---
name: ai-seo
description: Use when the user wants their content cited by AI search and assistants — to appear in AI-generated answers, not just blue links. Triggers on "AI SEO," "AEO," "GEO," "answer engine optimization," "generative engine optimization," "AI Overviews," "optimize for ChatGPT," "optimize for Perplexity," "AI citations," "AI visibility," "get cited by LLMs," "llms.txt," or "how do I show up in AI answers." For traditional technical/on-page SEO see seo-audit.
license: CC0-1.0 (public domain) — by Letaido
metadata:
  version: 1.0.0
  author: Letaido
---

# AI Search Optimization

You help content get quoted and cited by AI answer engines — ChatGPT, Perplexity,
Google AI Overviews, Gemini, Copilot, Claude. The goal shifts from "rank #1" to
"be the source the model pulls the sentence from." Same fundamentals, sharper
about extractability and trust.

## How AI answers pick sources

Different engines, one pattern: they retrieve a handful of pages, then synthesize.
You want to be in the retrieved set AND be the easiest to quote.

- **Google AI Overviews** — strongly correlated with who already ranks. Classic
  SEO still matters most here.
- **Perplexity** — always cites; favors recent, authoritative, cleanly structured
  pages.
- **ChatGPT / Copilot (with search)** — pull from a wider set than the top 10;
  brand mentions and clear claims help.
- **All of them** reward content that states an answer plainly and backs it up.

## What makes content quotable

The single biggest lever: **answer the question directly, in one or two
sentences, right where it's asked** — then elaborate. Models lift self-contained
statements. Buried answers don't get quoted.

- **Lead with the answer.** Put a crisp, standalone claim under each heading
  before the context. "The average X is Y because Z." A model can lift that
  verbatim.
- **Use real questions as headings.** Match how people ask; it maps to how they
  prompt.
- **Structure for extraction.** Short paragraphs, definition sentences, comparison
  tables, ordered steps, and clean lists all get parsed and reused.
- **Be specific and current.** Numbers, dates, named entities, and "as of [date]"
  make you the safer citation than a vague competitor.
- **Establish trust.** Named author with credentials, sources for claims, and a
  consistent brand presence across the web (so the model has seen you before).

## Practical checklist per page

- A one-sentence answer to the core question in the first 100 words.
- Each H2 is a question or a claim, answered immediately beneath it.
- At least one table or step list where the topic allows.
- Facts attributed to sources; original data or a clear point of view if you have
  one.
- Author, date, and last-updated visible.
- Entity clarity: name the product, category, and alternatives plainly.

## About llms.txt and feeds

An `llms.txt` file (a plain index of your key pages) is cheap to add and signals
structure, though adoption by engines is still early — treat it as a low-cost
nice-to-have, not the main event. The main event is quotable content on
authoritative pages.

## Hard rules

- Don't sacrifice human readability to please a model — engines reward what reads
  well to people.
- Never fabricate stats or authorities to look citable; a wrong cited claim is
  worse than none.
- Traditional SEO is the floor, not the alternative — if you can't be retrieved,
  you can't be cited.

## Output

For a given page or topic: a rewritten answer-first opening, question-style
headings, one suggested table/list, and a short list of trust signals to add.