My book came out this week. It’s called Get Recommended by AI, and it covers exactly what the title promises: how to become the business that ChatGPT, Gemini and Perplexity name when someone asks who to buy from.
So I did the obvious thing. I went to Amazon and typed “how to get recommended by AI“.
Amazon’s assistant answered the question. It gave a reasonable summary of what a business should do to appear in AI answers. It did not mention my book.
I know why it happened, and it’s the same reason your business is probably missing from AI answers about your category.
Why AI won’t recommend a brand new product
Recommendation runs on accumulated evidence, and a new listing has none. No sales history, no reviews, no coverage anywhere else. The page can be perfect and still have nothing outside it to back up what it says.
Key Takeaways
- Why AI won’t recommend a brand new product
- What happens when you run the framework on your own failure?
- Foundation: can AI reach and read it?
- Extraction: is there a clear answer to lift?
Two things were happening in my case. Neither is Amazon being unfair.
The first is that I asked a question rather than for a product. “How to get recommended by AI” is shaped like a request for information, so an assistant answers it. Retrieving a product is a different job, triggered by a different kind of query. A system that responded to every question by selling you something would be useless.
The second is that my listing was hours old. Recommendation systems run on accumulated evidence, and I had not accumulated any.
The page was not broken. Every field was filled in, the description was clear, the metadata was complete. The page was fine. There was simply nothing outside the page to corroborate it.
What happens when you run the framework on your own failure?
You find out which of the three layers actually broke. In my case the first two passed and the third did not, which is the layer you cannot fix by yourself.
The book breaks the problem into three layers, in the order they have to be fixed. It felt only fair to point it at myself.
Foundation: can AI reach and read it?
Yes. The listing exists, it is indexed, the metadata is complete and machine readable. This layer passed.
Extraction: is there a clear answer to lift?
Yes. The title, subtitle and description state plainly what the book is, who it is for and what it covers. Anything summarising that page would get it right. This layer passed.
Reinforcement: can it be verified by anyone other than me?
No. No reviews, no rank, no third-party mention. Every claim about the book, at that moment, came from the person selling the book.
Two out of three. And the third is the one that matters, because the first two are things you assert about yourself and the third is what everybody else says about you.
How common is it to be missing from AI answers?
It’s the normal condition, not the exception. For the State of AI Visibility Index we analysed more than 1,000,000 websites against 250+ signals each. 216 were accessible, understandable and verifiable enough for AI systems to use with confidence. That’s about 1 in 4,600.
The highest score recorded anywhere is 86.6 out of 100, so the standard every business is being judged against is one that no site has met yet.
Most of those failures sit at the Foundation layer. The site cannot be reached properly, or cannot be read cleanly, and no amount of reputation rescues it. You can be the best firm in your city and be invisible because a crawler cannot parse your pages.
My case was the other failure, and it’s the one nobody plans for. Foundation perfect. Extraction perfect. Nothing outside my own site saying I exist.
From the outside, these two problems look identical. You ask the question, your name is not in the answer, and nothing tells you which layer broke. That’s why the order matters and why guessing is expensive. Reinforcement cannot rescue a site that AI cannot read, and a technically perfect site with no external evidence still does not get named.
What can you actually do about it?
Fix the layers in order, and accept that the third one is built by other people rather than by you. There is no trick, and I would distrust anyone selling one.
In my case the honest levers are the slow ones. Sell copies, so there is behaviour to observe. Earn reviews, so there is evidence from people who are not me. Get written about, so there is something to cite. That’s Reinforcement being built the only way it can be built.
I will check again in thirty days and publish what changed, including if the answer is nothing.
What I am not going to claim
I cannot see inside Amazon’s assistant. Nobody outside Amazon can.
What I have written here is inference from how retrieval-based systems generally work, applied to a case where I happen to know every detail of the input. It is not a measurement of Amazon’s ranking and I am not going to dress it up as one. Anyone who tells you they know exactly how a specific assistant ranks is overclaiming what the evidence shows.
The difference between “here is what we measured” and “here is what we reckon” is the difference between a methodology and a horoscope, and the AI visibility industry currently produces a great deal of the second. Ours is published in full, so you can argue with it.
The number that made me write the book
216 out of more than a million. Not one site has reached the top band. The best score anyone has recorded is 86.6.
Being absent from the answer is the normal condition. Almost every business is in it, including, this week, the one selling a book about how to get out of it.
The difference is that it is fixable, the work is specific and mostly mechanical, and almost nobody has done it yet.
Get Recommended by AI is out now in Kindle and hardcover. Every copy includes the full toolkit and twelve months of AI visibility tracking. You can also score your own site free, no email required.
Frequently asked questions
Why does ChatGPT not mention my business?
Usually one of three reasons: AI cannot reach or read your site, it cannot lift a clear answer from your content, or nothing outside your site verifies what you claim. They look identical from the outside, so the only way to know which applies is to test each layer in order.
Does having a good website mean AI will recommend me?
No. A technically perfect site with no external evidence still does not get named. Of more than 1,000,000 websites analysed, 216 met the full benchmark, which is about 1 in 4,600.
How long does it take to become visible in AI answers?
The Foundation and Extraction work is mostly mechanical and can be done in weeks. Reinforcement is slower, because it depends on other people citing and reviewing you, and that cannot be manufactured honestly.
Is AI visibility the same as SEO?
They overlap but they are not the same. Ranking on Google means appearing in a list of links. Being recommended by AI means being the answer, which requires your content to be extractable and your business to be verifiable, not merely ranked.
I wrote a book about being recommended by AI. On launch day, Amazon’s AI would not recommend it.
