Who AI recommends for pharmaceutical process skids

Pharmaceutical and biopharma buyers increasingly build their supplier shortlist by asking an AI engine — before they contact anyone. The engine names a handful of fabricators, and that handful becomes the consideration set. There is no second page.

We ran the buyer's question across all four major engines, five times each, for eight high-value skid categories — CIP, buffer prep, SIP, filtration, cGMP/API, modular, WFI, chromatography, and bioreactor systems — plus two regional markets. Three findings define the landscape.

One — the engines disagree, and it is structural. In every contestable category, a fabricator dominant on one engine is frequently absent from another. The disagreement is stable across repeated runs because each engine reads from a different evidence base.

Two — visibility is won on the fabricator's own pages. Across six categories, the study found between 73% and 89% of all citations pointing to fabricators' own capability pages rather than to directories, marketplaces, or listicles. This is the empirical core of the report: in this world, the lever is the firm's own site, almost everywhere.

Three — categories divide into three types. Some are contestable (a mid-market fabricator can be named), some are technology-anchored (a dominant brand owns the buyer's attention and fabricators are secondary), and some are locked (a few global incumbents hold near-total share). Knowing which is which is where strategy begins.

How do you source custom stainless steel skids for pharma?

You source custom stainless steel skids for pharma the same way you source any validated system, with one addition that did not exist five years ago. The shortlist most buyers now start from was assembled by an AI engine, and this study found that it is a ranking of visibility rather than of capability. Two practical consequences follow. Run the enquiry on more than one engine, because the rosters barely overlap. Then write the specification tightly enough that the answer has to discriminate, because a generic enquiry returns the same generic names to everybody.

The specification items below are not our recommendations. They are the requirements the engines' own answers tell buyers to ask for, and the details that appear on the capability pages the engines actually cite, as documented later in this report.

One caution about the shortlist you get back. This study cross-referenced the engines' answers against ASME BPE certificate holders and industrial supplier directories, and found credentialed fabricators with decades of pharmaceutical project experience appearing in few or none of the four engines' answers for their core category. A fabricator's absence from an AI answer says nothing about whether it can build your skid. If the enquiry matters, the AI shortlist is a starting point to widen, not a supplier list to work through.

What does a custom stainless steel specification pin down on a pharma skid?

A custom stainless steel specification pins down four things, and only four things, in a form a purchase order can hold a fabricator to: the tubing specification, the product contact surface finish designation, the passivation treatment with the test that confirms it worked, and the electronic records rule the control system has to satisfy. Everything else in the phrase is adjective. Ask for all four by name and the field sorts itself, because a fabricator that routinely builds validated systems answers them off the quote, and one that does not has to go away and find out.

None of this is exotic, which is the point. Every item above is published, checkable before anyone is contacted, and answerable in a sentence by a fabricator that builds these systems weekly. It belongs in a visibility report because the same four details decide the other half of the problem. The capability pages the engines cited in this study state their standards, finishes and control platforms in readable text on the page. The pages the engines skipped describe the same capabilities as adjectives, or bury them inside a datasheet PDF where nothing can extract them. A buyer writing a tighter enquiry and a fabricator writing a more citable page turn out to be doing the same work from opposite sides of the table.

How should the tubing, finish and passivation lines actually be worded?

The tubing line should name ASTM A270 and then name the supplementary requirement, because the two are not the same request. A270 is the specification for seamless and welded austenitic stainless sanitary tubing, covering sizes up to and including 12 in outside diameter, and its published scope is written in the first instance for the dairy and food industry, with pharmaceutical quality available as a supplementary requirement rather than as the default (ASTM A270/A270M, standard scope). An enquiry that says "A270 tubing" and stops there has asked for the dairy grade.

The finish line is where "electropolished" stops being an adjective. ASME BPE designates product contact surface finishes SF0 through SF6. SF1 permits a maximum roughness average of 20 µin, or 0.51 µm, and is reached by mechanical polishing. SF4 permits 15 µin, or 0.38 µm, and requires electropolishing as the final preparation, which is the real difference between them: mechanical polishing leaves a worked surface layer that further abrasion only regenerates, and electropolishing removes it (published by Dockweiler in its ASME BPE surface specification and by Alfa Laval in its guide to pharmaceutical tube surface finish). A quote that says electropolished without an SF designation and an Ra value has committed to nothing measurable.

The passivation line has the same failure mode. ASTM A967 covers chemical passivation treatments for stainless steel parts, including both the nitric acid and the citric acid routes, and it carries alternative tests with acceptance criteria for confirming that a treatment worked (ASTM A967/A967M, standard scope). So the useful wording names the route and the acceptance test rather than the word. Riboflavin coverage testing sits beside this and answers a different question, which is whether a spray device wets every internal surface, not whether the metal is passive.

The records line is the one buyers tend to write as a slogan. Where the skid produces batch records, the governing rule is FDA 21 CFR Part 11, and the thing to ask for is not "Part 11 compliant" but the mechanism behind it: which platform, whose audit trail, how electronic signatures are bound to records, and who holds the validation documentation once the system is handed over. How a fabricator answers that tends to predict the rest of the documentation package, which is why it is worth asking early rather than at FAT.

What documentation should a pharma skid fabricator hand over, and when do you ask for it?

Ask for the documentation list at quote stage rather than at handover, because by then the records either exist or cannot be reconstructed. A custom stainless steel skid for pharmaceutical use arrives with three bundles of paper, and they answer different questions. Mill certificates, weld logs, installation coupons and examination records say how the hardware was built. The qualification protocols and their executed results say that the assembled system does what it was bought to do. The control system records say that the batch data it produces will stand up to an inspector. A fabricator that builds validated systems every week can name all three off the quote. One that does not has to go away and find out, which is the same sorting test the specification lines apply, run on the half of the job that happens after the steel is cut.

The stage names are not the fabricator's invention and they are not negotiable vocabulary. EU GMP Annex 15 sets out qualification as a lifecycle that runs from the user requirement specification through design qualification, factory and site acceptance testing, and installation, operational and performance qualification, and it allows a FAT to be supplemented by a SAT once the equipment reaches the manufacturing site rather than treating the two as duplicates (EudraLex Volume 4, Annex 15: Qualification and Validation, in force since October 2015). An enquiry that uses those names asks for a package the fabricator already knows how to price. An enquiry that asks for "full validation documentation" asks for something nobody can quote.

The weld records have their own published basis, which is worth knowing because they are the line item most often quoted vaguely. ASME BPE is organised into parts, and materials joining sits in Part MJ alongside general requirements, systems design, dimensions and tolerances, surface finishes, metallic and polymeric materials, and certification (ASME BPE-2024, contents). Part MJ is where the visual examination acceptance criteria for bioprocessing welds live, at MJ-8.3 and MJ-8.4, and it is also why installation coupons exist: small diameter sanitary tubing does not take radiographic or ultrasonic inspection usefully, so a coupon welded under field conditions and cut open is the evidence that the qualified procedure is still producing acceptable welds on site, and coupons are inspected against the same criteria as the production welds they stand for (ASME BPE welding requirements, orbital welding reference, read 23 September 2026). Asking how many coupons, cut by whom, and inspected against what is a more searching question than asking whether a shop is ASME BPE compliant.

This belongs in a visibility report for the same reason the specification does. The capability pages the four engines cited in this study name their documentation package in readable text: the protocols they write, the records they deliver, the standard the welds are examined against. The pages the engines skipped say "full documentation package provided" and leave it there, or put the detail in a PDF that nothing extracts. A buyer who cannot tell the two apart from the page will ask an engine, and the engine will hand back the firms whose pages said the specific thing.

How do you widen an AI shortlist into a real bid list?

You widen an AI shortlist by asking the same question four more ways, because each variation returns a partly different set of fabricators. The four are the engine you ask, the words you use, whether you name a region, and an independent credential list to check the answer against. This report and its 27-query follow-up ran all four, and each one surfaced firms the others missed.

The wording variation is the one buyers underestimate. Two phrasings of the same requirement, minutes apart, produced an open field in one case and a single supplier at total share in the other. A buyer who happens to type the second phrasing never learns that the first exists, and nothing in the answer signals that the question was narrow.

The regional variation is the one that adds the most names. Mako Industries, A&G Piping Services, Placer Process Systems and TM Process & Controls lead their state-level queries in this study and appear nowhere in the national maps. Whether any of them suits a given project is a question for the enquiry, not for the engine. The point is that a national query alone never puts them in front of you.

None of this makes the AI answer wrong. It makes it narrow in a way it does not disclose, and the width you need depends on what the skid is worth. For a repeat purchase of a known system the first shortlist may be enough. For a validated system running for a decade, the cost of asking four more questions is an afternoon, and the fabricator best suited to the job has no particular reason to be the one an engine found most legible.

How is the supplier shortlist assembled before anyone is contacted?

The shortlist is assembled by the engine, before a buyer contacts any supplier. Someone sourcing a multi-million-dollar process system no longer opens a dozen browser tabs. They ask an AI engine which firms to consider, and most of that research happens before any fabricator is contacted. The engine's answer is a short list of names — and a fabricator that isn't on it is not ranked lower. It is absent from the decision, and never learns the opportunity existed.

Because each engine reads from a different set of sources, a fabricator's visibility is not one number. It is four — one per engine — and they are rarely the same. Most fabricators have never measured any of them.

What a buyer actually sees. This is Google's literal answer — it names CSI, Paul Mueller, Zajac, and Brinox. Now compare that to the other engines for the same question: Claude leads with IPEC, GEA, Arcadia, and OCS; ChatGPT with IPEC, Brinox, and Kropman; Perplexity with GEA and Centrex. The rosters barely overlap. The same buyer, asking the same question on a different engine, is handed a different shortlist — which is the central finding of this study. (Note, too, that the capabilities Google tells buyers to require — ASME BPE, 316L with electropolishing, 21 CFR Part 11 — are precisely the page details that determine citation, examined later in this report.)

Which pharmaceutical skid categories are actually open to an independent fabricator?

Reading the engines' actual answers, not just the share-of-voice tallies, reveals that the eight categories fall into three structurally distinct groups. A category can look open by the numbers while being effectively closed to an independent fabricator in practice — so the distinction matters.

The practical reading for an independent fabricator: visibility effort belongs in the contestable categories you actually serve, where being named is achievable — not the locked ones, where a few global brands own the answer outright.

Where independent fabricators win directly: CIP, buffer prep, and SIP

CIP, buffer prep, and SIP share a structure: the engines name independent stainless-steel fabricators directly, mid-market firms appear alongside the global players, and citations point overwhelmingly to fabricators' own pages. All three are cleaned in place and reused rather than thrown away, which is why they stay in stainless steel and stay specified per site, and the format split between single-use and fabricated stainless biologics skids tracks the same line. They are where an independent fabricator's visibility effort pays off — and where the cross-engine divergence is most striking.

CIP skids — two co-leaders, reached by opposite routes

Clean-in-place skids are the clearest illustration of cross-engine divergence in this study's dataset. The two co-leaders by combined share of voice, IPEC and GEA, both reach 60% — but by completely different paths. IPEC scores 100% on ChatGPT and Claude yet 0% on Perplexity. GEA leads Perplexity (80%) and Claude (100%) but trails on ChatGPT. Central States Industrial tops Google at 80% yet is absent from ChatGPT entirely. No firm is named by all four engines.

Buffer prep skids — a split market of platform brands and custom fabricators

Buffer prep divides into two layers: large platform vendors that sell named buffer-management products, and custom skid fabricators that build to specification. In this study KeyPlants leads combined share at 65%, but the divergence persists: IPEC scores 100% on Google and 0% on Claude; ABEC and Exentec both top Google yet barely register elsewhere; Cytiva and Hamilton are Claude-heavy and absent on Google.

SIP skids — contestable, with a more settled top tier

In this study, sterilize-in-place skids are contestable but more concentrated at the top: Cotter Brothers (80%), Engineered Biosystems (75%), and DCI (75%) lead. Even here the divergence is total — Cotter scores 100% on Google, ChatGPT, and Perplexity but only 20% on Claude; DCI scores 0% on ChatGPT yet 100% on the other three. The fabricator field is real and named, just with a firmer leadership tier than CIP.

Across all three contestable categories, one pattern is constant: read any single row and the fabricator is invisible on at least one major engine where a competitor is named. There is no firm that wins everywhere — which means there is room everywhere.

Why do filtration, cGMP, and modular skids look open but stay closed?

Filtration, cGMP/API, and modular skids all look open on a share-of-voice chart, where no single firm dominates. But reading the actual answers reveals a different structure: a dominant technology brand, or a different class of equipment maker entirely, owns the buyer's attention, and independent fabricators appear only as a secondary tier. An independent fabricator can be named here, but it is competing for the runner-up slot beneath a brand the engine reaches for first.

Filtration skids — the filter-technology brands lead

In filtration, the engines lead with the firm that makes the filter, not the firm that builds the skid. Mott, a porous-metal filtration-element specialist, tops combined share at 65% in this study, and its own domain is cited 19 times, more than any fabricator. Independent fabricators (G&G Technologies, IPEC, CSI, Cotter, Bioneau, ACE, Arcadia, Noreside) are all present, but beneath the filter-technology names. The category is contestable, but a fabricator competing here contends with the filter brand for the buyer's first attention.

cGMP / API process skids — the reactor and chemistry giants lead

"API production" pulls chemical-synthesis equipment — glass-lined reactors, filter-dryers, reaction trains — and the engines lead with the large reactor and process-equipment specialists: GMM Pfaudler, De Dietrich, GEA. This is partly a different equipment category than sanitary skid fabrication. Arcadia tops combined share at 70% in this study and is the most-cited domain (22 citations), so independent fabricators are genuinely present — but they sit within a field defined by the reactor incumbents.

Modular skids for biotech startups — a heterogeneous field

By the numbers this is the most "open" category in the study: no firm exceeds 50% combined. But the openness reflects a heterogeneous field rather than an open lane for fabricators: large engineering firms (Re:Build Optimation), modular facility-platform providers, single-use technology specialists (CPi Biotech), and a few stainless fabricators all mixed together. The "startup" framing also skews toward the buyers least able to invest. A fabricator appears here, but among a crowd of different business types competing for the same query.

The lesson of the technology-anchored categories: a high contestability score is not the same as an open lane. Filtration, cGMP, and modular all look winnable on a chart — but in each, the buyer's first attention belongs to a filter brand, a reactor maker, or a crowd of adjacent business types, not to the independent fabricator.

Where the answer is owned outright: WFI, chromatography, and bioreactor

For contrast, this study found three categories owned outright: WFI, chromatography, and bioreactor skids are effectively closed. Where the contestable categories have co-leaders in the 60–80% range and no firm winning everywhere, these are defined by a handful of global incumbents holding near-total share — in two cases a literal 100% across all four engines at once. This is the single most persuasive exhibit in the report: the difference between a category where the leader is cited in 60% of runs, and one where the leader is cited in 100% of runs on every engine, is the difference between a winnable market and a closed one.

WFI systems — two firms own every engine

In water-for-injection systems, the study found Paul Mueller and MECO each holding 100% share of voice across all four engines. A fabricator trying to be named here is competing against total incumbency on every engine simultaneously.

Chromatography skids — Cytiva sweeps every engine

Chromatography is owned by the major life-science instrument brands. Cytiva holds 100% across all four engines; Sartorius follows at 90%. These are the only across-the-board sweeps in the study's dataset alongside WFI. A fabricator's chromatography-skid visibility is, in practice, not contestable.

Bioreactor / fermentation skids — concentrated, though less absolute

Bioreactor skids are concentrated rather than fully locked. In this study Rodem leads at 70%, GEA at 55%, and the large bioprocess names cluster here. There is marginally more movement than WFI or chromatography, but the category is still dominated by a handful of large specialists rather than open to independent fabricators.

The contrast is the strategy. Across this study, a contestable category's leader is named in 60% to 80% of runs and no firm wins all four engines. A locked category's leader is named in 100% of runs on every engine at once. Visibility effort spent on a locked category is spent against total incumbency; spent on a contestable one, it can move a firm onto the shortlist.

Which fabricators do AI engines name for regional queries?

National category queries are dominated by the global names. But buyers also ask regionally, and when they do, a completely different set of fabricators surfaces — smaller, local firms that never appear in the national maps. A fabricator invisible nationally behind GEA and Cytiva can be the named leader in its own region. For an independent fabricator, the regional query is often the more winnable battleground.

The names tell the story: Mako Industries, A&G Piping, Placer Process Systems, McKenna — fabricators that appear nowhere in the national category maps lead their regional queries. Nationally invisible, regionally winnable.

What sources do the engines actually cite?

Behind every AI answer is a set of cited sources. We classified them across six categories — and the result is the empirical backbone of this report. The share of citations pointing to fabricators' own websites, rather than directories, marketplaces, or listicles, lands in a remarkably tight band:

Between 73% and 89% across six independent categories in this study is not noise, it is structure. In this world, there is no dominant marketplace or directory standing between a fabricator and its citation. The engines read fabricators' own capability pages and cite the firms whose pages clearly describe what they build. That means AI visibility in this category is largely controlled by the fabricator — and it is fixable.

CIP skids — where the citations come from

Buffer prep skids — where the citations come from

What a cited capability page contains

Reading the pages the engines actually cite, a consistent pattern emerges. The fabricators that get named have dedicated pages that state, explicitly and specifically:

• The system types they build — single-tank, multi-tank, and no-tank CIP; CIP/SIP integration; inline dilution versus batch buffer prep
• Controls and automation by name — PLC/SCADA platform (Siemens, Rockwell/Allen-Bradley), S88 batch architecture, 21 CFR Part 11 electronic records
• Validation posture — cGMP, ASME BPE, IQ/OQ/PQ documentation packages, FAT/SAT protocols
• Materials and finish — 316L stainless for product contact, electropolished internals with stated Ra values
• Process science — conductivity/TOC rinse control, orbital welding with weld logs, material traceability

Fabricators whose pages describe their work in general terms — "custom stainless fabrication," "quality process systems" — tend not to be cited, even when their engineering is equal or better. The gap between the cited and the absent is frequently a content gap, not a capability gap.

Does an AI shortlist rank capability or visibility?

An AI shortlist ranks visibility, not capability. The previous sections measured who the engines name, and it is just as revealing to ask who they don't. We cross-referenced the engines' answers against an independent reference set of established pharmaceutical skid fabricators — drawn not from AI, but from ASME Bioprocessing Equipment (BPE) certificate holders and industrial supplier directories, the credentials and registries the industry itself uses to identify serious fabricators.

The result: a number of credentialed, established fabricators — firms holding the industry's gold-standard certifications, with decades of pharmaceutical project experience — appear in few or none of the four engines' answers for their core category. They are real, qualified, and capable. They are simply not visible to the engines their buyers now use.

Absence from an AI answer is not a verdict on a fabricator's capability. A firm can hold every relevant certification, build excellent systems, and still be invisible — because visibility is decided by how a firm's pages are written and read, not by how well it welds. That is the gap, and it is the opposite of permanent.

This reframes what the engines actually produce. An AI shortlist looks authoritative — a clean list of "the best" fabricators. But it is not a ranking of capability. It is a ranking of visibility: the firms whose pages the engines could read, extract, and trust. For a qualified fabricator that has been overlooked, that distinction is the entire opportunity — the gap is in the pages, and the pages are within the firm's control.

How was this measured?

Every figure in this report was measured the same way. Each query was run five times on each of the four engines (ChatGPT, Google AI Overview, Perplexity, and Claude) on their default consumer configurations, which is what a typical buyer experiences. Share of voice is the percentage of runs in which a firm was named. Citations were captured and classified by source type: a fabricator's own site, a third-party listicle, a directory, or media.

What should you do if you fabricate process skids?

If you fabricate process skids, start by measuring: you have gaps you cannot see. Your firm's visibility is four different numbers across four engines, and at least one is almost certainly zero where a competitor is named. The only way to know is to measure all four.

The gaps are largely fixable through your own pages. Because the engines cite fabricators' own capability pages, between 73% and 89% of the time across the six categories in this study, the primary lever is how clearly and specifically those pages document your systems, controls, validation posture, materials, and process science, in a form the engines can read and extract.

Where you compete matters most. Focus on the contestable categories you actually serve — CIP, buffer prep, SIP — and on regional queries, where the field is smaller and more winnable. The technology-anchored categories (filtration, cGMP, modular) are harder, and the locked ones (WFI, chromatography, bioreactor) are owned outright. Spending visibility effort against total incumbency is spending it in the wrong place.

AI visibility for pharmaceutical process skid fabricators

This report was produced by SIGNALS, a BlackSig Systems company. We measure how AI engines recommend suppliers in specific B2B categories, and we help firms close the visibility gaps the measurement reveals.

If you fabricate pharmaceutical process skids and want to see exactly where your firm appears — and doesn't — across ChatGPT, Google, Perplexity, and Claude, we can produce a firm-specific AI Visibility Snapshot mapping your share of voice against your named competitors, category by category.

If you fabricate process skids, here's where to start →