01What Is Actually Happening
Two different invisibilities, usually confused with one another
The Short Version
Not being found on Google and not being named by an AI are two separate failures. They have different causes, different tests and different fixes — and a company can have one without the other.
The first failure is about pages. A buyer types "supplier of X part for Y process" and your site has no page about that part, that process or that application — only a home page, an "about us", a product line with three lines of description and a PDF catalogue. There is nothing for the search engine to rank because there is nothing written. The second failure is about description. Even when you do rank, an assistant asked "who makes X in Brazil" will name whoever it has read the most about in a form it can repeat: companies described in text, in several places, with specification, capacity, certification and application spelled out. If your company only exists as a logo and a phone number, the model has nothing to say about you and stays quiet.
There is a detail that upsets the owner of a good factory, and it is worth saying plainly: neither system is judging how good your product is. Nobody at Google or at OpenAI compared your gearbox with the competitor's. What both systems measure is how much readable, checkable material exists about each company. A supplier with worse tooling and better documentation wins that comparison every time, because the comparison was never about the tooling.
Nobody compared your product with the competitor's. What got compared was how much readable material exists about each company — and that is a decision you make, not a verdict you receive.
The market gave names to the work of fixing each one: making the site answer the questions a buyer asks is what gets called AEO, and getting your company to be named inside the answers generative models produce is what gets called GEO. The names matter less than the order — there is no point chasing citations in an assistant while the site still has nothing written for the assistant to read. How the second half works is in industrial GEO.
02Why the Competitor Shows Up
Why the competitor gets named and you do not
When you look at the competitor that keeps getting named, four things are almost always true about them and not about you. None of the four is expensive. All four take time.
What the Cited Competitor Has That You Do Not
Text Instead of PDF
The specification is on an HTML page — dimensions, materials, tolerances, applications — not locked inside a catalogue download. A PDF behind a form is, for practical purposes, a closed door for both the crawler and the model.
→
One Page per Question
Instead of one generic "products" page, there are twenty pages each answering one real buying question: which material for which chemical, what lead time for what quantity, which standard applies to which sector. Twenty questions answered means twenty chances to be the source.
→
Mentions Outside the Site
The company name appears in trade association listings, sector magazines, supplier directories and technical talks. A model that reads the same company name in three independent places treats it as established; a name that only exists on its own site stays uncertain.
Consistent Data:
Company name, address, phone number and what the company actually makes are written the same way everywhere they appear. Three different company names across the site footer, the invoice, the trade listing and the Google profile turn one supplier into three half-known ones — and a model that is unsure about who you are names someone it is sure about.
The order in which those four get fixed matters more than the effort put into each. Text before mentions, because a mention pointing to a site with nothing on it converts nothing. Questions before volume, because ten pages answering real buying questions beat a hundred posts about market trends. The applied sequence for the assistant side, with what to publish first, is in how to show up in ChatGPT for manufacturers.
03The One-Afternoon Diagnosis
How to find out in one afternoon which problem you actually have
Step 1
Write the 12 Buying Questions
Ask the two most experienced sales reps what customers ask on the first call, and write down twelve of those questions as the buyer phrases them — never as your catalogue phrases them. "Which seal holds up to 180 °C with sulphuric acid" is a buying question. "Sealing solutions" is a brochure heading.
Input
Step 2
Run the Same 12 in Both Places
Type each question into Google and into ChatGPT, Gemini and Perplexity, in a logged-out window. Record two columns: does your site appear in the first ten results, and does your company name appear in the assistant answer. Twelve questions across four surfaces is 48 checks and fits in an afternoon.
Measurement
Step 3
Read the Two Columns Together
Absent from both columns: the problem is content — there are no pages answering those questions. Present on Google and absent from the assistants: the problem is description and external mentions, not ranking. Present in the assistants with wrong facts about you: the problem is inconsistent data, and it is the cheapest of the three to fix.
Verdict
Do this manually the first time, so you see the answers with your own eyes — the argument with the board is much shorter when the director reads the competitor being recommended in a screenshot. Repeating it every month by hand is where it falls apart, and where turning those 48 checks into a tracked number makes sense. What that number is and how it gets measured is in the share of model audit.
04The 10 Questions in This Cluster
The ten questions inside this problem
Each page below answers one specific question that comes up when a manufacturer discovers it is invisible. Start with the one that matches what your test showed:
05FAQ
FAQ
ChatGPT recommends my competitor and not my company. Why? +
Because it has read more usable text about the competitor than about you. The model is not comparing products — it names whoever it can describe with specification, application and evidence in more than one independent source. A company whose site is a PDF catalogue and a contact form gives the model nothing to repeat, so it repeats someone else.
My site ranks well on Google. Why does the AI still not cite me? +
Ranking and being quoted reward different things. A page can rank on brand and internal links while offering no self-contained paragraph an assistant can lift as an answer. Rewrite the top pages so each one opens with a direct two-to-four-sentence answer to a specific question, then keep the depth below it.
How long until an AI starts naming my company? +
Three to six months for the first consistent citations, counting from when the pages are published and not from when the project is approved. Assistants that browse the live web pick up changes in weeks; the ones answering from training data take much longer. In a low-competition industrial niche the window tends to be at the shorter end.
The AI states something wrong about my company. Can that be corrected? +
Yes, by correcting the sources, not the model. Find where the wrong fact is published — an outdated trade directory, an old address on a supplier listing, an abandoned social profile — fix it at the source, and publish the correct version on your own site in plain text. Assistants that browse the web update within weeks.
Do I have to pay something to appear in ChatGPT? +
There is no ad slot to buy inside the answer, and anyone selling you a guaranteed placement is selling something that does not exist. What you pay for is the work: publishing readable content, fixing the site structure and building external mentions. The cost is production, not media.
My whole product line is a PDF catalogue. Does that hurt? +
It hurts a lot, especially if the PDF sits behind a form. Keep the catalogue for whoever asks for it, but move the specification onto HTML pages — one page per product family, with dimensions, materials, applications and standards written as text. That single change is usually the biggest gain available to an industrial site.
How do I check today whether the AI knows my company at all? +
Open a logged-out window and ask three things in each assistant: "what does [company name] do", "who supplies [your product] in Brazil", and one of your real buying questions. If it invents facts, it read little. If it stays silent, it read nothing. If it names three competitors and not you, the problem is comparative and not technical.
Is this worth investing in if my customers do not use AI to buy yet? +
The work that gets you cited is the same work that gets you found in search: readable pages that answer real buying questions. Even if not a single customer ever asks an assistant, those pages still rank and still convert. The AI part is what you gain on top, and it arrives late for whoever starts late.
Want to know exactly where you are invisible?
We run your twelve buying questions across Google and the main assistants, show you who gets named instead of you, and come back with the publishing order that closes the gap. Then you decide the scope and ask for a quote.
