What a visibility score is actually made of

A number you cannot take apart is not a measurement, it is a mood. This one takes apart into seven pillars — ninety points of observation and ten of forecast.

The number is a sum, not a verdict

Every visibility tool ends at a score, and most of them stop there. But a number you cannot take apart is not a measurement — it is a mood with a decimal point. The honest test is simple: can the person holding it say what would make it move, and by how much?

So here is the whole of ours, pillar by pillar. What each one counts, what it is worth, and — the part that usually goes unsaid — what it refuses to count.

The score runs from nothing to a hundred, and it is the sum of seven pillars.

| Pillar | Points | What it counts | |---|---|---| | Answer presence | 30 | How often you are named, weighted by engine and by how often the question is asked | | Citation footprint | 20 | How many authoritative sources the engines lean on about you, and whether your own site is one | | Factual accuracy | 15 | Wrong hours, services and service areas that the engines repeat | | Competitive share | 11 | How often you win a question that somebody local wins | | AI readiness | 10 | Whether a machine can read your site and your profile at all | | Answerable content | 8 | Whether your site contains the answer to the question | | Sentiment | 6 | How favourably you are described when you are named |

Seven pillars, and no eighth held back for a sales call.

Ninety points of observation, ten of forecast

Six of those pillars describe things that were observed. An engine was asked a question, an answer came back, and the answer either named you or it did not, cited a source or it did not, said something wrong or it did not. Ninety of the hundred points are the tally of what actually happened.

Readiness is different, and the difference matters more than its ten points suggest. Nothing about your markup or your profile is a record of what an engine said. It is a forecast: how findable have you made yourself for the next probe, and the one after that. It is the only predictor in the score, and it is capped at a tenth of the total on purpose.

That line is under constant pressure. A complete Google Business Profile really does correlate with being named — the engines lean on it heavily — and it is tempting to let it earn presence points, because it feels like evidence. It is not. The moment a checklist can move a measured pillar, the score stops measuring anything: a business could improve its number without a single engine changing a word about them. So the checklist stays in its own box and nowhere else.

Answer presence — 30 points

The largest pillar, and the closest thing to the question you actually care about: when somebody asks an assistant for a business like yours, are you in the paragraph?

Two weightings shape it. The first is by engine — a mention where more people ask counts for more than a mention where fewer do. The second is by how often the question itself is asked. Every prompt is banded high, medium or low, and a mention in a high-band question is worth roughly three times a mention in a low one. Being named in questions nobody asks is not visibility in any sense that pays a bill.

And presence is a rate, not a flag. Every question is asked of every engine more than once, and the cell records how many of those runs named you rather than whether any of them did. Named in three runs out of three and named in one run out of three are different facts about your business, and a tool that flattens them into a tick is throwing away the most useful thing it learned. The reason for that is worth its own page: answers change between runs, and the change is not noise to be smoothed away.

Citation footprint — 20 points

When an engine answers a question about local businesses, it leans on sources. The footprint pillar asks two separate things about them.

Coverage is the breadth: how many distinct authoritative domains turn up carrying information about you. It saturates — past a handful of solid sources, another directory listing stops changing which businesses get named, and a score that kept paying for them would be paying for busywork.

Ownership is the harder half. Of all the times a source was pulled into an answer about you, how often was the source your own site? This is the local version of the hotel problem. If every answer about you is assembled from aggregators, the aggregators own your description and you are a line item inside somebody else’s page. Ownership is measured on pulls rather than on domains, because being cited once against forty aggregator citations is not shared ground, and counting domains would score it as though it were.

Which sources those are, and which ones keep coming back, is the subject of the sources that keep coming back.

Factual accuracy — 15 points

Points come off for wrong facts an engine repeats about you, priced by how badly each one misleads a buyer. A wrong closing time that sends somebody to a locked door costs the most. A slightly stale service list costs the least. Nothing here is a penalty for the engine being imperfect; it is a measure of how expensive its imperfection is to you.

Two decisions inside this pillar are worth stating plainly, because both cost the score points it could have claimed.

The first is that the detection is biased toward silence. A rule only fires when a claim asserts a value we hold and that value confidently disagrees. Anything vague, anything partial, anything we cannot ground against a real record is skipped. Telling you that an engine got your phone number wrong when it did not is far worse than missing one that it did — a miss costs you a fix you did not know about, a false accusation costs you your belief in every other number on the page.

The second is that not checking scores zero, not fifteen. An empty list of wrong facts is ambiguous: it can mean we looked and found none, or it can mean we never looked. Those are worth fifteen points apart, and defaulting to the generous reading would hand every audit a free fifteen per cent for work the product did not do.

Competitive share — 11 points

Of the questions where somebody was named first, how often was it you? This pillar only counts contested ground — questions where a business won and it was either you or a rival. Questions where the engine named nobody are not a loss here, because there was nothing to lose.

It used to be worth more. It was cut to fund the content pillar, on a simple test: what can the owner actually do about it? Share is largely a restatement of presence seen from the rival’s side, and the advice it generates is the same advice presence already gave you.

Answerable content — 8 points

Three checks, and the only ones in the whole score you can close by writing something this afternoon.

Service pages come first: a site that describes everything it does on one page can be cited as a general trade, but it cannot be the source for a question about one specific job. This scores in steps rather than as a yes or no, because going from one page to three is real progress and a tick box would tell the owner who did that work that they earned nothing.

Prices are second, and they are worth as much as the service pages together. The most-asked research question in almost every category is what the job costs. A page that answers it is a page an engine can quote. A site with no number anywhere on it cannot be the source for an answer about money.

Service area is third. Fewer questions turn on it, but the ones that do are unwinnable without it, because an engine will not name a business for a town the business never claims.

Content and readiness fail independently, and that pairing is the most useful thing the score tells most owners. A perfectly marked-up single-page site scores full readiness and almost no content — and it is the content number that explains why it loses every question about price.

Sentiment — 6 points

How favourably you are described in the runs that named you, normalised onto the pillar. It is the smallest pillar in the product and it is small for a reason worth admitting: it is the only one with no fix behind it. Nobody can act on the instruction to be described more warmly. It stays in the score because it is real information about how you are being represented, and it stays small because it cannot be turned into work.

AI readiness — 10 points

Split evenly between the two surfaces you actually control.

Five points sit on the website: structured data carries the most, then an llms.txt file, then whether the AI crawlers are allowed in at all, then a sitemap. Five more sit on the Google Business Profile — a name, address and phone number that match your site, a working website link, opening hours, categories, and enough reviews for a rating to mean something.

The profile half is flat, a point each, and that is deliberate. Each one is a single independently closable gap, so one point maps to exactly one thing to do and you can watch that one thing earn that one point.

The review threshold moves by country. Review volume outside the United States is far thinner, and a well-reviewed British plumber has tens rather than hundreds. A single global bar would mark down most of a market for a gap its own rivals do not have either.

When you are never named, two pillars go to zero

If no engine named you in any run, accuracy and sentiment both score nothing, with a note saying why rather than a number pretending otherwise.

This looks harsh and it is deliberate. Awarding a cheerful no wrong facts found to a business nobody mentions would hand full marks to invisibility — which is the precise opposite of what the score exists to tell you.

Why there is a version stamp

Every audit records the version of the scoring engine that produced it, and changing what is measured means bumping that version.

It has moved three times. Once when wrong-fact detection started running at all, once when the content pillar arrived and share and sentiment paid for it, and once when every question in the library was rewritten from search-box keywords into the sentences people actually type at an assistant. In each case a business that changed nothing would score differently, and comparing across the change would show progress nobody earned.

So scores from different versions are not compared, not charted together, and never mailed to you as movement. A score you did not earn is worse than no score, because the first time you catch a tool crediting you with an improvement you know you did not make, you are right to stop believing the rest of it.

What the score is not

It is not a ranking, and there is no league table behind it. It is not a prediction of revenue. It is not a guarantee of placement, and no honest tool can sell you one, because none of us controls the engines.

It is a measure of one thing: how a set of buying questions, asked the way a stranger would ask them, came back — and how much of the outcome is inside your reach. If you want to see the questions themselves, they are written down in the method behind every audit.