I ran a live GEO audit. Here’s what B2B distributors should actually look at.
I built a free GEO audit for B2B distributors and manufacturers because I needed it.
Every existing tool I tested was either a $400/month enterprise SEO platform with GEO features bolted on, or a free generic checker that scored your schema and called it a day. Useful work in both cases — but neither showed me what I actually needed in the first ten minutes of a client meeting: what AI says about you right now, and what to do about it.
So I built one. Four buyer-realistic prompts, three engines, transcripts verbatim. No score-only dashboard. No subscription. Ninety seconds. It’s not a competitor to the enterprise platforms — it’s a different shape of tool, built for the specific job of showing a B2B operator what their buyers see when they ask AI for a recommendation.
Run it here: https://humanafterall.ca/geo-audit/
To show what it actually surfaces, here’s what came back when I ran it against a leading North American electrical distributor — a billion-dollar operator, part of a global parent group, the kind of company every electrician in their territory has been ordering from for thirty years. The industry knows them. AI didn’t.
The score said 38 out of 100. The transcripts said something more useful.
A score-only audit would tell this distributor their site has problems and recommend a schema audit. The transcripts told me three patterns no score could surface — and these are the patterns I now see in almost every B2B audit I run.
Pattern 1: Entity collision
When ChatGPT was asked who this distributor’s competitors are, it confidently described a US telecommunications company. Edge computing, network solutions, cloud infrastructure. Wrong industry, wrong country, wrong company.
The brand name overlapped with a much larger US entity in a completely different sector. When AI is forced to choose between a regional B2B distributor with a thin entity graph and a multibillion-dollar telecom with decades of structured presence on Wikipedia, Wikidata, and the SEC, it picks the telecom every time. The distributor disappears.
This is the single most common failure mode for B2B distributors I audit. If your company name overlaps with a parent brand, an M&A legacy entity, a generic industry term, or a much larger company in a different sector, AI will default to whichever entity has the stronger graph. Your real category will be silently overwritten.
The fix is not technical SEO. It’s entity disambiguation: a Wikidata entry tied to your actual parent company, Organization schema with sameAs links to LinkedIn, Crunchbase, and industry directories, and the same category language across every page on your site so AI engines have something concrete to anchor to. None of which a score-based audit will tell you to do, because none of those signals show up in a robots.txt scan.
Pattern 2: Category drift
ChatGPT placed this distributor in telecommunications. Gemini placed them correctly in electrical distribution. Claude placed them nowhere — refused to engage at all.
Three engines, three different answers. That is not an AI problem. That is a content problem. Across the web, this company’s name does not co-occur with one consistent category term often enough for AI to converge. Each engine grabbed whatever signal it could find and ran with it.
For a $50M to $500M distributor, this looks like: ChatGPT calls you a wholesaler, Gemini calls you a manufacturer, Perplexity says you’re a logistics company. Your buyers ask different engines and get different stories. Your sales team spends the first ten minutes of every discovery call explaining what you actually sell.
The fix is editorial, not technical. Pick one category. Use the same words for it on your homepage, your About page, your case studies, your LinkedIn, your podcast appearances, your industry-publication bylines. Repetition across independent sources is what builds the entity-to-category association AI engines reward.
Pattern 3: The refusal as diagnosis
Claude refused to answer three of the four prompts. Variations on “I don’t have reliable information about this company in my training data.”
That looks like a tool limitation. It is not.
When an LLM refuses, it is telling you something specific: there is insufficient training-data co-occurrence between the company name and the category to support a confident answer. The model could guess. It chose not to. That is the cleanest possible signal that the third-party mention footprint is too thin.
For mid-market distributors, this is the most common audit result and the most actionable. It means no industry publication, no analyst report, no podcast transcript, no Clutch review, no Reddit thread, no LinkedIn longform, no G2 listing has tied the company name to its category at sufficient volume to enter the training corpus. The moat isn’t invisibility. It’s unrecognizability.
What made this audit particularly instructive: Claude actively told the buyer to “check Clutch or G2” when evaluating vendors in this category. The distributor has no presence on either platform. Every time Claude gives that advice, the deal walks.
And then there’s the technical layer
The audit also surfaced what most score-based tools focus on exclusively: the technical signals. For this distributor, the picture was equally bleak.
No structured data. No Organization schema, no sameAs links, no FAQ schema. The site has the basics — title tags, meta descriptions, canonicals, HTTPS, hreflang for multilingual — but nothing AI-specific. No llms.txt file. The robots.txt exists but has no rules for AI crawlers. Author signals are present but credentials are thin.
Two specific easy wins beyond what the audit flagged: FAQ schema with questions phrased the way buyers actually ask them, and a Quick Answer block — a 40-to-80-word direct response at the top of each key page that AI can extract verbatim. Both are emphasized across the GEO literature as disproportionately high-leverage. They take a developer half a day.
This is the layer score-based audits do well. It’s also the easier half to fix. Schema is a development ticket. An llms.txt file is a thirty-minute job. AI crawler rules in robots.txt are a one-line update.
But shipping all of it cleanly wouldn’t have changed Claude’s refusal, ChatGPT’s telecom confusion, or the category drift across engines. The technical layer is necessary. It is not sufficient. The editorial work — entity disambiguation, named case studies, third-party mentions — is where the score actually moves.
Why scores hide all of this
GEO isn’t a separate discipline from SEO. Yoast’s position, echoed by Google’s Gary Illyes, is that foundational SEO already covers most of it. What’s new is the emphasis: entity association, unlinked brand mentions, and structured extractability. Score-based audits check the SEO layer. They miss the new emphases.
A 38 out of 100 tells you something is wrong. It does not tell you that your brand is colliding with a US telecom in ChatGPT’s training data, that your category is split three ways across engines, or that Claude is actively redirecting your buyers to platforms where you have no profile.
Score-based audits are SEO tools wearing GEO branding. They were built for a world where the answer was “rank higher” and the signal was traffic. GEO is a different game. The signal is citation. The answer is editorial. The score is a smoke detector that tells you there is a fire but not which floor.
The transcript is the floor plan.
Three fixes that move the number
For any B2B distributor or manufacturer reading this, the order is the same.
Fix entity disambiguation first. Wikidata entry, Organization schema, sameAs links to your parent company and LinkedIn, consistent name-address-phone across directories. This is foundational. If AI doesn’t know who you are, nothing else matters.
Publish named case studies with hard metrics. “Reduced project lead time by 40 percent for a hospital build in Montreal.” Three case studies on your domain, syndicated to industry publications. AI training pipelines reward independent co-occurrence between your name, your category, and measurable outcomes. Generic homepage copy does not survive this filter. Academic research on GEO methods (Aggarwal et al., GEO-bench) found that adding concrete statistics and source citations produced 22 to 37 percent visibility lifts in AI-generated answers — with even larger gains for non-dominant entities, which describes most mid-market distributors.
Earn three to five third-party mentions where your name and your specific category term appear in the same sentence, on a domain that isn’t yours. Trade publication, analyst report, podcast transcript, conference recap. This is how the refusal turns into a recommendation.
None of this is fast. None of it is a plugin. All of it compounds.
Why this matters now
According to Yoast, 42 percent of users now start their research with an LLM rather than a traditional search engine. For B2B specifically, that figure is the leading edge of a shift that’s already changing how deals enter your pipeline.
B2B distributors spent thirty years building relationship moats — dedicated reps, account history, supplier loyalty, the Tuesday-morning phone call. Those moats are still real. But they’re being routed around. The buyer who used to call you first is now asking ChatGPT first, and ChatGPT is making a recommendation before your rep ever knows the deal is in motion.
You don’t lose the deal. You lose the discovery call. The deal goes to whoever AI named in the paragraph.
If you haven’t read your paragraph yet, that’s where to start.
Run yours: https://humanafterall.ca/geo-audit/
If your score comes back ugly, you’re in good company. Even billion-dollar operators with thirty years of brand equity are coming back at 38 out of 100. The score is the start of the conversation, not the end.
Subscribe to the B2B eCommerce Newsletter for weekly implementation pieces on Shopify Plus, Akeneo PIM, GEO, and what’s actually working in mid-market B2B commerce: b2becommerce.substack.com