RRF Top-n Playbook

Fresh

This content reflects Metehan Alp's RRF methodology as of 2025-2026.

The Reciprocal Rank Fusion (RRF) Top-n Playbook is a mathematical framework for understanding and engineering citations in AI search platforms. It answers a question most AEO practitioners are guessing at: how high do you actually need to rank, and for how many queries, to get cited?

The Mathematical Formula

RRF computes a fused relevance score across multiple ranked lists (sub-queries):

$$S(d) = \sum_{i=1}^{n} \frac{w_i}{k + r_i(d)}$$

Where:

• S(d) = fused score for document d
• w_i = weight assigned to ranked list i
• k = smoothing constant (typically 60)
• r_i(d) = rank of document d in list i
• n = number of ranked lists (sub-queries)

How to Read This

Each sub-query the AI system generates produces a ranked list of results. RRF combines these lists by giving higher scores to documents that appear near the top across multiple lists. The smoothing constant (k=60) prevents any single top-1 result from dominating the fused score.

A document ranked #1 in one list contributes 1/(60+1) = 0.0164 to its fused score. A document ranked #10 contributes 1/(60+10) = 0.0143. The difference between rank 1 and rank 10 is small. The difference between appearing in multiple lists versus one list is massive.

Citation Threshold

$$\tau = 0.020$$

The citation threshold (tau) is the minimum fused score a document needs to be cited by an AI platform. Based on Metehan's analysis, a document must accumulate an RRF score of at least 0.020 to cross the citation threshold.

Practical Rules for Meeting the Threshold

You do not need to compute RRF scores manually for every query. These practical rules translate the math into actionable ranking targets:

Rule	Condition	Fused Score
Rule 1	Rank in top 40 for 2 sub-queries	Exceeds tau
Rule 2	Rank in top 90 for 3 sub-queries	Exceeds tau
Rule 3	Rank #1 for 1 sub-query AND top 80 for 1 more	Exceeds tau
Rule 4	Rank in top 140 for 4 sub-queries	Exceeds tau

Core Takeaway

Ranking #10 for 3 different sub-queries beats ranking #60 for everything. Breadth across sub-queries matters more than depth in a single query.

4-Step Operationalization

Step 1: Sub-Query Cluster Mapping

Identify the sub-queries an AI system would generate for your target topic.

1. Take your primary query (e.g., "best project management tools for remote teams")
2. Map 8-16 sub-query variations:
   • Related queries: "project management software features"
   • Implicit queries: "remote team collaboration tools"
   • Comparative queries: "Asana vs Monday for remote teams"
   • Procedural queries: "how to set up project management for remote teams"
   • Contextual queries: "project management challenges for distributed teams"
3. Validate by checking Google AI Mode or ChatGPT to see which sub-queries they actually generate

Step 2: SERP Data Collection

For each sub-query, collect the top 60-100 results from Google:

• Use Screaming Frog, Semrush, or Ahrefs to pull SERP data
• Record your domain's position for each sub-query
• Note which competitors appear across multiple sub-queries

Step 3: Compute Fused Score

For each page on your site, calculate:

Score = sum of (1 / (60 + rank)) for each sub-query where the page ranks

If a page ranks #15 for sub-query A and #30 for sub-query B:

Score = 1/(60+15) + 1/(60+30) = 0.0133 + 0.0111 = 0.0244

This exceeds tau (0.020), so the page meets the citation threshold.

Step 4: On-Page Engineering

For pages that fall below the threshold, identify which sub-queries they rank weakest for, and optimize:

• Add sections that directly address missing sub-query intents
• Improve topical coverage for sub-queries where you rank beyond position 100
• Create supporting content that targets individual sub-queries and links back to the hub page

AYIMA Testing Data

Research from AYIMA tested RRF behavior across different query types:

Query Type	Avg Sources Retrieved	Sub-Queries Generated	Top Source Domain
Informational	6.2	4-8	Wikipedia, established publishers
Commercial	4.8	3-6	Review sites, brand sites
Navigational	2.1	1-3	Target brand site
Local	3.4	2-5	Local directories, brand sites
Comparison	5.7	5-10	Review sites, forums

Key Observations

• Informational queries generate the most sub-queries, creating more opportunities for RRF-based citation
• Comparison queries generate high sub-query counts but spread citations across more sources
• Navigational queries generate few sub-queries, making RRF less relevant (brand authority dominates)
• Commercial queries represent the highest-value citation opportunities with manageable sub-query counts

Core Takeaway

The math is clear: breadth beats depth for AI citations.

A page that ranks #10 for 3 different sub-queries earns a higher RRF score than a page ranking #1 for a single sub-query alone. This inverts traditional SEO thinking, where ranking #1 for your primary keyword was the goal.

For AI search optimization, the goal is to rank well enough across enough sub-queries to cross the citation threshold. Target the practical rules: 2 in top 40, or 3 in top 90, or 4 in top 140.

Further Reading

• Query Fan-Out Analysis -- How to identify the sub-queries AI systems generate
• AI Overview Optimization -- Structuring content for Google's AI features