Industrial AI Visibility Benchmark, Wave 1

Executive Summary

Across six industrial and commercial categories, 71 real brands, and 36 buyer questions put to four AI assistants, 28 of the 71 brands were never recommended by any assistant. That number splits two ways. Twenty-four never appeared in a single answer in their own category at all: not ranked low, not mentioned and passed over, absent. The remaining four were named but never once put forward as an option.

Being large does not protect a brand from this. Greenheck, Munters and Modine are absent from commercial HVAC. Konecranes, Combilift and Hytrol are absent from material handling. These are established manufacturers with real market share, and when a buyer asks an assistant who to consider, they do not come up. This study did not measure company size, so it cannot say what predicts visibility. It can say that being large plainly does not guarantee it.

The second finding is about where the risk sits. The assistants name almost no vendors until the buyer asks a vendor-shaped question. The 24 education and evaluation questions were put to all four assistants, 96 probes in total, and produced 10 brand mentions between them. Not one was a recommendation. Claude and ChatGPT named nothing at all. Brand visibility exists only at the comparison and vendor-selection stages, which means the entire early half of the buyer journey is being answered without a single company in the room.

Third: the leaders are not the same leaders across assistants. No category has a single brand that all four assistants place first on its own. In three of the six, one brand is at least tied for top on every assistant. In the other three the assistants disagree outright. Absence is a consensus, so a brand missing from four systems cannot dismiss it as one engine's quirk. Leadership is not, so a brand sitting top of one assistant should not read that as a position it holds.

---

How to Read This Study

"Recommended" is a higher bar than "named." A brand is named when it appears anywhere in an answer. It is recommended when the assistant puts it forward as an option the buyer should consider. The gap between the two is where some of the sharpest findings sit. PTI Security Systems is named on 18 of 24 probes, exactly as often as OpenTech Alliance, but recommended on 6 against OpenTech's 15. The assistants know PTI. They reach for it less.

Rankings run on the 36 decision-stage questions, not all 60. The frozen question set carries a buyer stage on every question, set before any probe ran. Education and evaluation questions produce almost no brand mentions on any assistant, so scoring visibility across all 60 would dilute every brand's rate with 24 questions where no brand was going to be named regardless. That measures question mix, not visibility. The early-stage questions are reported separately below, because what happens there is a finding in its own right.

Every brand's denominator is 24. Six decision questions per category, each put to four assistants.

Wave 1 is a snapshot, not a trend. It shows where things stood in August 2026. Trend claims become honest when Wave 2 exists to compare against.

---

Per-Category Results

Recommendation rate is recommendations divided by 24 probes. "Named" counts every appearance.

Industrial Coatings

The only category with two brands above 70%. Note that this is not the same as being the most concentrated: coatings spreads its recommendations more widely than any other category, and cross-category finding 5 gives the numbers.

Never recommended: Cardinal Paint & Powder, VersaFlex, Cerakote (NIC Industries), Denso, National Coatings & Supplies.

Belzona is the category's clearest near-miss. It surfaces once and is never put forward.

Commercial HVAC

Never recommended: Modine Manufacturing, Greenheck Group, Munters.

Three brands tie at the top on 14, and the tie holds inside every assistant: Carrier, Daikin and Trane are level on all four. No assistant separates them. Johnson Controls shows the named-but-not-recommended pattern: 11 mentions, 7 recommendations.

HVLS / Industrial Ceiling Fans

One recommendation separates first from second, and the more-mentioned brand is the one that loses. Only commercial HVAC is closer, where the top three are level.

Never recommended: Entrematic (Typhoon HVLS), Brock HVLS, Barron Equipment, Envira North Systems (Jazz), Altra Air (Sailfin).

Worth noting precisely: Big Ass Fans is named more often (19 against 16) and still ends up recommended less, 14 against MacroAir's 15. MacroAir leads the category on recommendations while being named three fewer times. MacroAir is a Synthesis Insights client, disclosed here rather than footnoted.

Five of eleven brands in this category are invisible, second only to warehouse racking.

Material Handling Equipment

Never recommended: Konecranes, Combilift, Hytrol.

Toyota is top-ranked on all four assistants, outright on three of them and tied with four other brands on Perplexity, which recommended nothing more than twice in this category.

Self-Storage Access Control Vendors

Never appeared at all: Paxton Access (Net2), BearBox.

Named but never recommended: QuikStor, Stor-Guard.

This category carries the study's sharpest named-versus-recommended gap. PTI matches OpenTech on mentions and trails it by nine recommendations. SpiderDoor is a Synthesis Insights client, disclosed.

Warehouse Racking / Storage Systems

Six of thirteen brands here were never named at all, the highest count in the study.

Never recommended: Tennsco, Republic Storage Products, Husky Rack & Wire, Bulldog Rack Company, Cogan, Konstant.

Frazier is named eight times and recommended twice, the sharpest named-but-not-recommended gap in the category.

---

Cross-Category Findings

2. Almost nobody is named until the buyer asks who. Mentions per question by buyer stage, all four assistants pooled:

Across 96 early-stage probes there were 10 brand mentions. Claude and ChatGPT returned none. Perplexity returned one, a passing reference to Toyota inside an explanation of aisle widths. The other nine were Gemini's, and all nine are parenthetical examples of the shape "e.g., PTI, OpenTech, PDK" rather than answers to who the buyer should use. Every one of the 10 is labelled named; none is labelled recommended.

The questions were not the constraint. Several of them are natural places to name a brand, including what a facility manager should look for when choosing an HVAC company, and whether to buy new or used pallet racking. The assistants had the opening and did not take it. This is the study's most actionable finding: the early buyer journey is a category with no vendors in it. A manufacturer that becomes the answer to "what kind of coating should I use for steel exposed to salt air" is competing against nobody.

3. The assistants agree on who exists and disagree on who wins. 30 of the 47 brands named by anyone were named by all four assistants, and only four brands were named by a single assistant. Absence is the same story everywhere: 24 brands are missing from all four. But the top of each category is not stable. No category has a brand that every assistant places first on its own. In three, one brand is at least tied for top on all four: Toyota in material handling, Interlake Mecalux in racking, and a three-way tie between Carrier, Daikin and Trane in commercial HVAC. In the other three, HVLS, coatings and access control, the assistants disagree outright.

The two halves point in opposite directions and both are useful. Invisibility is a consensus verdict, so a brand missing from four independent systems cannot write it off as one engine's quirk. Leadership is not a consensus verdict, so a brand sitting top of one assistant should not read that as a position it holds. For a manufacturer the practical reading is that the floor is real and the ceiling is soft.

4. Being named is not being recommended, and the gap is a real diagnostic. PTI Security Systems (18 named, 6 recommended), Johnson Controls (11 and 7), Frazier Industrial (8 and 2) are all known to the assistants and reached for less than their visibility suggests. This is a different problem from invisibility and needs a different fix.

5. Concentration varies more than expected. The top three brands hold 86% of recommendations in HVLS and 76% in access control, against 67% in material handling and 64% in coatings. Categories are not equally winnable.

---

Methodology

What this measures. Whether AI assistants recommend real brands when industrial buyers ask the questions those buyers actually ask.

The instrument. Probes ran on a purpose-built harness rather than a commercial AI-visibility tool. Every raw answer is stored in full and read against the category's complete brand list, so every brand that came up gets credit for coming up, rather than tracking one configured target brand at a time. Transcripts are retained, which makes every number here re-checkable against the text that produced it.

The engines. Four: ChatGPT with search (OpenAI's Responses API), Claude, Perplexity Sonar, and Gemini with Google Search grounding. Google AI Overviews is not included in Wave 1.

Gemini's answers carry search metadata the other three do not expose. It ran a live search on 29 of its 60 questions and answered the remaining 31 without searching, and the searched questions were concentrated in the later, vendor-oriented stages. All four assistants show the same decline in brand mentions at the early stages, which suggests the pattern is not specific to Gemini. Wave 1 does not establish why it happens.

The three original engines were probed on 18 and 19 August 2026 and Gemini on 24 August. That six-day gap is stated rather than smoothed over.

How the questions were built. Sixty questions across six categories, every one traced to real buyer evidence rather than to an assistant's own idea of what buyers ask. Sources were search-volume and related-query data, distributor and marketplace buyer-question pages, trade-association buyer guides, industry forum language, and questions appearing independently across many competing vendors' FAQ pages. The set was frozen before probing and is version-locked, so Wave 2 asks identical questions.

What "recommended" means. Each brand gets one tag per probe, read from the answer text: absent, named, or recommended. Named means the brand appears. Recommended means the assistant puts it forward as an option to consider. Classification runs on a language model reading the answer text, three independent passes per label with the majority taken, and every label stores the quote it was drawn from, so any tag can be audited. Of 433 labels, 401 were unanimous across all three passes and 32 carried a two-of-three majority.

Scoring scope. Every brand is scored against the questions in its own category, which is how a category ranking stays a ranking. A brand can therefore be absent here and still surface in another category's answers. Rankings use the 36 comparison and vendor-selection questions. The 24 education and evaluation questions are reported separately in cross-category finding 2. Stage labels were fixed when the question set froze, not derived from results.

Brand selection. Each category's brands come from named public sources: trade directories, association member listings, market rankings. Every inclusion traces to a stated source.

Limitations.

• One snapshot. No trend claim until Wave 2.
• One round per question. Assistants vary their answers between runs, and that variance is not measured in Wave 1. The harness supports multiple rounds and future waves should use them.
• Brand lists come from public directories, which skew toward companies with a findable public presence. A brand buyers know but directories miss would not be on the list.
• Every finding states what appeared in an answer. Where a pattern looks causal it is labeled an observed correlate. No claim is made about why a model named or omitted a brand.
• Absence in an answer means the brand did not appear, not that an assistant judged it inferior.

---

About

Synthesis Insights runs paid media for industrial and commercial manufacturers. This study is practitioner work, not vendor research, and Synthesis Insights does not score its own site in it.

---

Cite This Study

This study is free to cite, quote and reproduce with attribution. No permission needed. If you are writing about AI search visibility in industrial and commercial markets, these are the figures and the attribution line.

Key findings, verbatim:

• 28 of 71 industrial manufacturers were never recommended by any AI assistant. Twenty-four of those never appeared in a single answer in their own category at all.
• The 24 education and evaluation questions, put to all four assistants across 96 probes, produced 10 brand mentions between them and not one recommendation. Claude and ChatGPT named nothing at all.
• No category had a single brand that all four assistants placed first on its own. Absence is a consensus across assistants. Leadership is not.
• Being large does not protect a brand. Greenheck, Munters and Modine are absent from commercial HVAC. Konecranes, Combilift and Hytrol are absent from material handling.

Wave 1 covered six categories, 71 brands and 36 decision-stage questions drawn from a frozen set of 60, put to ChatGPT, Claude, Perplexity and Gemini between 18 and 24 August 2026. Every probe transcript and every labelled classification is available on request.