Reading the Gaps

Measuring tells you how often you show up. This step is about why you don't, when you don't, because each kind of absence has its own fix. A low score on its own is not something you can act on. A named gap is.

The number everything hangs off

Everything here is a way of explaining a recommendation rate that is lower than it should be: how often you get recommended across a lot of runs, per engine, per channel. The findings below take that number apart.

Are you unknown, or known and passed over?

There are two reasons you might be missing from an answer, and from the outside they look the same. You are simply not there. But they have opposite fixes.

A knowledge gap means the model does not know you. You are missing because you are not showing up enough in what it learned from, or in what it can pull up live. The fix is getting into those sources: the content, citations, and listings models read. A knowledge gap does not close on its own as models get better. It closes when you change your presence in those sources, so you have levers here, not a waiting game.

A selection gap means the model knows you but does not pick you. Ask about you by name and it can describe you fine, but it never brings you up for the category. That is a ranking problem, and the fix is whatever is making it choose competitors instead: stronger citations, deeper content, better associations. This is the one people mean when they say a brand is getting ghosted.

There is an easy test for which one you have. Ask the model about you by name, the brand level from Scope. If it can describe you but won't recommend you, you have a selection problem.

Which channel is the gap in?

Recall and live search often disagree, and which way they disagree tells you where to put your effort.

Strong with web off, weak with it on. The model knows you from training, but live search is not surfacing you. Your standing is not being backed up by current content and citations. Put the effort into the live layer: the pages and sources retrieval pulls from.

Weak with web off, strong with it on. Live search finds you, but the model does not know you on its own. Your visibility is real but borrowed, leaning on whatever the web looks like right now, and it can slip when that changes. Put the effort into building presence that sticks, so you are known without a search.

The reason this split matters is that the two fixes run on opposite clocks. Earning your way into a model's training takes a retrain on a web that already talks about you, which is a 12 to 18 month effort of building consistent presence. Getting cited in live search can happen within weeks of publishing the right page. So the channel gap tells you two things at once: where to put the work, and how long that work takes before it pays off.

The case that catches people out is strong with web off and weak with it on. The model already knows you and already wants to recommend you. It just cannot find a page of yours to back that up when it searches. Read the weak live-search number on its own and you would assume you have to earn the model's trust from scratch and budget years for it. The web-off test is what tells you the trust is already there, and the real job is handing the model something to cite, which is a publishing push rather than a multi-year brand campaign. That one read stops you from spending a brand-building budget on a sourcing problem.

Either way, the gap between the two channels is the instruction.

How deep does it go, and is the hole yours to take?

Your recommendation rate for the broad question hides what happens service by service. Two things to look at.

Depth and breadth. Depth is how strongly the model ties you to a given topic. Breadth is how many topics it ties you to at all. A business can be broad and shallow, known a little for everything, or narrow and deep, owning one thing. The fix is different for each, so a single tier label can point you the wrong way.

Missing or displaced. When you are absent on a service, either nobody owns it, in which case you can take it by publishing, or a competitor owns it, which is harder and slower. We always check service lines on their own, because a strong broad score routinely hides a hole on one specific service that someone else has taken.

The thing you do that a competitor gets credit for

Most businesses have at least one real differentiator: open Saturdays, sedation for nervous patients, a service the others in town don't offer. The check is simple. For each true differentiator, ask the unbranded question, "who around here offers this," many times, and count who gets the credit.

The result is often the most lopsided number in the whole diagnosis. We watched a business that is genuinely open Saturdays get named in one out of twenty "open Saturday" answers, while the engines assembled neat tables of competitors' weekend hours pulled straight from their websites and listings. The model wasn't wrong about anyone it named. It just never found anything tying this business to the question.

That last part is why this gap is so attractive to find: it is usually cheap to close. The differentiator already exists, so there is no service to build and no reputation to earn. There is just no crawlable page, listing, or profile field that states it plainly where the engines look. A misattributed differentiator is a found lever, and it is often the first thing worth fixing.

One nuance worth knowing: ask the model about the business by name and it will often mention the differentiator just fine. The knowledge is there. It only goes missing on the unbranded question, where retrieval has to connect the differentiator to you on its own. That is the same retrieval-versus-knowledge split as the channel gap, showing up at the level of a single fact.

Is what it says about you accurate, and what does it say you're known for?

Two more reads come from the same branded questions.

Accuracy. Take what the engines say about you and check every verifiable claim against the facts: hours, locations, staff, services. With web search on, the accuracy rate for a local business tends to be high, often nearly perfect. The errors that do show up are systematic rather than random: the model stretches one service you offer into a neighboring one you don't, or occasionally invents a staff member. Worth checking and worth a correction when it's wrong, but it is rarely where the visibility problem lives.

Attributes. Separately from whether the facts are right, there is the question of what descriptors attach to you: family-friendly, modern, gentle, affordable. Tracked over time, this is your reputation in the model's own words, and it is the early-warning instrument. A descriptor that vanishes, or a new one that appears, is a change in how the AI describes you to buyers, and you want to notice it before the buyers do. The absences are just as informative: if "affordable" never comes up and your prices are nowhere on your site, those two facts are usually the same fact.

Recommended, but is it your site getting cited?

You can get recommended all the time and still have your own site barely cited. When that happens, your standing is resting on sources you do not control: directories, aggregators, somebody else's article. Two findings come out of this.

The owned-citation gap: you get recommended, but your own pages are not the source. The fix is making your site the one worth quoting, with real answers, real prices, and the specifics a model wants to lift, on clean pages.

The citation map: who is actually getting cited for your category. Those directories, publishers, and competitor pages are your shortlist, both for where to get listed and for what kind of page to become.

What this feeds

Every named gap points to a lever. Pulling the levers maps them to the specific tactics that close them.