Published . 3 min read.

How do you rebuild 300 pages so AI answer engines can cite them?

By , Group AI Marketing Executive at Archetype Group, Ho Chi Minh City

Short answer

Fix the structure before the pages: audit what exists, merge pages that compete with each other, and remove what has no reason to be indexed. Then write passages that make sense on their own, attribute claims, and state plainly what each page is and who stands behind it. Google says no special markup is needed; the gain comes from clear, indexable, specific pages.

Archetype Group had 300-plus web pages across 20 country markets, with no shared structure and several pages competing for the same terms. Search was moving from ten links toward generated answers in Google AI Overviews, ChatGPT and Perplexity, so rebuilding only for the old ranking target would have meant doing the work twice. This post describes what I did and which parts I think carry over. The case study is at Search & GEO rebuild.

Start from what Google actually says

Before any optimisation, it is worth reading what the search engine itself requires. Google's documentation says a page must be indexed and eligible to appear in Search with a snippet to be shown as a supporting link in AI features, and that there are no additional requirements or special optimizations for AI Overviews and AI Mode. It also says site owners do not need new machine-readable files or special schema for this. I treat that as the floor: ordinary technical health comes first, and anything marketed as an AI trick should be tested against it. Other answer engines do not publish equivalent guidance, so what follows for them is my working practice, not a documented rule.

Step 1: audit before touching anything

I crawled and inventoried the whole estate: what existed, what ranked, which pages cannibalised each other, and which had no reason to be indexed at all. A meaningful share of the estate was competing with itself. A twenty-country firm accumulates pages the way a building accumulates cabling: every office adds what it needs and nobody removes anything.

Step 2: fix the architecture, not the pages

Consolidation came first. Deciding that four pages should be one is worth more than optimising four pages, and no amount of on-page work rescues a structure that splits its own authority. Pages with no reason to be indexed were identified and dealt with. This is slow, unglamorous work, and I put it before any rewriting.

Step 3: write for extraction, not just for ranking

Answer engines quote passages, not pages. That changes the unit of writing. I wrote each page so that its key claims stand on their own:

  • a direct answer near the top that makes sense without the rest of the page
  • self-contained claims, with the subject named rather than implied by "it" or "this"
  • attribution: who says this, based on what
  • structured data that states what the page is and who stands behind it, consistent with what the visible page says

This is the part of the rebuild most projects skip, because it is editorial work and cannot be automated well. It is also the part a competing page cannot copy by adding a plugin.

Step 4: instrument it, honestly

I tracked positions and technical health in Semrush and RankMath, and what visitors did after landing in Google Analytics. One caution on the numbers: RankMath's score, which moved from about 20 to 50-plus across the pages, is an on-page checklist score from a WordPress plugin. It shows that pages meet a set of on-page criteria. It is not a ranking, and it is not a measure of being cited by an AI engine. I do not have a published citation-rate figure for this project, and I would be wary of anyone who claims a precise one: citation behaviour varies by engine, by question wording and from week to week.

Step 5: hand over the standard

Twenty offices publish content and I do not. The rebuild only holds if the person adding page 301 knows the rules, so the output included the written standard, not only the pages.

What carries over to other sites
PracticeEvidence it worksConfidence
Index health and unique, specific pagesStated in Google's own documentationHigh
Consolidating pages that competeMy project outcome: self-competition removedHigh for classic search, plausible for AI answers
Self-contained, attributed passagesMy working practice; no engine documents itMedium
Special files or AI-specific schemaGoogle says none are neededLow; do not rely on it

What I would tell a team starting this

Do the audit and the consolidation even if the AI part is the reason you started. They improve classic search too, so the work pays off whichever way the engines behave. Then add citable writing on the pages that matter most, and measure with a monthly check: ask the major engines your target questions and record which sources they name. That is manual and imperfect. It is also more honest than a dashboard that implies more precision than exists.