The four signals that decide whether AI cites you

These signals are dependent, not additive. That single property is what makes the framework useful and what most GEO advice misses.

An engine has to fetch and parse your page before it can extract anything from it. It has to extract a clean claim before it can attribute that claim to anyone. It has to know who you are before corroboration means anything. Each stage is a gate on the next.

The practical consequence is uncomfortable: money spent on a later pillar while an earlier one is failing produces nothing at all. Not a smaller return, nothing. A site that renders its content only after JavaScript executes can commission the best writing in its category and remain uncited, because no part of that writing ever reaches the engine.

This is also why generic GEO scores are less useful than they look. A single number tells you how you are doing. It does not tell you which gate is shut.

The first pillar asks a yes or no question. Can an engine fetch this page and recover its meaning from what comes back.

Google states plainly that content must be indexable to appear in its AI features, and documents separate controls for preview behaviour and training use (Google Search Central, 2025). The failure modes are unglamorous: content that only exists after client-side rendering, crawlers blocked in robots.txt, canonical tags pointing somewhere else, pages that return the shell of a template and fill it in later.

Structured data belongs here too. Schema.org types such as Article, Organization and FAQPage let you state what a page is, who wrote it and when it was published, without an engine having to infer any of that from your layout (Schema.org, 2025). It is not a ranking lever, and marking up content that is not visible on the page violates Google's guidelines. What it does is remove ambiguity, which is exactly what an engine needs before it will attribute something to you.

Technical failure is the cheapest to fix and the most expensive to ignore.

What it looks like when it fails

Your domain does not appear in an engine's consulted sources at all, for any query, including ones about your own product.

The second pillar is the most commonly failed and the hardest to see, because a page failing it looks perfectly good to a human reader.

Generated answers are assembled from spans of text. A span that depends on the previous three paragraphs for its meaning cannot be lifted, so it will not be. This is a property of how the answer is built, not a judgement about your writing.

The fix is a rewrite of shape rather than substance. Put the direct claim in the first sentence under each heading, then qualify it. Phrase headings as the questions people actually type. Make each section survive being read on its own, because that is how it will be read.

Compare two openings. "There are several factors to consider when evaluating whether schema markup is worth implementing" gives an engine nothing to work with. "Schema markup does not raise rankings; it removes ambiguity about what a page is" gives it a complete claim it can quote and credit to you. Same information, different extractability.

What it looks like when it fails

Your domain appears in the list of sources an engine consulted, but never in the answer body itself.

The third pillar is about identity. Attribution needs a stable entity, not just a hostname.

An engine that is going to write "according to Acme" needs to be confident that the Acme on your homepage, the Acme in your structured data, the Acme in a directory listing and the Acme mentioned in a forum thread are the same organisation, doing the thing you say you do. When those descriptions conflict, the safe move is to quote somebody else.

Entity work is unglamorous maintenance: one organisation name, one description, one founder, one location, published consistently and corrected wherever an old positioning is still live. It is slow to move because it depends on other people updating their pages, and it is hard to fake for the same reason. That is precisely why it carries weight.

What it looks like when it fails

You are cited, but described inaccurately, or confused with a similarly named company.

The fourth pillar is corroboration. Models are tuned to avoid asserting things on the word of a single unfamiliar source, so a claim that appears only on your own site, in your own words, is one an engine will hedge on or drop.

What helps is being referenced where an engine already looks: industry roundups, publications, comparison pages, forum threads where practitioners discuss the problem. This overlaps with link building without being the same activity. A mention in a source an engine consults regularly can outperform several links from pages it never reads.

Trust is the slowest lever and the one least under your direct control, which is an argument for starting it early and never treating it as a launch task.

What it looks like when it fails

You are cited for uncontroversial claims but passed over for the ones that would actually win you a customer.

Diagnose before you buy anything. Ask, in order: can an engine fetch and parse these pages, can it lift a clean claim from them, does it know who we are, and does anyone else say so.

The first "no" you hit is your entire roadmap for the next month. Everything after it is premature.

The most common mistake is starting at pillar two because content is the work marketing teams know how to commission. Publishing more pages into a site that fails pillar one produces more pages that do not get cited, which reads as evidence that GEO does not work when it is really evidence that the order was wrong.