The four gates
Every AI answer about your business passes four checkpoints. Miss one and the others stop mattering.
synthesis · April 7, 2026 · 4 min read
This is the framework everything else on this site is built on, so it is worth setting out properly rather than as four bullets on a landing page.
The observation behind it is simple: the reasons a business fails to appear in AI answers fall into four groups, they are strictly sequential, and almost every business we look at fails at least one without knowing.
Sequential is the important word. A failure at gate one makes gates two, three and four irrelevant — you cannot be understood if you cannot be read. So the order is also the order of work.
GATE ONE — Can it get in?
Before anything else, a machine has to be able to fetch your page and find words in it. Three things stop that, and all three are invisible from a browser.
The instruction file. One file on your server tells every crawler what it may and may not read. It has usually been there for years, written by someone who has moved on, and it frequently blocks things nobody meant to block. There is a whole article on this — the distinction between the crawler that trains a model and the crawler that fetches pages to answer a question is where most published advice is wrong.
The security layer. The service protecting your site from attacks often cannot distinguish an AI assistant from a scraper and blocks both. This is the most common invisible failure we find, and it is not in your robots.txt at all. It is in a dashboard your web host controls.
Rendering. If your text only appears after JavaScript runs in a browser, then to any crawler that does not execute JavaScript, your page is blank. We fetch every site twice — once as a raw HTTP request and once in a headless browser — and the difference between the two is itself the finding.
GATE TWO — Can it understand you?
A page can be perfectly clear to a person and useless to a model. The difference comes down to whether a paragraph can be lifted out and dropped into an answer without breaking.
The evidence here is unusually good. Across 15.7 million AI Mode citations, the passages that get reused are a median of 117 words, state their answer in the first sentence about 80% of the time, and are self-contained about 85% of the time. And separately, across 252,000 controlled trials, stating concrete facts — prices above all — turned out to be a gatekeeper while formatting sat in the weakest tier.
So gate two is not about structure in the way the industry means it. It is about being quotable: answer first, paragraphs that stand alone, real numbers instead of “call for a quote”.
GATE THREE — Can it find you where it actually looks?
This is the gate most people don’t know exists, and for a local business it is often the largest.
Most of what a model says about you does not come from your website. In our own city study, guides and media accounted for 48% of citations against 27.8% for the businesses’ own sites. Meanwhile, for anything local, the factors that decide whether you appear at all are led by the verified address and its proximity to the searcher, the primary category on your business profile, and the volume and recency of your reviews — none of which live on your site.
So gate three is the profile, the listings, the aggregators in your trade, the “best in town” articles, and being named in ordinary prose without a link.
GATE FOUR — Can you see any of it?
Google only broke this out as its own report in June 2026. Before that it was buried inside your ordinary search numbers with no way to separate it. Most business owners have still never opened it.
And there is a second, harder half to this gate: knowing what cannot be measured. Ask a model the same question twice and you frequently get a different list of companies — one study of 2,961 runs found under a one-in-a-hundred chance of two identical brand lists. What can be measured is a rate across repeated asks, tracked over months. What cannot be measured is a position, on a Tuesday, on a dashboard.
Why sequential matters commercially. It is also the reason our first tier exists. Gate one and most of gate three are fast — days, not weeks — and they are where the largest failures usually are. Rewriting a site’s content before checking whether anything can read it is the most expensive way to make no difference at all.
