I had left far too late. I had not noticed any of them.

I have no technical background in any of this. That accounts for most of the first three, and it is worth saying plainly, because if you are in the same position these are the faults you are most likely to have too.

The first fault

Writing that was published, and invisible.

My free articles sit on a page that lists each one with a short summary. On screen, they are all there. I have looked at that page a hundred times.

Those summaries were being assembled by JavaScript after the page had finished loading. Your browser runs that code without being asked, so you see the finished result. A crawler often does not. A crawler is the small program a search engine sends out to fetch a page and read it, and many of them take the file exactly as it arrives.

In the file that arrives, that part of my page was empty. Five articles, published, and invisible to anything that did not run the code.

Fixing it had a second effect I was not looking for. Because the summaries arrived late, everything below them was shoved down the page as they appeared. Google measures that shove and calls it Cumulative Layout Shift. My page scored 1, where anything above 0.1 counts as poor. It now scores 0, and the page no longer moves while they are reading it.

The second fault

The best thing I had written, filed where nobody would look.

Somewhere on your site is the thing that makes you worth choosing. For me it is the method: how I check a figure, when I will name a company and when I will not, and what happens when I get something wrong.

All of it was written. All of it sat inside a page called About.

Think about what happens when somebody asks an AI assistant how to tell trustworthy research from the rest. The assistant goes looking for pages that explain a method. A page called About is where a business talks about itself, so that is not where it looks. My best argument was sitting there anyway.

It now has a page of its own, named after what it contains rather than after me.

The third fault

A number that was not a measurement.

I ran an automated check on how visible the business is to AI assistants. It came back with a score of 8 out of 100, and a blunt conclusion: the site had not been indexed in thirty days.

Indexed means a search engine has read your page and added it to the list it looks through when somebody types a question. If your page is not on that list, it cannot appear in results at all.

My pages were on the list, and had been for a week. Google Search Console, which is Google's own record of your site, said so plainly.

What had gone wrong is worth knowing, because the same trap sits inside a lot of tools. The check had not asked an AI assistant anything. It had run ordinary web searches and drawn its conclusions from those.

One of those searches returns nothing for a new domain, even when the pages are there. The flaw takes ten seconds to see. Search for wikipedia and you get Wikipedia results, and the words there are no results, on the same page.

I am not naming the tool, because the fault is not really the vendor's. It is that a number arrives looking like a fact. Before believing a reading, it is worth asking what the instrument actually measured, and whether it could have measured the thing it claims to.

The fourth thing

And one step I had never taken.

The three faults above were things that were wrong with the site. This one is different. It is something I had left too late, and of the four it is probably the most common.

I launched the site on 21 August. I did not tell Google or Bing that it existed until 6 September. Sixteen days.

You hand each search engine a sitemap, which is just a list of your pages, and switch on the free tools they give you for watching what happens next.

I knew it needed doing. It sat on a list behind finishing the report, setting up payments, writing the legal pages, and getting delivery and email working. None of those could wait. This one could, so it did.

So the four faults are two different kinds. The first three I did not know about. This one I knew about, and it still lost to the jobs that had deadlines. That is what running the whole thing on your own looks like.

The site may well have been found anyway in those sixteen days. That is the other half of the problem: I cannot tell. Google Search Console only begins collecting on the day you set it up. Whatever happened before that, I have no record of it, and I never will.

One appearance is not a result. It is not a failure either. It is simply too little to tell anyone anything, and it would be dishonest to present it as either.

Google only began collecting on 6 September, because that is the day I set it up. My two main pages were added to its list on the 6th and the 7th. No other website links to me yet. Six days of data, on a site nobody had been told about until then, gets you about this.

One more thing belongs here, because it pulls the other way. For its first month the site was caught by corporate security filters at some large firms and blocked. Not for anything on it. The main filtering vendors treat any domain registered in the last month or so as a risk in itself, whatever it contains. The registration date is the one that counts for this, not the launch date. Mine was registered on 13 August, so it clears that window on 14 September. That made no difference to how often Google showed the site, but it did mean some of the people I write for could not open it even if they found it.

Two limits on all of this, which matter if you are thinking of copying it. The average position shown beside these figures is an average of one observation, so it is ignored here. And I have checked how one AI tool reads the site, not how several do. Asking the assistants directly costs nothing, because they have free tiers. What I have not done is ask them repeatedly, over time, and write down what came back. That is the difference between an anecdote and a measurement, and it is the same distinction the third fault above turns on.