GEO-041·Issue 04·Strategy & Market·41 / 132
Which Page First? Prioritize by Purchase Intent
Not every page deserves this quarter. Score page groups on question volume, margin, citation gap, and competition, then commit to three pages and re-test.
Issue focusAnswer Units: Writing Category and Product Pages for GEO
A DTC sleep brand with US and German storefronts opens Q3 planning with 312 pages: 240 product pages, 18 category pages, 40 blog posts, 14 comparison pages. Three content people, six weeks. The argument starts where it always starts: category pages first, because category keywords carry volume. The retention lead pushes back: last quarter’s rewritten blog series never surfaced in a single AI answer the team sampled. Nobody has asked the question that decides the quarter: which page is being quoted wrong today, and would the same twenty questions prove a fix?
GEO capacity is always smaller than the catalog, so the decision is which three pages this quarter gets — and the wrong order renovates pages nobody asks about.
Why search volume is the wrong sort key
Volume ranks demand, not your chance of being quoted: an AI answer assembles sources per question instead of allocating clicks by position.
Classic SEO sorts pages by volume because clicks follow rank. AI search cuts that chain in two. For one question the model pulls a sentence or two from a few sources; a category page can lose to a marketplace listing, a retailer, and two review sites, while a comparison page becomes the only independent storefront source.
Value per sentence is uneven: a material and use-case line gets lifted into “who is this for,” while a long explainer outside the question chain is rewritten ten times and never quoted. The constraint is verification, not writing: re-sampling is weekly work, so a quarter moves two or three page groups.
Four dimensions and one gate
Do not fold the dimensions into a weighted total: it lets “easy to fix” outrank “worth fixing” and puts the biggest-traffic page first.
| Dimension and question | Score 1 | Score 5 |
|---|---|---|
| Question volume: is this asked? | Brand terms only | Non-brand repeats |
| Margin and basket: what is a conversion worth? | Traffic driver | Hero or bundle |
| Competitive gap: is there room? | Retailers own it | No independent source |
| Citation gap: are you quoted correctly? | Absent | Misquoted |
Separate brand from non-brand questions: brand volume shows memory; GEO competes at comparison. Margin is not snobbery: low-margin pages wait; the same hours return faster on the hero. The competitive gap is inverted: fewer strong sources is a reason to enter. The smallest citation gap is cheapest: one paragraph closes it.
The gate sits outside the table: crawlable, renderable, quotable. If the no-JavaScript view says “enable JavaScript,” or specs live in images, fix that first; scoring a page AI cannot read picks from the wrong pool, as the AI crawler map shows.
From a ranking to a decision in four steps
A quick win is not “easy to edit” but “verifiable in four weeks by the same questions” — the question in the sample, your domain in the answer, a content-block edit, not a template rebuild.
| Step | What you do | Owner | Passes when |
|---|---|---|---|
| 1 Sample | Log 20–30 questions, the answer, its sources | Growth | A baseline, not an impression |
| 2 Score | Score groups; run the gate separately | Content, engineering | One judgment per group |
| 3 Edit | Change answer blocks on three pages | Content | One conclusion survives being lifted whole |
| 4 Re-test | Re-run the questions after four weeks | Growth | Something moved, or you can say why not |
Step 3 is where discipline breaks. A scoring table invites “while we are here” pages; past three, the re-test dissolves into a feeling that cannot decide next quarter. High-traffic, zero-citation, template-rebuild pages are quarter projects, not quick wins.
Example: four page groups, one quarter
Example — a hypothetical cross-border home-goods brand; scores are invented for method only, no real brand or figures.
| Page group | Volume | Margin | Gap | Citation | This quarter |
|---|---|---|---|---|---|
| Hero PDPs (12) | 4 | 5 | 3 | 4 | First; missing specs and use cases |
| Comparison (4) | 3 | 5 | 4 | 5 | First; questions exist, sources do not |
| Category (6) | 4 | 3 | 2 | 2 | Second wave; reuse PDP blocks |
| Blog (30) | 2 | 1 | 2 | 1 | Hold; keep the support-quoted one |
The comparison pages trail the category pages on volume and still ship together. Retailers and review sites hold the category slot, and three edits will not change that structure, while one evidence page may be the answer’s only independent brand-owned source. Edit thirty posts at once and a month later no citation change can be attributed to any single one. Scores exclude candidates; they never prove demand, as the market priority matrix also insists.
Re-rank on three triggers, not a calendar
Re-sort when a signal moves, not when the quarter ends: a fixed ritual becomes a fresh guess with old scores attached.
Three moments justify a new order. Citation structure changed: a new strong source appears, or the page you were cited on is gone. Margin structure changed: freight, duty, or returns costs move, and a group’s value moves with them. The gate lifted: a template fix ships, and previously blocked pages enter the pool.
Split capacity about seventy percent to two or three groups, thirty to catch-up, and log what changed and why. Re-rank every two weeks or you trade noise for a strategy.
Where this breaks
Five cases. Demand has not formed: a new category has no AI questions yet, so every group scores 1 — seed independent reviews instead. Sales happen in the marketplace: fix listings, feeds, and reviews before page order. Evidence-heavy categories: medical devices, children’s goods, and batteries hit certification first. No sampling baseline: without recorded answers, scoring becomes a room vote — start with an AI visibility baseline. Technical debt outstanding: JavaScript rendering, a stray robots block, prices in images.
The order owns one question: which page first. How a page should read is separate work, starting with the answer unit it has to carry. This is the strategy and market lane’s job, not a content brief.
If you take one action this month: pull twenty questions from support and site search, log the current answer and its sources, score three groups, and re-test three pages four weeks later. A quarter that edits thirty pages ends with “it seems a bit better” and nothing to re-rank.