“The paid unit of value is the screenshot of ChatGPT, Perplexity, or Gemini recommending your competitor instead of you.” — from the Suede product spec
There is a message I have sent to founders more times than I can count. It goes like this:
“Hey, I ran ‘best tool in your category for your exact customer’ through ChatGPT and Perplexity this morning. Two of your competitors are cited in both answers. Your product doesn’t appear in either. Screenshot attached.”
Nobody argues with the screenshot.
Founders argue with analytics dashboards. They argue with SEO reports full of scores and acronyms. They argue with consultants who talk about domain authority. Nobody argues with a picture of ChatGPT, asked a question their actual buyer actually asks, recommending someone else by name.
The screenshot works because it is not an opinion. It is not a projection, a trend line, or a maturity model. It is the exact answer a real buyer received when they asked a machine for a recommendation in your category. If your name is not in that answer, then for that buyer, in that moment, you did not exist. There was no second page to scroll to. There was no listing to skim past. There was an answer, the answer had names in it, and yours was not one of them.
The morning this became a business
I did not set out to write a book about AI visibility. I set out to figure out why smart founders with good products were quietly losing deals they never knew existed.
The pattern kept repeating. A founder builds something genuinely good. They do the responsible things: a clean site, some content, maybe an SEO retainer. Their traffic is fine. Their rankings are fine. And meanwhile, a growing share of their buyers have stopped searching in the way any of those numbers measure. Those buyers open ChatGPT, or Perplexity, or Google with AI Overviews turned on, and they ask a question. The machine answers with three to five names. The buyer evaluates those names. The deal happens inside that shortlist.
If you are not in the answer, you are not losing the deal. Losing implies you competed. You were never in the room.
So I started running the check for people. Ask the engines the questions their buyers ask, several different ways. Capture every prompt and every answer with timestamps. Count who gets named. Put the screenshots in front of the founder. Then, for the ones who wanted it, ship the actual repairs the same day: the structured data, the machine-readable files, the rewritten page sections that give an engine something worth citing.
That service became Suede’s paid audit work. This book exists because the diagnosis half of that work should not be a secret. You can run it yourself, today, for free, and Part II shows you exactly how. What I sell is speed and execution. What I am giving you here is the sight.
Why a screenshot and not a report
I want to be precise about why the screenshot is the unit of value, because the reasoning is the spine of this whole book.
An SEO report tells you about a system you already understand: search results, rankings, clicks. You have twenty years of intuition for what page one means. You have no intuition yet for what it means that an answer engine synthesizes one response and your company is structurally absent from it. The screenshot builds that intuition in five seconds. It converts an abstract risk into a specific, dated, undeniable event: this machine, asked this question, on this day, named these companies and not yours.
The second reason is harder to hear. Reports let you postpone. A score of 61 out of 100 feels like a project for next quarter. A screenshot of your top competitor being recommended to your exact customer, this morning, feels like a fire. It should. Chapter 5 is about what the delay actually costs, and the short version is that these answers harden over time. The competitor who owns the answer today is training thousands of buyers, and arguably the engines themselves, to treat them as the default.
One caveat, and I will repeat it throughout because the honest version of this field requires it: every screenshot is a point-in-time observation. Ask the same engine tomorrow and the answer may differ. That cuts both ways. It means a good answer today is not a permanent asset, and a bad answer today is not a permanent verdict. What persists are the inputs, and the inputs are what you control.
What this book will and will not do
By the end of Part I you will understand why this happened: where the buyers went, how answer engines choose who to name, and why absence compounds. By the end of Part II you will have run a real audit of your own AI visibility: whether the crawlers can read your site, how each major engine selects sources, whether your pages are extractable, whether your evidence is citable, and a checklist you can rerun monthly.
What this book will not do is promise you citations. Nobody can. The engines do not publish their selection logic, they change constantly, and anyone who guarantees you a spot in an AI answer is charging you for weather. What I can promise is narrower and more useful: you will know exactly what the machines say about your category right now, you will know which inputs are broken, and you will know how to fix every one of them.
That is the whole trade. Evidence, shipped inputs, and measured deltas. It is enough.
Check this yourself right now
Open ChatGPT and paste this, filled in for your business:
“What is the best [your category] for [your ideal customer]? Recommend 3 to 5 options and briefly explain each.”
Read the answer once as a founder. Then read it again as a buyer who has never heard of you. Screenshot it, note the date, and keep it. That screenshot is your baseline, and by Chapter 11 you will be measuring against it.
If your name is in the answer, good. You have something to protect, and most of Part II applies to you with equal force. If it is not, you now know precisely what this book is for.