“Impressions are vanity; DMs are signal.” — from the Suede launch campaign
Every founder understands compounding when it works for them. Content compounds. Reputation compounds. Distribution compounds. This chapter is about the version nobody budgets for: absence compounds too.
Being missing from AI answers is not a static condition, like a billboard you have not bought yet. It is a position that deteriorates while you wait, through three loops that feed each other. None of them requires the engines to be malicious or even particularly smart. They just require time.
Loop one: the buyer’s default hardens
Start with what an AI answer actually does to the person reading it. It does not just inform them. It frames the category. When a thousand buyers ask about your space this quarter and the same two competitors appear in most of the answers, those companies are not merely getting leads you are not getting. They are being installed as the reference points, the names against which everything else in the category gets compared.
Buyers repeat what machines tell them. The founder who asked ChatGPT mentions the shortlist in Slack. The consultant pastes it into a client deck. The junior analyst’s market summary is one-third synthesized answer. Each repetition is invisible to you and cumulative for them. By the time you finally meet this buyer, months from now, you are not a candidate. You are an unfamiliar name being measured against their defaults, and the defaults were written by a machine you never audited.
Sales teams have a phrase for the vendor who arrives after the frame is set: column fodder. The new mechanism producing that old outcome is that the frame now gets set in four seconds, before you have ever heard of the deal.
Loop two: the evidence trail thickens on one side
Chapter 4 established that engines lean on third-party evidence: reviews, comparisons, discussions, reference pages. Now watch what happens to that evidence over time when one company is in the answers and another is not.
The cited company gets the buyers, which means it gets the users, which means it gets the reviews, the community threads, the “we switched to X” posts, the comparison articles that include it by default. Every one of those artifacts is a new page that tomorrow’s retrieval can find and cite. The absent company generates none of this, not because its product is worse but because the flywheel that produces public evidence starts with buyers, and the buyers are being routed elsewhere.
This is the loop that should genuinely worry you, because it operates on the inputs rather than the outputs. A stale answer can flip tomorrow. A category’s entire evidence trail, tilted three years toward your competitor, cannot. The longer you wait, the more the public record of your category is written by people who have never used your product, describing a market that does not include you.
Loop three: today’s answers are tomorrow’s training data
The third loop is slower and quieter. The text of the public web is what these systems learn from. The answers being generated today, and the articles, posts, and summaries humans write downstream of those answers, become part of the corpus future systems train on. A category narrative that hardens in this era gets inherited by the next one.
I will not overclaim here, because this loop is the least measurable of the three and this book does not deal in unverifiable numbers. Nobody outside the labs can tell you precisely how much today’s answer shapes next year’s model. But the direction is not in serious doubt: these systems learn from the written record, the written record is being written now, and you are either in it or you are not. Absence, left alone, archives itself.
Why waiting feels safe and is not
Put the three loops together and the cost of waiting stops being abstract. A quarter of delay is not a quarter of missed leads. It is a quarter of buyers trained on a shortlist without you, a quarter of evidence accruing to the companies on it, and a quarter of the public record hardening in a shape you will later have to argue with.
Yet waiting feels safe, and it is worth naming exactly why. Nothing hurts. The loops run in the silence Chapter 3 described: no alert fires, no chart dips, no deal is visibly lost. The line at the top of this chapter is from my own launch notes, where it meant something narrow about marketing metrics: impressions are vanity, DMs are signal. It generalizes into the operating principle for this whole problem. The numbers that feel good and arrive automatically are vanity. Signal is what a real buyer, or a real buying machine, actually does. And the signal here, the answers themselves, never arrives on its own. You have to go collect it, which by now you have done three times.
Be suspicious, too, of the comfort on the other side. If you ran the exercises and found yourself in the answers, the loops are running for you, which is leverage, not safety. Point-in-time, as always: answers move between days and sessions. An incumbent who stops tending the inputs is exactly the kind of incumbent the decoupling in Chapter 4 exists to punish.
The turn
Here is where Part I lands. The buyers migrated to a layer you were not watching. That layer writes shortlists, binary and silent. It selects for citation logic, not ranking logic. And its verdicts, left alone, compound.
Everything in that paragraph is outside your control except one thing: the inputs the machines encounter when they come looking. Whether they can read your site. What your pages give them to quote. What evidence exists that you are real. Those are concrete, inspectable, fixable properties, and fixing them is not a dark art. It is a checklist, and it is the entire second half of this book.
The stakes were the hard part. The fix is work. Let’s work.
Check this yourself right now
One last stakes exercise, and it is the one that reliably ends the debate inside a company. Ask ChatGPT:
“Compare [your company] and [the competitor you most often lose to] for [your ideal customer]. Which would you recommend and why?”
Read what the machine believes the comparison is. Founders regularly discover the engine misstates their pricing, their features, or their category, or politely declines to say much about them at all while speaking fluently about the competitor. Screenshot it, date it, add it to the panel. Then turn the page, because everything from here forward is repair.