Measure Later. Fix First.

There is a particular kind of email arriving in small business inboxes at the moment. It says your brand is invisible to AI. It offers a dashboard. On the dashboard there are numbers: share of model, citation share, inclusion rate, prompt-to-conversion attribution. The numbers are real numbers. Somebody measured something.

I have stood at doors for a very long time, and I recognise the shape of this. It is the sale of a mirror to a man who has not yet washed his face.

Let me be precise about what I am not saying. Measurement is not a scam. The metrics on those dashboards describe something that genuinely exists, and for a company with a marketing team, a category to defend, and competitors worth watching by name, tracking them is sensible work. If you have twelve people whose job is demand generation, go and buy the dashboard. This article is not for you.

It is for the person running a physio practice in Zürich, or a joinery in Bristol, or a two-person accountancy that has been quietly good at its job for eleven years. For that person the tracking-first approach is not merely premature. It produces a specific, expensive kind of paralysis.

The problem with measuring first

Say you buy the monitoring. You get your baseline. It tells you that when someone asks an assistant to recommend a physiotherapist in your city, you appear in eleven percent of answers.

Now what.

The number is true and it is useless, because it does not contain a next action. You cannot fix eleven percent. You can only fix the things that produce it, and the dashboard was not built to show you those. It was built to show you the score.

There is a second problem, less obvious and more annoying. Answers from AI assistants are not stable. Ask the same question twice and you may well get two different sets of recommendations, and two different sets of citationsCitation: A citation is when an answer engine names your business or links to your site as the source of what it just told someone — the AI-era version of being recommended. Read the full entry in the glossary.CitationA citation is when an answer engine names your business or links to your site as the source of what it just told someone — the AI-era version of being recommended.Read full entry behind them.

Research through 2025 and 2026 has put numbers on this repeatedly: only around thirty percent of brands hold their visibility from one answer to the next, and roughly twenty percent survive across five runs of the same query. A single reading tells you almost nothing. A trend tells you something only after months of readings.

Which is to say: you have bought a subscription to watch weather.

Meanwhile your services page is a PDF.

What is actually wrong, usually

I look at a great many small business websites. The failures are not exotic. They cluster, and they cluster in the same places, and most of them have been sitting there since the site was built.

The site tells AI crawlers to go away

A line in a file called robots.txtrobots.txt: A plain-text file at the root of a website that tells automated visitors — search crawlers and AI bots — which parts of the site they may visit and which to leave alone. Read the full entry in the glossary.robots.txtA plain-text file at the root of a website that tells automated visitors — search crawlers and AI bots — which parts of the site they may visit and which to leave alone.Read full entry, put there by a theme, a plugin, or a developer three years ago who was thinking about something else entirely. It blocks the assistants outright. Nothing else on this list matters until it is fixed, because everything else assumes someone can read the page.

The content needs a browser to appear

Some sites assemble their text only after code runs. A person's browser runs that code without complaint. Machines reading at scale often do not. The page looks perfect to you and arrives empty to them.

The services live in a PDF

This is the most common one and the most frustrating, because the owner has done the work. The treatments, the prices, the specialisms, all written out carefully, then put behind a download link. What is inside that file is largely invisible to the thing deciding whether to recommend you.

Nothing on the page says where you are

No city, no address, no plain text mention of the place you actually operate. If anyone ever looks for a business like yours by location, and for most local trades that is nearly everyone, you have made yourself hard to place.

The page explains before it answers

Three paragraphs of context and then, eventually, the point. Assistants tend to quote whatever answers the question first. Bury the answer and something else gets quoted.

The content does not cover what people ask

Of everything on this list, this one moves the needle hardest. When someone asks an assistant about a physio, they ask about specific things: sports injuries, back pain, whether you take insurance, how long a session runs, whether you do home visits. If your site never answers those questions, it does not get chosen, and no amount of monitoring will tell you which answer was missing.

Notice something about that list. Every item names a thing you can go and change this afternoon. None of them requires a subscription to discover. All of them are on your own website, under your own control, sitting there whether or not anyone is watching.

Two of these you can fix in the next ten minutes

Not everything on that list is quick. Writing the answers your customers actually want takes real work. But the two access failures are usually a matter of a few lines, and they are the ones blocking everything else.

First, check whether you are turning the assistants away. Open your browser and go to your own domain with /robots.txt on the end. If you see something like this, you have found your problem.

# yourdomain.com/robots.txt
User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: PerplexityBot
Disallow: /

Three assistants told, in plain language, that they are not welcome. Every page you have written is invisible to them, and no dashboard on earth will improve that number until the line changes.

What it should say instead is this.

# yourdomain.com/robots.txt
User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

Sitemap: https://yourdomain.com/sitemap.xml

Whoever manages your site can make this change in under a minute. If nobody manages your site, most website builders keep this under a settings page called something like search engine visibility, and the fix is a checkbox.

Second, tell machines plainly who and where you are. This one is a small block of JSON-LD schemaJSON-LD Schema: A small block of machine-readable code you add to a web page that spells out the facts a page is making — who you are, what you sell, where you are — in a format built for machines to read cleanly. Read the full entry in the glossary.JSON-LD SchemaA small block of machine-readable code you add to a web page that spells out the facts a page is making — who you are, what you sell, where you are — in a format built for machines to read cleanly.Read full entry that sits in the invisible part of your page. It is not glamorous and it is not the highest-impact thing you will ever do, but it is cheap, and it removes any ambiguity about your name, your address, and your trade.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "LocalBusiness",
  "name": "Beyond Physio",
  "description": "Physiotherapy practice specialising in sports injury rehabilitation.",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "Baslerstrasse 12",
    "addressLocality": "Zürich",
    "postalCode": "8048",
    "addressCountry": "CH"
  },
  "telephone": "+41 44 000 00 00",
  "url": "https://beyond-physio.ch"
}
</script>

Swap in your own details and paste it before the closing head tag. One caution worth stating: this only helps if the same information also appears as ordinary readable text somewhere on the page. Labels describe content. They do not replace it.

Limen does the first job

This is where I stop being neutral, so let me be obvious about it rather than coy.

Limen scans your website and reports what is wrong with it in language you do not need a developer to interpret. Nineteen checks, organised into four territories.

The four territories, and the question each one answers. Content carries the most weight by a wide margin, which is inconvenient, because it is also the one that cannot be fixed with a snippet.

The list is deliberately short. A website offers thousands of measurable data points and collecting all of them is precisely the mistake I have been describing. A check earns its place on our list only if knowing the result changes what you should do. Anything true but inert stays out, however easy it would be to collect.

The free scan covers the checks a machine can verify without judgement: whether crawlers are allowed in, whether your pages need code to run, whether your services are readable text, whether your location appears, whether the basic machine labels exist. That is hygiene, and hygiene is where nearly every small site is losing.

The subscription adds the checks that need judgement rather than detection: whether your content covers the ground a customer expects, whether your pages lead with their answers, whether your questions match how people actually ask, and the rewrite that shows you your own text beside what it should say. Those are the ones that move the outcome, and they need judgement no automated crawler can supply on its own. That is the honest reason for the line between the tiers. It is not more findings. It is the findings that matter.

Every failing check comes with a fix, at every tier. Some you paste in. Some you write yourself, and where you have to write, we tell you what is missing rather than leaving you to guess.

Then wait. Genuinely wait.

Here is the part that gets left out of every article on this subject, because it is unwelcome.

You will fix your site and nothing will happen. Not the next day, not the next week. The pages have to be crawled again. The new content has to work its way into how these systems represent your business. And the answers themselves wobble, which means the first few times you check, you will not be able to tell improvement from noise.

The honest shape of the wait. I cannot give you a firmer number than this, and I distrust anyone who does, because nobody outside these companies knows the reindexing cadence, and it changes.

What this means practically: patience is part of the method, not a personality trait you either have or lack. The work is doing the fix and then leaving it alone long enough to mean something. Checking daily will teach you nothing except how to feel anxious on a schedule.

And only then, measure

Once the fixes are in and a few months have passed, tracking starts to earn its place. And your first tools are ones you almost certainly already have, at no cost.

Google Search Console

Shows you which queries surface your pages and what people click. It is not AI-specific, but the questions people bring to assistants and the questions they bring to search overlap heavily, and Search Console shows you the gaps in what you have written. Sort your queries by impressions with a low click rate: that list is a to-do list of things your pages are being offered for and failing to answer well.

Google Analytics

Will show you referral traffic from AI assistants. The picture is partial. A great deal of AI influence happens without a click, when someone reads a recommendation and later types your name directly. You will see a rise in direct traffic and not know why. Accept the imprecision rather than buying a tool to eliminate it, because the tool cannot eliminate zero-clickZero-Click Search: A search that ends with the answer shown right on the results page, so the person gets what they need without clicking through to any website. Read the full entry in the glossary.Zero-Click SearchA search that ends with the answer shown right on the results page, so the person gets what they need without clicking through to any website.Read full entry behaviour either.

Between them, those two will tell a small business what it needs to know for a long time. When you outgrow them, you will know, because you will have a specific question they cannot answer. Buy the answer to that question. Do not buy a dashboard in the hope that it will suggest one.

The advantage worth naming

I will not promise you an outcome. I can promise death and taxes, and beyond that I am as much at the mercy of these systems as you are.

What I will point out is timing. Almost nobody is doing this yet. The small businesses in your category are mostly still where you were before you read this: a good business with a site that machines cannot properly read, unaware that anything has changed. Fixing yours now is not a competitive edge you have to defend against fifty others. It is an empty room.

The measuring can wait. The door is the thing.