FieldNoteHi there,
Let's start with three numbers worth knowing.
|
|
|
So what does this mean for you? Your customers are getting answers instead of links. When a customer asks AI to compare products, shortlist suppliers or check a price, your business may be assessed before a single person visits your website. That applies whether you sell industrial equipment, software or consumer goods. An AI assistant gathers information, and an agent can act on it too, preparing an enquiry or placing an authorised order.
That creates a new question: can AI find your offer, understand it, and do something useful with it?
So we built Hyperize, a MING Labs venture, to test exactly that: how AI finds a business online, reads what it finds, and acts on it.
To show you how it works, we ran it on five appliance brands in Germany.
We asked ChatGPT and Perplexity 77 questions a German household asks before buying an appliance. Same questions, twice, on the same day, for Bosch, Miele, Samsung, LG and Electrolux. Then we knocked on each brand's website the way an AI system does.
This is a sample, not a census: five brands, one country, one day.
Here's what surprised us: the brands AI names most are the ones whose websites it can barely read. The brands whose websites are wide open are the ones it almost never names. Not one of the five manages both. That split is where the money leaks, and it is unlikely to be unique to appliances.
Most brands treat these as one thing. They are won in different ways.
Named. AI says your name when the customer never asked for you.
Cited. AI takes the facts from your website, and says so.
Here is how the five brands did on both.
|
Takeaway 01
AI can mention your brand without citing your website.
When the question did not name a brand, this is how often each brand came up, and how often the facts came from its own site.
Bosch is the most named brand in the test. Its own site supplies the facts in fewer than a quarter of those answers. Test portals and comparison sites supply the rest. Bosch also turns away the named AI readers at the front door, so the assistants cannot fetch Bosch's own version even when they want to. What this means for you
Being cited is the only way you control whether what AI says about you is correct. A third-party description can be outdated or simply wrong, and those errors reach your customers as fact. They can also settle into future models. Make your own website the source AI can find, verify and cite. |
|
Takeaway 02
Being found and being usable are separate jobs.
The split does not stop at who gets credited for the facts. Being found. Appearing when AI researches a customer's needs. Being usable. Letting that customer's agent act: request a quote, upgrade a service, reorder a spare part. In our test the split was clean. Bosch and Miele are named far more often, and their websites turn away several kinds of automated access. Samsung and LG let every AI reader in, and are the least named of the five. What this means for you
Behind every agent is a person trying to get something done. Blocking the agent can mean losing their business. Give it a reliable way to act, through your website or through a connection to the customer's assistant. Tools such as WebMCP support this, and payment integrations such as Stripe's can let an agent buy with the customer's permission. Make it possible to do business with you through an agent. |
|
Takeaway 03
AI asks follow-up questions of its own.
Opening the door is only the first ask. One customer question turns into several searches, and more searches as new details emerge. This branching is called query fan-out, and it varies by model and by conversation. In our appliance runs, a broad question about washing-machine test winners led to searches for exact Bosch and Siemens model numbers. Another run searched for a model's water-protection rating. The customer had asked for none of that. What this means for you
Give AI facts it can use along the way: exact model numbers, specifications, compatibility, service conditions, evidence. Keep them current and structured so machines can read and connect them. The detail a customer would never think to ask for is often the one AI needs to check a claim. Your product data has to answer the questions AI asks, too. |
| When somebody else gets cited talking about you, they are speaking in your name. |
People need a website that helps them understand an offer, trust a brand and choose with confidence. Their agents need that same offer in a form they can fetch, verify and act on.
For us this widens the design brief. Build something compelling for people, and a reliable way for their agents to get things done. Right now almost nobody has both, which is the whole opportunity.
|
The full study
All five brands, the 77 questions, every answer and every source. Read the home-appliance Snapshot →
Method: 77 frozen questions, ChatGPT and Perplexity via API, 2 September 2026, two runs each. Percentages cover only the no-brand-named answers. Access checked with test clients, not verified crawlers, one page per brand. One market, one day. Brand names used for identification only, not a product-quality rating.
|
Give us your website and a work email. Within 48 hours you get the same treatment: where AI finds you, what agents can actually use, and what to fix first.
|
Free Hyperize Snapshot
See how often AI names you, and whose answer it uses.
Then send the report back to us and we will go through what it means for your business. That conversation is where most people find the surprise.
Get My Free Snapshot →
Free. No commitment.
|