SourceCited / Phase 1 · On-page / AEO
1 Phase 1
AEO is about being the single extracted answer. That comes from three things: a front-loaded self-contained answer, real information gain, and extraction-friendly structure.
Do three things to every important page: front-load a self-contained answer with a number in the first 200 words; add information gain (original data, testing, a real statistic, an expert quote) because rehashing consensus gets demoted; and restructure into question headers, self-contained chunks, and comparison tables that engines lift cleanly.
Put a complete, self-contained answer in the first ~200 words, carrying a statistic or a named entity — because answer engines extract the passage that resolves the query on its own, not the one that promises to explain later. Cut the throat-clearing intro. The opening should make sense if it were the only thing quoted.
Rewrite the first 150–200 words of the page I'm pasting so it works as a standalone answer an AI engine could quote directly. Requirements: - Sentence one directly answers the page's core question, with a number or a named entity in it. - No "in this article we'll explore…" preamble; no throat-clearing. - Self-contained: it must make sense if it were the only paragraph shown. - Keep my facts; don't invent statistics. If a claim needs a source, mark it [needs cite]. Return the rewrite, then a one-line note on what you changed and why.
Pages that merely restate the consensus get displaced; pages that add information gain — original data, first-hand testing, a specific number, an expert quote — get cited. The founding Princeton GEO study measured the winning tactics directly: adding statistics, quotations, and cited sources lifted visibility by up to 40%, while keyword stuffing hurt. Write claims that can be lifted whole — cited sentences skew short, with a median around 10 words — and treat information gain as defense too: the March 2026 core update demoted pages that summarize without adding anything, and the 2026 spam updates specifically target scaled AI content.
Review the page I'm pasting for fact density and information gain. Do three things: 1. Mark every sentence that is generic consensus a reader could get from any competitor — these are candidates to cut or upgrade. 2. Suggest, per section, one concrete addition that would add information gain: an original data point I could gather, a specific statistic to source, a first-hand test to run, or an expert to quote. 3. Where I give it real data or numbers, weave them in with inline attribution. Do NOT fabricate statistics — propose what to add and where, and I'll supply the real figures.
Structure so each chunk answers one question on its own. Use question-format H2/H3 headings that mirror how people ask; keep sections self-contained (roughly 75–225 words); and add comparison tables and short Q&A blocks, which engines extract cleanly.
| Instead of | Use | Why |
|---|---|---|
| "Overview" | "What is X, in one sentence?" | Matches the query; extractable heading |
| A wall of prose comparing options | A comparison table | Engines lift tables into answers cleanly |
| "Benefits" | "When should you use X over Y?" | Answers a real decision, not a label |
| Buried conclusion | Answer in the section's first sentence | Inverted pyramid = quotable passage |
Restructure the page I'm pasting for answer extraction, keeping all my facts: - Convert section headings into natural question form that mirrors how people search. - Make each section self-contained (~75–225 words) with its answer in the first sentence. - Identify one place a comparison table would replace prose, and draft that table. - Add a short FAQ (4–6 Q&As) using real questions, formatted for FAQPage schema. Return the restructured outline first, then the rewritten sections.
Front-load a self-contained answer to the exact question with a number or named entity in the first ~200 words, add information gain (original data, a real statistic, a test, an expert quote), and restructure into question-format headings, short self-contained sections, and comparison tables that engines extract cleanly. Then earn off-site corroboration.
AEO (Answer Engine Optimization) is structuring a page to be the single extracted answer — the descendant of the featured snippet. GEO is the broader goal of being cited across generative engines. In practice they overlap heavily; AEO is the on-page half.
Roughly 75–225 words — long enough to fully answer its heading, short enough to be lifted whole. Each should stand on its own if it were the only passage an engine quoted.
Not the way it used to — stuffing actively backfires. What matters now is fact density and entity clarity: real numbers, named entities, and consistent terminology.