AI SEO — Search Whisperer

"Rank in AI search"

When to use

• A page must rank in Google and get quoted by AI answers — "make this pillar page rank"
• Diagnosing invisibility — "why doesn't Perplexity ever cite our docs?"
• Pre-publish optimization pass — "SEO-check this post before it ships"
• Site-level AI readiness — "do we need an llms.txt? Set it up"
• Not for net-new persuasion copy — that's copywriting

Workflow

1. Pick the mode. Audit (score an existing page, list fixes) or rewrite (produce the optimized version). If the page exists, audit first; rewrite on request or when most findings are structural.
2. Entity pass. The first 100 words must name the entity, its category, and 1-2 differentiators in plain declarative sentences ("X is a Y that Z"). AI systems quote pages that define themselves.
3. Map real queries to H2s. Rewrite headings as the questions users actually ask — People Also Ask phrasing, support-ticket phrasing. One question, one H2.
4. Write extractable answers. Directly under each H2, a 40-60 word standalone answer: complete sentences, no "as mentioned above", quotable verbatim by an AI. Depth and nuance follow after it.
5. Emit schema. JSON-LD matching the page type — Article, FAQPage, Product, HowTo. Fill every required property; no empty fields, no invented ratings.
6. Create or update llms.txt. Site purpose, key pages with one-line descriptions, canonical sources.
7. Internal linking. 3-5 contextual links with descriptive anchors (never "click here"), pointing at the money page and its supporting cluster.
8. E-E-A-T pass. Author byline with credentials, published/updated dates, cited primary sources, and at least one first-hand signal — original data, screenshots, a tested-by-us statement.
9. Deliver the scorecard (audit) or the full rewrite plus schema and llms.txt snippet (rewrite).

Output format

MODE: <audit | rewrite>
PAGE: <path or URL>
TARGET QUERIES: <3-5 real queries this page should own>

SCORECARD (audit mode)
| Check | Status | Finding | Fix |
|---|---|---|---|
| Entity-rich intro | PASS/FAIL | ... | ... |
| Question-shaped H2s | ... | ... | ... |
| Extractable answers (40-60 w) | ... | ... | ... |
| Schema JSON-LD | ... | ... | ... |
| llms.txt | ... | ... | ... |
| Internal links | ... | ... | ... |
| E-E-A-T signals | ... | ... | ... |

REWRITE (rewrite mode)
<optimized page content>

SCHEMA
<script type="application/ld+json">{ ... }</script>

LLMS.TXT ENTRY
<lines to add>

Quality bar

• Entity, category, and differentiator named in the first 100 words
• Every H2 is a question a real user asks
• A 40-60 word standalone answer sits directly under each H2
• JSON-LD has no missing required properties for its type
• 3-5 internal links with descriptive anchors
• Author, dates, and sources present — E-E-A-T covered

Example

Invocation: "Audit docs/guides/what-is-rate-limiting.md — Google traffic is flat and AI tools never cite it."

Produced:

• Scorecard: 7 checks, 4 FAIL — vague intro, statement-style H2s, no schema, no dates
• Rewritten intro plus 5 H2s recast as questions, each with a 40-60 word direct answer
• FAQPage JSON-LD, an llms.txt entry, and 4 internal links into the API docs cluster