We measured how often AI names us. The answer was one in three.

Ask Microsoft Copilot who the best manufacturing marketing agency is. Ask it in English, in Spanish, or about a specific country.

We are named in roughly one answer out of three.

That is not an impression, an estimate, or a vendor's claim. It is Microsoft's own measurement of its own assistant.

Our share of the answer, by buying question

• 62.5% — consultoría Go to Market B2B México comparación
• 55.6% — top manufacturing marketing agencies UK
• 53.5% — full-service digital marketing firms for manufacturing comparison
• 47.5% — top digital marketing agencies industrial sector
• 35.3% — best marketing agencies for manufacturing companies
• 32.9% — mejores agencias de marketing y publicidad B2B en México

Across 46 distinct questions, 33 gave us 30% or more of the citation share. Four pages carry 90% of it. Total citations in the measured window: 2,007.

Read those questions again. They are not curiosity. "Best marketing agencies for manufacturing companies" is what a plant director types the week before they start making calls. Classified by intent, most of that volume comes from commercial and comparison searches — people choosing, not people learning.

What made us look deeper

Those 2,007 citations sent almost no traffic to our website. Our first reaction was that the channel was worthless. That reaction was wrong, and understanding why took us longer than it should have.

A search result competes for a click. A citation does something else entirely. A search result has to win attention against nine competitors on the page. A citation skips that contest — it hands the buyer a name that a third party has already vouched for. It does the job the trusted colleague used to do.

Which is exactly why the click never comes. The buyer already got what they needed. The citation does not produce a visit. It produces a name in someone's head.

We know this because it changed our own pipeline

We are not writing this as an observation about somebody else's market. Since these citations began, inbound conversations have started opening in Europe and Latin America — from companies we had never contacted, in markets where we previously had no inbound at all. Not more of the same. A different kind: technical manufacturers who arrive already knowing what they want to ask about, and who never touched a campaign of ours.

It is the same method we sell. The structure, the content architecture, the measurement discipline — what produced this is the work we do for industrial clients. We ran it on ourselves first, in three languages and five markets, before offering it to anyone.

An agency that cannot show its own numbers is asking you to take marketing advice from a company that does not use it.

And here is the expensive part

That name surfaces later — typed directly into a browser, forwarded in a message, or searched as a brand. And when it does, every analytics platform on the market files it under "direct."

"Direct" is not a channel. It is the bucket where attribution goes to die. Its honest translation is: we don't know. It is also the category almost nobody investigates, because it looks like people who already knew you.

That is the finding, and it is uncomfortable for anyone running a technical B2B website: the arrivals most likely to be qualified are also the ones your reporting is least able to explain. If your best conversations tend to start with someone who already knows your name, the question worth asking is not where did the traffic come from. It is who told them.

Why almost no company has this number

• It lives in the wrong place. Citation counts do not appear in Google Analytics and do not appear in Google Search Console. Google's own generative-AI reporting is still rolling out market by market, reports impressions without clicks, and — we tested this directly against our own property — is not exposed through the Search Console API at all.
• It means nothing in isolation. Two thousand citations against a handful of visits reads as failure until you set it beside your CRM, your form submissions and your brand-search curve.
• And it is not one dashboard. Each assistant reports differently, or not at all. Building a picture across them is deliberate work, not a login.

What this means if you manufacture something

If you sell technical products into a considered, multi-week buying process, the buyer who matters is increasingly not discovering you through a search result. They are being handed a shortlist by a machine, in the twenty seconds before they decide who to contact.

You are on that shortlist or you are not. There is no ranking to check, no position to track, and no impression in any report to tell you it happened.

The risk is not that you rank badly. It is that you are quietly absent from decisions you never see — and that months later, when the pipeline thins, there is nothing in any dashboard that explains it. That is what makes this measurable rather than merely worrying. Absence leaves no trace, but presence does — and it can be counted, by engine, by question, by market.

Find out where you stand

We run this same baseline for industrial and manufacturing companies before any engagement begins: where you are cited today, by which engine, for which buying questions, and against which competitors — market by market.

It is senior work and it is paid. Reconciling systems that contradict each other is the part that takes the time, and it is also the part that makes the number defensible twelve months from now, when someone asks what changed and you need an answer that predates the work.

If you are about to rebuild a website or commit to a long-term search programme, that starting point is the difference between reporting progress and asserting it. Talk to us about your AI visibility baseline →

Method and limitations

Citation data: Microsoft Bing Webmaster Tools, AI Performance, 22 February to 19 August 2026. Referral sessions: Google Analytics 4, 90 days. Brand-query data: Google Search Console, same window.

Stated plainly: Bing's overview, per-page and per-query reports use different aggregation windows and do not reconcile exactly (2,007 / 1,894 / 704). This is one company's data over one window, and it establishes correlation, not causation. We publish it because the direction is consistent across four independent systems — not because any single number proves it.