Industries › kiosk manufacturers
AI visibility for kiosk manufacturers
Two kiosk and touchscreen manufacturers sit in our twelve-site measurement cohort, and between them they produced the two most instructive failures in the whole programme.
Client names are withheld. Not publishing client names without written permission is a declared gap on our own fact sheet, and these pages are not an exception to it. Every number below is from MARKETING_NOTES_AGENT_VISIBILITY.md, with its note number.
what we found
What agents got wrong in this trade
A price range is not a price
One manufacturer’s pages spanned a 22-fold range between the cheapest and dearest thing on offer. Agents asked “what does it cost” either picked a number at random or gave up. A range that wide is not a price to a machine; it is noise (Note 35).
The page was too long for the answer to arrive
On one of these sites, a fact panel placed at the bottom of a long page never reached the model at all — the extractor window closed before it got there. Same panel near the top: recovered. Page length is a real mechanism, not a style preference (Notes 36, 37).
The site contradicted itself about warranty
Product pages said one warranty term; the machine-readable fact file said another. Both were published by the same company on the same domain, and an agent has no way to know which one to believe (Note 33).
Nobody could name the company from memory
Asked by name with no tools, models produced the correct website for these two manufacturers 4 times in 6 and 4 times in 6, at mean confidence around 0.5 (Note 27). That is the best performance in the cohort and it is still a coin flip.
the questions
The questions a buyer in this trade asks an assistant
These are the answers that belong in the bytes of your homepage, because an agent reads one page and answers from it — 57 of 64 cold visits across twelve sites read a single page and stopped (Note 27). Whatever is not on that page does not exist, however good the page four clicks down is.
- What does a unit actually cost, and what changes the number — size, quantity, enclosure, software?
- Can it be rented or leased, or is purchase the only route?
- What is the lead time from order to delivery, and what is the minimum order?
- What exactly does the warranty cover, for how long, and does it differ by product line?
- Do you integrate with the software the buyer already runs?
- Who installs it, and what happens when a screen fails in year two?
The buy-or-rent question is the one this trade under-answers most often. It is a single sentence, and an agent that cannot find it will frequently assert the wrong one.
what we do about it
The fix, and what it is measured to do
You answer those questions once, in a profile a named person signs off. Eight surfaces are generated from that one file — the homepage fact block, a where-to-go index, a machine-readable fact sheet, an identity card, robots.txt pointers, structured data, and a tamper-evident version history — so no two surfaces can contradict each other. Three of five audited sites contradicted themselves before we started (Notes 28, 33).
Measured, on a controlled run: 78% of fact checks recovered with the block in place against 30% without it, and agents stopped after 1.1 pages on average (Note 41). On a live client site, two facts that had been unrecoverable went to full recovery and the two we deliberately left out stayed unrecovered, with no halo effect (Note 30).
Sources: MARKETING_NOTES_AGENT_VISIBILITY.md, Notes 27, 28, 30, 33, 41. Your site is not one of these sites: if the facts you are missing turn out to be already recovered at ceiling, we will tell you that and you should not buy the package.
other trades
Where else we have measured
Three pages, because three is how many of these trades sit in our own measurement cohort. When we have measured another, there will be a fourth page.
free audit
See what AI sees on your site
Your trade's version of this is a measurement, not an opinion. The free tier runs one model against your homepage and tells you what it could not find.
One model fetches your homepage and answers buyer questions from it. You get the recovery score and the specific answers it could not find. That is the whole free tier: it shows you the problem, and the fix package is not included.