AI visibility · Diagnosis
Why AI doesn't recommend your products.
By Josh Redecop · Published
An AI assistant does not recommend your products for one of four reasons: it cannot read your product data because the numbers live inside PDFs or images; it cannot corroborate you because no independent source mentions you; it cannot resolve who you are because your details differ across the web; or it cannot tell a buyer where to get your product. Each is a separate failure, each is fixable, and the first step is finding out which one applies to you rather than guessing.
Four different problems that look identical from outside
From where you sit, all four look the same: you ask an assistant about your category and somebody else gets named. But they have different causes and different fixes, and work aimed at the wrong one produces nothing. Naming which applies to you is most of the job.
- 01
It cannot read you
Your dimensions, spans, ratings and finishes are published as PDF brochures or images of tables. An engine reading your site finds marketing copy where the numbers should be, so it has nothing of yours to quote and quotes somebody else. This is the most common blocker and the most straightforwardly fixable: the data already exists, it is just in the wrong format.
- 02
It cannot corroborate you
Your own site says you are excellent. So does every competitor’s. An assistant assembling a recommendation wants a source that compares options rather than a source that is one of them, which is why roundups, trade directories and industry coverage carry disproportionate weight. If nobody independent mentions you, there is nothing to weigh your claim against.
- 03
It cannot resolve who you are
Your company name, location, founding year and description differ between your site, your directory profiles and your social pages. Faced with records that disagree, an engine treats the entity as uncertain and reaches for a competitor whose details are consistent. Contradiction is worse than absence here, because it actively lowers confidence rather than merely failing to raise it.
- 04
It cannot tell anyone where to buy
Your product may be perfect for the application, but if the assistant cannot resolve a regional question about supply it will hesitate to name you. A recommendation a buyer cannot act on is a bad recommendation, and engines behave as though they know that. Dealer and distribution coverage is part of being recommendable, not a separate marketing problem.
Check it yourself in ten minutes
You do not need anyone's help to find out which problem you have. Run these five, in order, and write down what happens.
Ask ChatGPT, Gemini, Grok and Claude the question your buyer would actually ask — not your company name. Something like "what should I use for this application, and who supplies it in this region." Record who gets named.
Ask each one what it knows about your company by name. If it invents details, contradicts itself, or confuses you with another business, your entity is the blocker rather than your content.
Open your most important product page and try to copy a specification number out of it as text. If you cannot select it, an engine probably cannot read it either.
Search the web for your company name and see what a third party says about you. If the only results are your own properties, you have a corroboration problem.
Check that your name, location and founding year are identical on your site, your directory listings and your social profiles. Any disagreement is a confidence penalty.
If the assistants describe you accurately but never volunteer you, the problem is corroboration. If they get your details wrong, it is your entity. If they cannot quote a number from your own pages, it is readability. Those are three different pieces of work.
Questions this raises
Is this the same as SEO?
It overlaps but is not the same. Search engine optimization competes for a position in a list of links, and the buyer chooses from that list. An AI assistant returns a short answer naming a couple of options, and everything else is absent rather than ranked lower. The pages that get quoted in those answers are frequently not the pages that rank first, so winning one does not guarantee the other.
Can we just add our products to a prompt or pay for placement?
No. There is no legitimate way to pay a model to recommend you, and anyone selling that is either describing sponsored placement on a site the engine happens to read, or making a claim they cannot support. What actually moves an answer is being readable, corroborated and resolvable — the same qualities that make a buyer trust you, which is not an accident.
How do we know which of the four reasons applies to us?
Run the self-check above; it separates them reasonably well. If the engines name competitors but describe you correctly when asked, you have a content problem. If they get your details wrong or confuse you with someone else, you have an entity problem. If they know you but never volunteer you, you have a corroboration problem.
Our products are sold through distributors, not direct. Does this still apply?
It applies more. A buyer asking an assistant what to use and where to get it is asking a question your distribution network exists to answer. If the assistant cannot connect your product to a place that stocks it, it tends to recommend something it can source — which means your channel strength becomes invisible at the exact moment it should be decisive.
How long does fixing it take?
It depends which of the four applies and where you start. Readability and entity problems are the fastest, because they are fixed on properties you control. Corroboration is slower, because it depends on other people publishing about you. Engines that retrieve live from the web reflect changes sooner than engines leaning on training data, so improvement tends to appear unevenly rather than all at once.
Which assistants should we be checking?
ChatGPT, Gemini, Grok and Claude, checked separately rather than as a group. They read substantially different sources, so being named by one says very little about the others. Checking only the most popular one is the most common mistake, and it produces a picture that is reassuring and wrong.
If you make building products and the check above points at something you would rather not fix in-house, the work itself is described at GEO for building-product manufacturers. If you sell software, see GEO for SaaS companies.
Thirty minutes. We run your category live on the call.
No pitch deck. We ask the engines what your buyers ask, in front of you, and show you which competitors get named. You keep the findings either way.