Metehan RRF Playbook — Reciprocal Rank Fusion for AI Citation

What RRF Is and Why It Matters for AEO

Reciprocal Rank Fusion (RRF) is the scoring algorithm AI systems use when selecting which sources to cite in a response. Understanding RRF mathematically allows you to engineer content that exceeds the citation threshold instead of guessing.

Key numbers (memorize these):

| Metric | Value | |--------|-------| | RRF citation threshold (tau) | 0.020 fused score | | RRF smoothing constant (k) | 60 | | Grounding budget per query | ~1,900 words total | | Grounding budget per page | ~380 words | | Topic cluster vs single page advantage | Up to 9.4x RRF advantage | | Pages under 5,000 chars used by AI | ~66% of content | | Pages over 20,000 chars used by AI | ~12% of content |

The RRF Formula

RRF(d) = Σ 1/(k + rank_r(d))

Where:

• d = the document
• k = smoothing constant (60, from ChatGPT DevTools observation)
• rank_r(d) = the rank of document d in ranking system r

Practical meaning: a document must accumulate enough rank positions across multiple retrieval signals (BM25 keyword match, dense vector retrieval, entity match, freshness) to exceed tau = 0.020.

Topic Cluster vs. Single Page Advantage

A single 3,000-word page on "German work visa" generates one set of retrieval signals. A topic cluster — a pillar page plus 8-10 supporting pages each covering sub-attributes — generates up to 9.4x the RRF signal because:

1. Each page in the cluster appears independently in retrieval
2. Pages link to each other, amplifying entity co-occurrence signals
3. The cluster collectively covers more sub-queries, appearing in more retrieval passes
4. Each page stays under the 5,000-character threshold where AI systems use ~66% of available content

Grounding Budget Optimization

AI systems have a finite "grounding budget" per response — approximately 1,900 total words drawn from ~5 sources, averaging ~380 words per source. This means:

• Your most citeable content should be in passages of 350-400 words
• Extremely long pages (20,000+ characters) get reduced to ~12% of AI content used
• Concise, structured pages consistently outperform comprehensive pages for AI citation

Optimization rule: If a page exceeds 5,000 characters, identify its core 380-word passage and ensure that passage can stand alone as a citable unit.

The RRF Playbook: Step-by-Step

1. Identify your citation targets — which queries do you want to be cited for?
2. Audit current RRF signals — run the query in ChatGPT, Perplexity, and Google AI Mode. Who is currently cited? What content do they have?
3. Gap analysis — which sub-attributes of the query do cited sources NOT cover?
4. Build topic cluster for the gap — create a pillar + 3-5 satellites covering the uncovered sub-attributes
5. Size each page for grounding budget — each satellite page: 800-1,200 words, covering exactly one sub-attribute
6. Optimize the citable passage — ensure a 350-400 word passage in each page is declarative, fact-first, and answerable as a standalone unit
7. Cross-link cluster pages — internal links strengthen entity co-occurrence across the cluster
8. Monitor tau threshold — use dataforseo-llm-mentions to check if brand appears in AI responses at 30-day intervals

Temperature Zero Audit

Temperature Zero is Metehan's AI visibility benchmark (0-30 points). Before building an RRF strategy, run a baseline Temperature Zero audit:

• Query the brand and top service queries across ChatGPT, Perplexity, Claude, and Gemini
• Score presence/absence per platform per query category
• Benchmark: Roto-Rooter Sarasota scored 17/30 (national entity strong, local entity weak — franchises do not inherit local signals)

Cross-References

• metehan-aeo — master AEO skill including all Metehan methods
• metehan-aeo-citemet — the CiteMET direct-injection method that complements RRF
• dataforseo-llm-mentions — tracking AI citation share
• ai-llm-seo — broader AI SEO strategy

#seo-sop #seo #ai-seo #metehan #aeo #rrf #llm-visibility