Flow · 6 steps · 5 known jams
A page you wrote, travelling from your own server to a sentence in somebody else's search result or assistant answer.
Finished when
Somebody who was not looking for you reads a claim from your page, either on the results page or inside an answer, and either arrives or does not.
Done when Returns 200, is in the sitemap, and its canonical points at itself.
Done when An automated reader has requested it at least once. Server logs say so; analytics never will.
Held up 16 discovered and never fetched
Where it stops
Search Console lists the page under “discovered — currently not indexed”, with no error beside it.
Not a fault in the page. The engine knows the address, has decided a request is not yet worth spending, and the decision is made on the site's standing rather than on anything in the markup.
Cost. The page cannot be indexed, ranked, or cited, and there is no technical remedy — which is exactly why this is the step most often billed for monthly.
Done when Stored and eligible to be returned. The engine's own report says indexed rather than the site: operator, which is an estimate it declines to stand behind.
Where it stops
Crawled — currently not indexed. One page on this site sits here.
An editorial judgement about the page itself: thin, duplicated by something the engine prefers, or answering a question it already has a better answer for.
Cost. Everything downstream is unavailable, and unlike the previous jam this one usually does have a remedy — the page has to be worth storing.
Done when Appears in a result set somebody was served. Position is not a condition; rank ninety counts.
Where it stops
Impressions rise and clicks do not move at all.
The appearances are happening deep — an average position of 35.9 across 192 queries is the fourth page — or inside a feature that answers the query on the results page.
Cost. Nothing is wrong and nothing is arriving. This is the state most sites are in and the one most often reported as progress.
Done when A declared automated client has fetched it, and the purpose of the fetch is recorded rather than reconstructed later.
Where it stops
The logs show training crawls and nothing else.
Training returns no citation, no link and no visit ever. Only an index build makes a citation possible later, and only a user-triggered fetch means somebody is waiting now.
Cost. The work is being consumed and cannot return anything. Worth knowing before renewing anything.
Done when Somebody clicked, or an answer named the source. The first is in your logs and the second is not.
Where it stops
Nobody can say whether the citation happened.
Citation occurs on the assistant's surface. There is no log to query, and the only way to establish it is to ask the questions and read the answers, which makes the prompt set the measurement.
Cost. The last step of the flow this practice sells is, for us and for everybody, the one with no first-party instrument. We say so rather than reporting fetches as citations.
The reasoning looks sound: more pages, more chances to be found. It fails at the second step, which is not sensitive to how much you published — a site fetching a fraction of what it has does not get fetched more by having more. The new pages join the same queue and the older ones lose their turn.
InsteadEstablish which step is actually binding before writing anything. If pages are being fetched and not indexed, the pages are the problem. If they are indexed and never shown, the demand is. If they are shown and never clicked, the position or the feature above it is. Four different problems, one symptom, and only one of them is solved by writing more.
Every process has a written form and an actual one, and the queue is always in the actual one. We map what happens, from the records rather than from the meeting.