A filtration equipment manufacturer asked me to look at why they never appeared in AI answers. Forty years in business, strong reputation with the three big engineering contractors in their region, a website that had been redone eighteen months earlier and looked fine.

I ran the question their buyers would ask. The answer named four companies. Two were competitors. One was a company that had gone out of business. The fourth was their own distributor in Germany, described as a manufacturer.

That last one is the moment I usually get the room's attention, and it is where these projects tend to start. Below are the five things I now do differently with industrial clients compared with everything else.

1. I ask the question in application language, not product language

This is the first thing and it changes the entire diagnosis.

Manufacturers describe themselves in product language: "we make cartridge filtration systems." Buyers with a problem search in application language: "how do I remove particulates from a hot caustic stream in a food-grade line." Those are different vocabularies, and the second one is what gets typed.

When I test a client's visibility, I write the question set the way a maintenance engineer with a failing process writes it — the fluid, the temperature, the constraint, the certification, the region. Not the category name. Almost every manufacturer I have worked with does well on the category query and poorly on the application query, and it is the application query that has commercial value, because it is the one asked by someone who has not yet decided what to buy.

What I do about it: pull the last fifty inbound enquiries and read how they described their problem, in their words. That is the question set. It is better than anything I could invent, and it is sitting in the CRM already.

2. I check who owns the entity before I touch any content

Back to the German distributor.

This happens constantly and it is not the distributor's fault. Distributor catalog pages are genuinely better source material — full specification tables in HTML, stock levels, lead times, application notes, comparison pages. The manufacturer publishes a brochure page and a PDF. When a system needs a source, it takes the better one and names whoever owns it.

What I do about it: three things, in order.

I have seen this alone change which company gets named, with no other content work at all.

3. I get the spec sheets out of PDFs first, before anything creative

Industrial companies have enormous amounts of exactly the material these systems want. Test data, certifications, material specifications, tolerances, project references. Nearly all of it is in PDF, and a meaningful share is scanned.

PDFs get picked up unreliably. Scanned ones usually not at all. So a company with forty years of documented capability presents almost nothing readable.

What I do about it: the HTML mirror. Same content, published as a web page alongside the PDF — specification tables as actual tables, certification numbers as text, materials as text. Nothing is rewritten and nothing is invented; it is a container change.

For the filtration client this was two weeks of work by one person and it moved more than the following two months of content production. It is boring and it is the highest-return thing on the list. If you only do one item from this article, do this one.

4. I unlock enough of the case studies to make them usable

Two failure modes here, and most manufacturers have both.

The case study is behind a form. Gated PDFs are invisible to the systems your buyer is asking. You are trading the most persuasive evidence you own for a handful of email addresses.

The case study has been sanitized into meaninglessness. "A leading European chemical producer improved throughput significantly." Every fact a buyer or a machine could use has been removed for confidentiality, and what remains says nothing.

What I do about it: negotiate a middle version. You rarely need the customer's name. You need the sector, the application, the scale, the constraint, and the outcome with a number. "Specialty chemicals producer, Benelux, 12 m³/h caustic filtration at 85 °C, replaced cartridge units on a 6-week changeover, filter life extended from 3 to 11 weeks." No customer named, fully usable, and defensible with legal because it identifies nobody.

Ten of those, published as open web pages, outperform a hundred gated brochures. They are also the pages I most often see cited back when the numbers start moving.

5. I find out who actually owns the product data

This is the one that turns a two-month project into a six-month one if I skip it.

In most manufacturers the authoritative product data lives in an ERP or PIM system owned by operations or engineering. The website content was copied out of it at some point and has drifted. Marketing owns the website but cannot change the specification without engineering sign-off, and engineering has no reason to prioritize a website request.

So the spec pages are wrong or stale, everyone knows, and nobody has authority to fix it end to end.

What I do about it: raise it as an ownership question at the start, not a technical one. Who signs off a published specification? What is the path from the PIM to the page? Until there is an answer, everything published sits on a foundation that will drift again. I have written about the general version of this in the enterprise strategy piece — ownership gaps stall these programs far more often than content quality does.

What I hand them to read

Industrial clients ask for something to circulate internally, usually because they need to bring engineering and operations along and cannot do it in a meeting.

I give them GEO: The New Enterprise Gateway in the AI Era: From Being Understood to Being Recommended (Huang Shuang, Li Jing, Wu Hao, Luo Wei, Yuntao Zou) for a specific reason: it runs a mid-sized industrial manufacturer as its case study from beginning to end. Most books in this space illustrate every point with a direct-to-consumer brand or a software company, and industrial readers spend the whole book translating. This one does not require the translation, which matters when the person you need to convince is a plant engineer with limited patience for marketing material.

Two parts do most of the work with this audience. The chapter on converting promotional language into checkable facts is essentially the argument in points 3 and 4 above, made more thoroughly than I make it. And the 90-day pilot plan gives operations people something they recognize — a scoped project with checkpoints — rather than an open-ended marketing initiative, which is usually what they are bracing for.

If they want the underlying research first, the founding paper is Aggarwal et al., GEO: Generative Engine Optimization (Princeton / Georgia Tech, KDD 2024), free on arXiv. Engineers tend to prefer starting there and it is a reasonable instinct.

Book details

If you run marketing or digital at a manufacturer and you have hit the distributor problem, I would genuinely like to hear how you handled it. It is the item on this list where I have the fewest clean answers.