R2 · Retrieved, does your site enter the consideration set for the demand that matters?

A buyer asks an assistant about the thing you sell. The assistant does not go looking for that question. It breaks it into narrower ones nobody typed, fetches candidate sources for each, and writes its answer out of whatever came back. Your best-ranked page can sit out that whole process. Nothing on your side looks wrong.

R2 · Retrieved is the second of the Quattr Method's five questions: do you enter the consideration set for the demand that matters? There are two consideration sets now. One is the ranked results page, where demand still finds you on Google. The other is the source set an assistant assembles before it writes a word.

A site can be healthy on the first and absent from the second. Most teams measure only the first. The two drift a little further apart every quarter.

Retrieval is not ranking

This is the shift that defines the era, and it is mechanical rather than philosophical.

When someone asks an assistant a question, the system expands it into many retrieval queries behind the scenes, pulls candidate sources for each, and composes an answer from what it fetched. You never see that expansion. So you are competing for sub-questions the model asked on the buyer's behalf. No rank tracker holds those.

A page that ranks well for the typed query and never gets fetched into the source set has entered one consideration set and missed the other. Nothing in your reporting says so.

The unit is smaller than a page, too. Assistants pull passages, judge them out of context, and keep the ones that answer cleanly. So a page can be excellent read end to end and still lose every sub-question underneath it, because the assistant never reads it end to end: it lifts one section out and judges that section alone.

What the Retrieved number measures

CTR-modeled search market share of tracked demand. Every word in that phrase is load-bearing.

It is your estimated share of the clicks available across the keywords you track, against the competitors you track. Modeled from rank and your own click-through curve, so it points a direction rather than settling an absolute. It is search share, never revenue share. And it is a share of tracked demand, not of some vendor's whole keyword universe.

Read it the way the method reads it. Who leads, how big the gap is, which segments carry the movement. Non-branded demand is the lens that matters for opportunity, because demand that already typed your name was never up for grabs.

That directional label is not a hedge to skim past. A modeled share moves for reasons that include the model's own inputs, so the method reads it for direction and distance instead of a second decimal place. That is what keeps it a number you can decide on.

See also: what the share number covers and what it does not →

The workflow that does this: Non-branded share by intent →

Tracked demand means your demand

The denominator is first-party: the queries your own site already earns impressions for, organized by an intent model written in your business language.

Not a hand-picked keyword list, which drifts toward whatever the team already believes. Not a borrowed category list either. That one measures somebody else's market.

Tracked also means organized. The intent model groups that demand so share reads per intent instead of per keyword soup. When the number surprises you, the first question is which intents moved, and the model answers that in one step.

The same recipe builds the prompt baskets one question up at R3 · Referenced, the question of whether you are the source answers use. A prompt basket is the set of questions replayed in AI engines to measure you, generated from what your buyers search for. Whichever demand you measure against decides whether the number can be trusted.

Why does most of the work aim here?

Four kinds of work name Retrieved as their primary target: demand modeling, refresh and content quality, internal linking, and net-new content. That concentration is no accident. Entering the consideration set is where most of the method's work lives.

Sequencing is fastest-proof-first: run what you can prove in weeks before what takes quarters, so the slower work earns its budget on evidence. That is why the two fastest levers, internal linking and refresh, start here. Their proof shows up on this question soonest.

When your share number sags, the reading order is fixed. Separate the market's motion from your own first, because a growing category can float every boat while your share quietly thins. Then work down toward R1 · Reachable, the question of whether search engines and AI bots can fetch, render, and index your site, to rule out the physical gates.

The market-share-review skill runs that first comparison as one conversation, a packaged analysis you run by name in your assistant. The article on telling the market's motion from your own carries the decomposition in full.

See also: how the method sequences the work →