tmj-90 Optimise content for LLM citation
AEO (Answer Engine Optimisation) optimises content to be cited in LLM-generated responses — distinct from SEO, which optimises for click-through rankings. The signal sets differ.
AEO vs SEO
| SEO | AEO | |
|---|---|---|
| Optimises for | Click-through ranking | Citation as authoritative source |
| Success metric | Position 1-10, organic traffic | Citation count across LLMs |
| Key signals | Backlinks, keywords, page speed | E-E-A-T, structured facts, primary-source signals |
| Update cadence | Weeks–months | Days–weeks (LLM training cycles) |
Both coexist — a page can rank #1 on Google AND be cited by Perplexity.
E-E-A-T signals for LLM citation
LLMs prefer content that looks like a primary source to their training data:
| Signal | Implementation |
|---|---|
| Experience | First-person case studies; dated real examples ("In our 2026 audit of 50 repos…") |
| Expertise | Author bio with credentials; technical depth; cite primary sources |
| Authoritativeness | External backlinks from authority domains; schema.org markup; Wikipedia presence |
| Trustworthiness | HTTPS; contact info; transparent corrections; verifiable claims with data |
Factual density — LLMs prefer pages with a high ratio of verifiable claims per 1,000 words. Prose that states facts > prose that describes opinions.
Content structure for citation
LLMs extract from:
- Direct-answer paragraphs — a question as a heading followed immediately by a 1–2 sentence direct answer.
- Structured data — FAQPage and HowTo schema (see
schema-markup). - Definition blocks — "X is [concise definition]" sentences at the start of sections.
- Comparison tables — LLMs reproduce tables well; use for vs. comparisons and feature matrices.
Citation-hostile patterns to avoid
- Gated content (LLMs can't read it during training crawls).
- JavaScript-rendered text (crawlers often miss it).
- Opinion without evidence ("we believe X is important").
- Content updated without a
dateModifiedschema field. - No author or institutional attribution.
Steps
- Identify citation-worthy content. Pages that answer specific questions with unique data, first-party case studies, or authoritative comparisons are candidates. Brand-voice content without factual claims is not.
- Audit E-E-A-T gaps. For each target page: score Experience (0–3), Expertise (0–3), Authoritativeness (0–3), Trustworthiness (0–3). Identify the lowest-scoring dimension and fix it first.
- Add direct-answer structure. Rewrite the first paragraph of each target section as a direct answer to the implied question. Add schema (FAQPage/HowTo where appropriate).
- Improve factual density. Replace opinion sentences with evidence sentences. Add specific numbers, dates, and attributable sources.
- Track citations. Query each target LLM (ChatGPT, Perplexity, Claude) with the exact question the page answers. Note whether your domain is cited. Repeat monthly.
- Verify. Rich Results Test for schema; check
dateModifiedis present and accurate; confirm author bio is visible and crawlable.
Review checklist
- Direct-answer paragraph in each target section — question heading + immediate direct answer.
- E-E-A-T score ≥ 2/3 in all four dimensions — no dimension at 0.
- FAQPage or HowTo schema on pages with question-and-answer structure.
- Author bio visible and crawlable — not loaded via JavaScript after parse.
dateModifiedin Article schema and visible on page.- No gated content for citation-target pages — if it needs a login, it won't be cited.
Rules
- AEO applies only to pages with verifiable factual content — brand/opinion content is out of scope.
- Never fabricate statistics or case studies to improve E-E-A-T — citation by LLMs of false claims is a liability.
- Track citation rates monthly; treat uncited priority pages as an audit finding.