Skip to main content

Build log · collimer

How does AI position your brand? Three checks a visibility score cannot see

·

Illustration for the build log "How does AI position your brand? Three checks a visibility score cannot see"

AI engines position your brand in three ways a visibility score cannot show: whether they describe you as yourself, whether they bring you up before a buyer knows your name, and whether the pages they cite mention you at all. A brand can score well and still lose on all three. Collimer’s “How AI positions you” section checks each one and gives a one-line verdict, up to four insight cards, and a link from every card to the matching fix in the report’s 30/60/90 roadmap.

Why is a visibility score not enough?

A visibility score counts how often engines name you. That is the right first question, but it treats every mention as equal. A mention that describes a different company with your name counts the same as a recommendation. A brand that only shows up when the buyer already typed its name counts the same as one that shows up when the buyer only has a problem. The positioning checks split those apart, because each one has a different cause and a different fix.

What are the three checks?

CheckThe question it answersWhat a problem usually meansWhere to start
IdentityAre you described as yourself, or as someone else?Engines mix you up with a same-name business or describe you as a feature of a bigger companyA short, distinctive description of who you are, what you do and where, repeated on your site and the profiles engines read
Cold-query presenceDo you show up when the buyer has not named you yet?You are known to people who already know you, and a competitor is named to everyone elseContent that answers the problem-led and category questions your buyers ask before they have a shortlist
Cited sourcesDo the pages engines cite name you?The engines trust a set of sources for your category, and you are not on themGet listed or mentioned on the pages that already get cited, instead of only publishing on your own site

The first two are written up in more detail: mentioned, and described as someone else and what a pooled score hides.

When does each check flag a problem?

Each check has a threshold before it says anything. The identity check flags when more than 15% of at least 5 classified mentions are confused with another entity or cannot be resolved. The cold-query check flags when your best warm category, where buyers already have your name, is at 40% or higher and the two cold categories, where they only have a problem, average 15% or lower. In one case we measured, a brand held about 70% on warm queries and 0% on problem-led ones, and a named competitor was returned instead.

Why does it matter which one you have?

Because the fixes pull in different directions. An identity problem is fixed by being more specific about who you are. A cold-query problem is fixed by answering questions that never mention you. A cited-source problem is mostly fixed off your own site. A team that only sees “visibility is low” tends to publish more of whatever it already publishes, which helps with none of these if it is the wrong one.

Where does it fall short?

The checks need enough evidence to say anything, and small or new brands often do not have it. A check with fewer than 5 classified mentions behind it stays silent rather than guessing, so some reports show no positioning section at all. That avoids a false alarm, but a reader cannot tell “nothing wrong” from “not enough data yet” without asking.

The thresholds are also judgment calls. The 15%, 40% and 15% lines rest on a small number of measured cases, and we have not checked how often they flag brands that are actually fine. A flag is a reason to look, not a verdict.

What we’re still figuring out

  • How many reports show no positioning section at all.
  • Whether a one-line “not enough evidence yet” note would help readers without reading as a finding.
  • Whether 5 mentions is the right floor for all three checks.

For agents: try this yourself

  • Split your own mentions. Ask five AI engines “what is [brand]?” and a problem-led question in your category with your name left out. Record whether each answer describes you correctly, and whether you appear at all on the problem-led side.
  • List the cited pages. For three questions your buyers ask, note every page the engines cite, then check which of those pages mention you.

How this was made

Drafted by the Chronicler from Collimer commits and a published guide, then edited and published by Chase.

See how the Chronicler works →

Try this with your own agent

2 prompts you can hand to your own agent (or run by hand) to work with what this post documents. Edit the bracketed parts for your context.

Find the empty states that still say something

Here is the code or a screenshot of our report or dashboard: [paste]. List every card, chart or sentence that still renders when its data is empty or below a sample threshold. For each, say what a reader would conclude from it and whether that conclusion is supported.

Check that the query loads the field

Here is a feature that reads [field] from [records]: [paste code]. Here is the query that loads those records for this feature: [paste]. Confirm the field is selected. If it is excluded or lazily loaded, say what value the feature will actually see.

More in Collimer Build