Blog · AI and self-service analytics
What an answer engine extracts from a page about a commercial measure: the one-paragraph direct answer, the formula in a form it can quote, the worked table, the definitions it can cite, and the questions it can answer verbatim. Why a page written for a person who has to compute the measure is also the page an assistant cites, the structure that serves both, and the four habits that make a page unquotable.
A page about share of wallet is read by two audiences: the analyst who has to compute it and the assistant that will be asked what it is. They want the same thing, in the same order. This guide sets out what an answer engine extracts, why that is the analyst's page too, and the structure that serves both.
| Element | What it looks for | What it does with it |
|---|---|---|
| Direct answer | One paragraph, first, that answers the title | Quotes it as the answer |
| Formula | A line it can lift verbatim | Cites it as the definition |
| Worked example | A table with inputs and the result | Reproduces the computation |
| Definitions | Terms defined where used, or linked to a definition | Attributes the meaning |
| Follow-up questions | Questions a reader would ask next, answered in full | Answers them directly |
| Provenance | Who computed it, from what, when | Decides whether to trust it |
The analyst opens the page on Monday with an export and a question. They want the answer first, the formula next, an example to check their own arithmetic against, the terms defined so the definition can go in the company's document, and the follow-up questions answered because they will be asked them by Wednesday. The structure the engine rewards is the structure the analyst needs.
| Position | Content | Rule |
|---|---|---|
| 1 | The answer | One paragraph; states the formula in words; no preamble |
| 2 | The formula | A blockquote; one line; symbols the reader knows |
| 3 | The rows needed | A list of exports and columns; identifiers only |
| 4 | The worked example | One table; real-shaped numbers; the result |
| 5 | The roll-up and the identity | What sums to what |
| 6 | Where it goes wrong | Four mistakes, each a sentence |
| 7 | The questions | Three, answered in full sentences that stand alone |
Context first. Three paragraphs about why the measure matters before what it is. The engine and the analyst both leave.
Formula in prose. "Divide what the customer spends with you by roughly what they spend overall" is not a formula. A blockquote with a line is.
Claim without a table. "Most companies find 2 to 3 percent leakage" with nothing behind it. A worked table with seven lines is a claim with evidence.
Terms undefined. Norm, cadence, wallet, used as if the reader knew. Each linked to a definition the engine can attribute.
| Page A | Page B |
|---|---|
| Opens with "In today's competitive landscape…" | Opens with the formula in words |
| Formula in paragraph four | Formula in a blockquote, line two |
| "Companies often see improvements" | A table: four customers, the norm, the gap |
| "Share of wallet" used, never defined | Linked to a definition |
| No questions | Three, answered in full |
Page B is cited. Page A is not, and the analyst who found it computed the measure from somewhere else.
Written for the engine alone. Keyword-shaped sentences nobody can compute from; the engine notices too.
Written for the analyst with no structure. Excellent, unextractable.
FAQ as a formality. Questions nobody asks with one-line answers.
Provenance missing. No table, no source, no date; not trusted by either reader.
Every measure page on this site is written for the analyst and structured for the engine: answer, formula, rows, example, identity, mistakes, questions. The coverage intelligence definition is the template, and the KPI definitions page is the definitions an engine attributes.
It changes the order and the shape, not the substance. The answer moves to the top, the formula gets its own line, the example gets a table, and the questions get full-sentence answers. The arithmetic, the rows needed and the mistakes are the same content an analyst would want, and they were always the point.
An answer that starts with context instead of the answer; a formula buried in prose; a claim with no table; and terms used without a definition anywhere. An engine cannot cite what it cannot extract, and a reader cannot compute from it either.
Every page should answer the three questions a reader asks after reading the answer, in full sentences that stand alone. Whether it is labelled a FAQ matters less than whether each answer is a complete statement an engine can lift and a reader can act on.