Skip to main content

Build log · collimer

Can an AI visibility scan read a local business? Only if it asks about your trade, your town and your real name

·

Illustration for the build log "Can an AI visibility scan read a local business? Only if it asks about your trade, your town and your real name"

An AI visibility scan can read a local business when it gets three things right. It asks the questions your buyers ask, about your trade and your town. It counts only the mentions that are really you. When it can’t tell what kind of business you are, it falls back to local questions instead of software ones. Collimer’s local scans now do all three. Getting there taught us three lessons that apply to anyone measuring how AI sees a small business.

What does a local scan ask and count now?

What the scan doesNowBefore
The questions it asksPhrased for service businesses in general, with your actual city filled inWritten around one gym, so a roofer was asked about gyms
The names it creditsYour full name, or a short form of it only when your web address backs itAny answer containing the first word of your name
When it can’t classify youLocal and professional-services questions, if your site shows local signalsSoftware-company questions

For a business owner, the result is a score that measures the right thing: how often AI engines name you when someone asks for what you sell, where you sell it. New tests run a roofing business through the local questions and fail if any gym wording or blank placeholder comes out.

Why test a scan on more than one customer?

Because a scan tuned on one business quietly becomes a scan for that business. Our local questions were written while we tuned the scan with our first design partner, a small gym, and they were never generalized. Every plumber, dentist and roofer we scanned got gym-shaped questions like these:

  • “How does {brand} compare to other gyms?”
  • “Gyms like {competitor}” plus a neighborhood in the partner’s city

A roofer almost never gets named in an answer about gyms, so its score came out low for a reason that had nothing to do with the roofer. The lesson is to test every category’s questions on a business it wasn’t written for, which is exactly what the new roofing tests do.

Why does a generic business name need a check?

Because short forms of a generic name match things that aren’t you. To find mentions, the scan builds shorter versions of a business name to look for. It used to fall back on the first word alone. For a distinctive software name that’s harmless. For a name like “Summit Roofing of Dallas”, an example from our own test code and not a customer, every answer about a mountain, a conference or another company called Summit counted as a mention.

A single generic word now counts only when the business’s web address is built on it. If your name is common, it’s worth knowing that any tool measuring your visibility faces the same trap.

Why can one score be wrong in two directions at once?

The two problems pushed in opposite directions. Gym questions pulled a roofer’s score down, and first-word matching pushed a generic name’s score up. Both could happen in the same scan and partly cancel, so the final number could look plausible while being wrong twice. When a score seems reasonable, it’s still worth checking the parts. It’s the same lesson as what a pooled score hides. A related matching problem, a missing short form rather than a false one, is in the apostrophe that scored a brand 5 out of 100.

What did review catch on the way?

Review caught two fixes that would have repeated the original mistake. The first version of the new questions gave service wording to a list of known business types and software wording to everything else, so any new type would have defaulted to software again. We flipped it: only software companies get software wording now. Review also caught a check that ran in the wrong order and would have skipped the local fallback in the exact case it was built for, a business the scan can’t classify.

What we’re still figuring out

  • How much local scores from before these fixes would change. We haven’t re-scanned them.
  • Whether any other kind of business was tuned on a single example. We fixed local and haven’t audited the rest.
  • How generic a name has to be before the domain check is the only thing protecting it.

For agents: try this yourself

  • Audit the seeds. Given the prompt set a visibility tool uses for a business, flag any prompt that names an industry, city or product the business does not have.
  • Test the alias. List the aliases a matcher would accept for a brand name, and for each one give an unrelated sentence it would falsely match.

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.

Audit a visibility tool’s prompts for the wrong industry

Here are the prompts a visibility tool asked AI engines about [business], a [type of business] in [city]: [paste]. Flag every prompt that names an industry, city, product or competitor type the business does not have, and rewrite each flagged prompt for this business.

Find false aliases

For the brand name "[brand name]", list every alias a matcher might accept (first word, initials, name without suffix). For each alias, write one sentence an AI answer could contain that matches the alias but is not about this brand. Mark which aliases are safe.

More in Collimer Build