Updated August 2026

The Citabld Method

The method exists because three measured facts broke the old playbook. First: engines rewrite. Retrieval queries share as little as 13% of their words with the prompt, and change run to run.[src] Second: retrieval is stage-gated. Across the buyer journey, retrieval and citation rates sit near zero at awareness and jump to roughly 48% at decision moments, where the same brands then appear with 100% consistency.[src] Content cannot influence retrieval that is not happening; upstream, the work is shaping what models already believe. Third: citations live off-site more than brands expect. Multiple studies find the majority of AI citations point to third-party sources for comparison-type queries, while owned pages dominate local and branded ones.[src] Placement is a decision, not a default.

The four phases

What are the four phases of the method?

Map, Library, Score, Produce. Map and Library and Score make up Run 1, the architecture you approve before any content exists. Produce is Run 2, where we write the content ourselves at full depth. Each phase has its own page with the tables, rubrics, and arithmetic it runs on.

Run 1 and Run 2

Architecture first. Nothing gets written until you approve the build sheet.

Run 1 produces the map, the library, and the scored priorities. Run 2 is production against what you signed off on, not against a hunch.

What is a head query?

A head query is the question a buyer actually types or says to an AI assistant: "best warehouse scheduling software," "is X worth it," "how do I fix Y." Head queries are sourced from real buyer language such as sales calls, support tickets, and communities, because keyword-tool phrasing produces keyword-shaped fan-outs that miss how people talk to assistants.

best warehouse scheduling softwareis X worth it for a 40 person teamhow do I fix Y without downtime

What is a subquery library?

A subquery library is the structured database of fragments an AI engine may retrieve against for a brand's head queries: 500+ rows spanning nine engines and every buyer persona, clustered into themes, with each row carrying its provenance label and each theme carrying coverage, winnability, and priority scores. The library is simultaneously the content plan and the tracked-prompt inventory for measurement.

subqueryenginepersonaprovenance
warehouse scheduling software pricing per siteperplexityops leadMEASURED
soc 2 compliant scheduling vendorschatgptcomplianceMEASURED
claude brave organic top results schedulingclaudeallDETERMINISTIC
best shift planning tool mid marketgeminibudget ledMODELED

Why personas change the fragments

Engines condition retrieval on who is asking. Memory systems feed the query rewriter directly; the same question from a budget-led buyer and a compliance-led buyer produces different fragments. The measured effect concentrates in the middle of the market: category leaders stay recommended across personas, while mid-market brands can see most of the recommendation set swap as the persona changes.[src] If you are not the category leader, persona coverage is not optional. It is where your citations live.

The engagement

How is the work structured?

run 1

Architecture

Entity resolution, citation landscape, personas, head queries with written verdicts on every query you supply, the full library, scored priorities. You review before anything is produced.

run 2

Production

Content at full expansion against the approved priorities: pages, blocks, posts, placements, feeds. Each unit through extraction QA and your voice standard before it ships.

When Citabld is the wrong buy

If you are the runaway category leader, engines already recommend you to everyone; spend elsewhere. If your category is answered almost entirely by institutions such as government or medical bodies, the play is entity accuracy on those surfaces, not content volume. We will tell you that in the mapping call and it is a short engagement. And if you want a vendor who promises exact fan-outs and guaranteed rankings, that vendor exists and it is not us, because that promise is not honest on unobservable engines.
Where the method runs

Read the four phases in full, or watch the method run on one of your queries.

Definition

Extraction engineering is the practice of structuring content so an AI engine can lift a passage and have it survive out of context: answer-first ordering, entity self-containment, one topic per block, dated claims. Within the Citabld method, the unit of retrieval is the passage, not the page.