GEO for Manufacturing and Industrial B2B: The 84% Category Absence Problem

Industrial and manufacturing B2B is the worst-performing vertical tested in Maria Dykstra's Algorithmic Authority Index to date: 84% of the 120 vendors tested receive zero AI citations, and when a company has rebranded, AI engines return the old name 78% of the time, with a documented 18 to 24 month recognition lag. If your industrial company has renamed, spun off, or merged in the past two years, budget for AI search actively working against you for the better part of two years.

Industrial buyers research equipment, control systems, and industrial software the same way SaaS buyers research CRMs now: they open ChatGPT or Perplexity before they open a vendor's website. The difference is what happens next. In SaaS, the AI has a decade of G2 reviews, comparison blogs, and Reddit threads to draw on. In industrial B2B, it mostly has twenty years of analyst coverage about eight conglomerates, and not much else.

That gap is the subject of Wave 2 of Maria Dykstra's Algorithmic Authority Index, published July 2026. The study tested 120 industrial B2B vendors, across four tiers from hardware conglomerates to 90 startups founded between 2020 and 2025, against 900 buyer-intent query sessions covering 12 sub-categories: industrial AI platforms, predictive maintenance, digital twin, IIoT and edge, MES, APM, SCADA and DCS modernisation, OT security, QMS, industrial supply chain visibility, energy management, and operational intelligence. Sessions ran across ChatGPT, Perplexity, and Google AI Overview from US-based IPs between 15 June and 6 July 2026, using each platform's default model. Share of Model, the study's composite metric, combines Mention Share, Citation Share, and Recommendation Share per vendor; a score above 5% marks a Category Winner, below 1% marks Category Absence.

Why industrial B2B is a different GEO problem than SaaS

Most GEO advice, ours included, was written for categories where AI models have plenty to work with: thousands of G2 reviews, comparison posts, Reddit threads, product-led growth content. Industrial B2B doesn't have that training data. The vendors are older, the sales cycles are longer, the products are spec-driven, and the internet's record of any given company is thinner and more likely to be twenty years old.

Two consequences follow directly from that thinness.

First, AI models fall back on whatever signal is strongest, which in this category means decades of analyst coverage about the handful of conglomerates that have been showing up in Gartner Magic Quadrants and ARC Advisory Group reports since before most of today's buyers were born. Second, without enough clean, current data to tell close categories apart, models frequently guess wrong about what a company actually does. Neither problem shows up this severely in SaaS, fintech, or e-commerce GEO, which is why the industrial playbook has to be different.

Part 1: Category absence is worse here than anywhere else in the Index

84% of the 120 vendors Dykstra's team tested scored below the 1% Category Absence threshold, meaning they received effectively zero citations across the full 900-session query set. For comparison, the Index's cybersecurity wave, tested under the same methodology, recorded a 73% absence rate. Industrial B2B is the weakest category measured so far.

The absence rate is not evenly distributed. Among Tier 4, the 90 vendors founded between 2020 and 2025, it climbs to 89%. Funding doesn't fix it: the study names specific well-capitalised startups that raised significant rounds and still score as functionally invisible in their own category. Visibility earned in one sub-category doesn't transfer either. Uptake holds 22% Share of Model in predictive maintenance, a genuine Category Winner result, and then disappears entirely from supply chain queries.

The exception the study identifies is vertical specificity. Startups that describe themselves in industry-specific, sector-verticalised language (a named use case, a named sector, a named regulatory context) score 28% higher Share of Model than horizontal generalists making the same underlying pitch in abstract terms. A company selling "industrial AI for pharma serialisation compliance" gives the model something concrete to attach to. A company selling "the operating system for industrial data" doesn't.

Part 2: AI engines confuse your category 68% of the time

Even the vendors that do get returned for a query aren't safe. 68% of vendors that appeared in the query set were operating in a category adjacent to the one actually asked about, not the one queried.

The study documents this directly. A ChatGPT session asked for top MES vendors for pharma and returned Tulip Interfaces, a connected-worker platform, as a primary recommendation, alongside Siemens Opcenter and GE Vernova Proficy, both of which carry the regulatory validation an MES buyer in pharma actually needs and Tulip does not. A separate Perplexity session asked for industrial AI platforms for oil and gas and returned Rockwell Automation and Emerson, both hardware-heavy control system vendors rather than native industrial AI software.

The mechanism behind this, per the study, is inconsistent terminology across a vendor's own marketing surfaces. When a company describes the same product differently on its homepage, its case studies, and its analyst briefings, the model averages the conflicting signals and files it into the wrong competitive set. MES and MOM collapse into one execution layer. Predictive maintenance queries return generic MES providers. SCADA and DCS modernisation queries return historian database vendors. None of this is visible to the buyer asking the question. They just get an answer that quietly points them at the wrong shortlist.

Part 3: Why the incumbents keep winning

Siemens, Rockwell Automation, and GE Vernova alone capture 67% of Share of Model across mature software segments. That concentration tracks physical footprint and institutional validation more than product quality: Siemens wins digital twin platforms (38%) and MES (36%); GE Vernova wins APM (33%); Emerson wins SCADA/DCS (32%).

The pattern isn't absolute. Software pure-plays do win, but only in emerging or tightly specialised niches: C3 AI takes industrial AI platforms at 34%, Claroty takes OT security at 41%, ComplianceQuest takes QMS at 44%. That's a useful distinction for anyone planning a GEO strategy in this category: broad, horizontal categories are where the conglomerates' trust inheritance is strongest, and narrow, newly-defined niches are where a focused vendor still has a real opening.

Part 4: Technical depth is a genuine moat here

This is the part of the study that should change how industrial marketers think about content, not just track it as a data point. The study measured a 100% failure rate on content extraction among the corporate content it tested: gated PDFs and whitepapers score zero citations, because a model can't extract what it can't crawl. High-volume short blog posts get treated as repetitive filler and ignored in favour of longer, data-dense analysis. Abstract horizontal messaging gets excluded from vertical queries entirely.

What does get cited is specific: analyst inclusion (Gartner, ARC Advisory Group, LNS Research), independent trade press (Automation World, Control Engineering), standards and regulatory sources (IEEE Xplore, CISA advisories), and original vendor-published research with a stated methodology, the kind of un-gated compliance data or benchmark study a model can quote directly. The study notes that cybersecurity vendors can earn citation-worthy signal from Reddit and G2 in a single quarter; industrial vendors need an ARC write-up or a Gartner mention, sources with real editorial gatekeeping, and there's no fast substitute for that.

The practical read: in a vertical this thin on training data, real specification tables, honest tradeoff comparisons against named alternatives, and precise, consistent category terminology carry more weight than they would in a more commoditised vertical, because there's comparatively little competing content for a model to weigh them against. Publish the data your gated whitepaper is currently hiding. Say the same thing about your product the same way everywhere. Get into the trade press and analyst reports that actually feed these models, not just the ones your PR team already tracks for humans.

A 90-day roadmap for industrial and manufacturing GEO

Days 1-30: audit and identity cleanup. Test your top 20 to 30 buyer-intent queries per sub-category you compete in, across ChatGPT, Perplexity, and Google AI Overview. Record whether you appear, what category you're filed under, and whether any older brand name still surfaces. If you've rebranded, merged, or spun off in the last three years, add Organization schema with alternateName pointing at every prior identity in the same week. Pull every whitepaper currently behind a lead-gen gate and identify which ones contain data worth publishing openly.

Days 31-60: bridge and publish. Build permanent bridge pages that explain any rebrand or acquisition in plain language, linking old and new identities explicitly, "formerly known as" framing on every page a buyer or a crawler is likely to land on. Un-gate the highest-value technical content: spec tables, benchmark data, compliance documentation. Rewrite category-defining pages so the same product description, the same sub-category terminology, appears consistently across the homepage, case studies, and any analyst briefing materials.

Days 61-90: earn institutional signal. Pitch trade press (Automation World, Control Engineering, sector-specific publications) and relevant analyst firms (Gartner, ARC Advisory Group, LNS Research) with the technical data you just un-gated, not a press release. If you rebranded, this is the step that actually moves parametric memory: third-party coverage under your new name is what teaches the next training cycle who you are, a schema fix alone only repairs live retrieval. Re-run the query audit from day one and compare category placement and citation presence against the baseline.

AY Rank builds GEO programmes for B2B categories where AI models have thin, uneven training data, industrial and manufacturing among them. For the general enterprise AI-search playbook this piece builds on, see our GEO for B2B guide.

FAQ

Why is industrial B2B the worst-performing category in the Algorithmic Authority Index?

Because AI models have comparatively little clean, current training data to draw on for this vertical. Industrial companies are older, sales cycles are longer, and most of what's publicly indexed is decades of analyst coverage about a handful of conglomerates. Wave 2 of the Index recorded an 84% Category Absence rate across 120 tested vendors, versus 73% in the study's cybersecurity wave under the same methodology, making industrial the weakest category measured to date.

My company rebranded two years ago. Why does ChatGPT still use our old name?

Because AI models run on two update cycles, and only one of them moves fast. Live retrieval, what a model pulls from current web pages, updates within two to four weeks. Parametric memory, what gets baked into the model's weights during training, updates on an 18 to 24 month cycle. A schema fix or a website update repairs the first loop immediately but does nothing for the second. Dykstra's study documented this directly: across ten tested industrial rebrands, AI engines returned the old name 78% of the time, with individual cases (Wonderware to AVEVA InTouch, OSIsoft PI to AVEVA PI) still surfacing the old identity 20-plus months after the rebrand.

What can we actually do about the rebrand lag, rather than just wait it out?

Three things, in sequence, not simultaneously. First, add Organization schema with alternateName referencing every prior identity, in the first week. Second, within a month, publish permanent bridge pages that state the old and new names together with explicit "formerly known as" language. Third, over the following months, run a deliberate push for third-party coverage under the new name, trade press, analyst mentions, press coverage, because that's what actually teaches the next training cycle who you are. Schema and bridge pages fix live retrieval; only third-party coverage under the new name moves parametric memory.

Does more content volume help in this category the way it does elsewhere?

Not on its own. The study found a 100% extraction failure rate among the gated PDFs and short blog posts industrial vendors typically publish; models can't cite what they can't crawl, and treat high-volume short posts as repetitive filler. What moves the needle is un-gated, data-dense technical content (spec tables, benchmark studies, compliance documentation) and consistent, vertical-specific terminology. The study found verticalised vendors score 28% higher Share of Model than horizontal generalists making the same pitch in abstract language.

Source: Maria Dykstra, "Algorithmic Authority Index Wave 2: Industrial B2B", published July 2026.