Twenty minutes to find out what a machine knows about your business
Somebody asks an assistant for a good accountant in your city. It answers in a paragraph, and your business is either in it or it is not. You can find out what that machine has to work with in about twenty minutes, without a consultant and without learning anything technical.
Here is what changed, in one sentence. For twenty years the job was to get somebody to click. Now a machine reads your site first, and what it can carry away decides whether the person ever hears your name.
The uncomfortable part is that this is not a design problem. Sites that look excellent fail this, and plain sites pass it. We measured 54 therapy practice websites and found the whole field could be read easily by machines and quoted by almost none of them. Nothing about that is visible from looking at the pages.
So look at what the machine looks at. Set aside twenty minutes and do these five things on your own site.
First, read your own homepage the way a machine does
Open your homepage, right click anywhere that is not an image, and choose View page source. You are now looking at what gets sent before any of the site runs.
Use your browser's find function and search for a sentence you know is on the page. If it is there, good. If your headline, your services and your description are missing from that view, they are being drawn after the fact by JavaScript, and a share of the machines reading your site will never see them.
You are not checking whether the site works. You are checking whether the words exist before the site does anything.
Second, look at the file that tells crawlers what they may read
Type your domain into the address bar and add /robots.txt at the end. Most people have never seen this file and it takes ten seconds to read.
You are looking for two things. Any line that says Disallow: / closes off everything below whichever agent it sits under, so check which name it follows. And if you see a list of names like GPTBot, ClaudeBot or Google-Extended, notice whether anything under them is actually blocked.
In the practices we measured, thirty sites named those agents and twenty blocked none of them. That is a hosting platform shipping a panel with the switch left off. The point is not that the setting is wrong. The point is that almost nobody knows it is there, in either position.
Third, find out whether your site says a human being works there
Back in the page source, search for application/ld+json. That is the structured data, which is the part of your site written for machines rather than people.
Inside it, look for the name of a real person, a street address and a list of links to your profiles elsewhere. More than half of the sites we measured named no person at all in that data. More than half linked nothing outside their own domain, which leaves the machine with a business whose existence is confirmed by exactly one source: itself.
If you find nothing at all under that search, that is a finding too, and it is the cheapest of these to fix.
Fourth, time yourself finding your own price
Start from your homepage and try to reach a number. Any number: a rate, a range, a starting from, an hourly figure. Count the clicks.
If you cannot get there in two clicks, neither can a stranger, and neither can a machine summarising you against three competitors who do publish one. Eighteen of the 54 practices we measured published no price anywhere we could find.
This is the check that owners argue with most, and the argument is always some version of it depends. It usually does. A range with the conditions attached still beats an empty page, because the alternative is not a better conversation later. The alternative is not being in the comparison.
Fifth, ask an assistant about your own business and read the answer coldly
Open ChatGPT, Claude, Gemini or Perplexity and ask what it knows about your business by name and city. Then ask it to recommend somebody in your category in your area.
Read the answer for three things: what it gets wrong, what it leaves out, and where it says it found you. Wrong facts usually trace back to an old directory listing or a profile you forgot. Absence usually traces back to the four checks above.
Do this one twice, on different days. These systems are not consistent, and one bad answer is an anecdote rather than a measurement.
What the answers actually mean
| What you found | What it means | Roughly how hard to fix |
|---|---|---|
| Your words are missing from the page source | Part of your site is invisible to anything that does not run JavaScript | Depends on the platform, sometimes a rebuild |
| Robots file blocks agents you did not choose | A default is making a business decision for you | Minutes |
| No person named in the structured data | The machine has a company but nobody accountable for the work | An hour |
| No links to profiles elsewhere | Nothing outside your own site confirms you exist | An hour |
| No price reachable in two clicks | You are left out of comparisons rather than losing them | A decision, then ten minutes |
| The assistant gets your facts wrong | Something stale outranks your own site as a source | Hours, spread over weeks |
What to do with a bad result
Very little of this is a website project. Four of the six rows above are an afternoon, and they are the four that move the most, because they are the ones a machine can actually act on.
Start with the cheapest: name a human being, link the profiles you already have, publish a number. Then write one page that answers a real question in its first sentence instead of its fourth paragraph, which is the difference between a page that reads well and a page that can be quoted.
What we would leave for last is the rebuild, and most of the time it turns out not to be necessary.
Questions
Do I need any tools to do this?
No. A browser is enough for four of the five steps, and a free assistant account covers the fifth. Anything that asks you to pay before showing you your own site is selling you a report you can produce yourself in twenty minutes.
My site looks great and my clients like it. Can it still fail this?
Easily, and that is the most common case. Design quality and machine readability are unrelated. Some of the worst results we have measured came from expensive, genuinely beautiful sites, because visual polish often arrives through the exact techniques that leave nothing in the served page.
Is this the same thing as SEO?
It overlaps and it is not the same question. Search asks which page deserves the click. An assistant asks what it can say about you without sending anybody anywhere. A business can rank well and still be unusable to the second kind of system.
How often should I repeat this?
Twice a year is plenty for most businesses, plus once after any site change, because the checks that break silently are exactly the ones a redesign breaks.
What if I find something wrong and my web person disagrees?
Ask them to show you the same view you looked at rather than describing it. Every one of these checks produces something you can both see on a screen, which is the reason they are worth doing yourself.
The short version
Twenty minutes, five checks, no tools. Your words in the page source, your robots file, a named human in your structured data, a price within two clicks, and what an assistant already says about you. Most businesses fail two or three of them, and most of the failures are an afternoon of work rather than a project.
If you would rather have somebody run the full version of this and hand you the list in priority order, that is what we do. What it covers and what it costs is on the services page, with more detail on the FAQ page. The call is thirty minutes.
