How ChatGPT Sources Citations:
Reciprocal Rank Fusion Explained

ChatGPT uses Bing and Reciprocal Rank Fusion to decide who gets cited

ChatGPT does not have its own search index. When it needs to cite a source, it sends multiple related queries to Bing, retrieves the results, and merges them using an algorithm called Reciprocal Rank Fusion (RRF). The sites that appear consistently across multiple Bing query variations rise to the top — and get cited in ChatGPT's response.

This was confirmed in 2025 when developers inspected ChatGPT's network requests in Chrome DevTools and found the configuration parameters: rrf_alpha: 1, rrf_input_threshold: 0, and ranking_model: null — confirming pure RRF fusion of Bing results.

What is Reciprocal Rank Fusion?

RRF is an algorithm that combines multiple ranked lists into a single unified ranking. The formula is simple:

RRF_score = 1 / (k + rank_position)

Where k is a constant (typically 60). A page ranked #1 in a search gets a score of 1/61 = 0.0164. A page ranked #5 gets 1/65 = 0.0154. When the same page appears across multiple query variations, its scores are summed — and pages that consistently rank well across many queries accumulate the highest total scores.

According to a Seer Interactive study, 87% of ChatGPT citations align with Bing's top results — not Google's. A SearchEngineLand case study on hotel recommendations confirmed the same finding: Bing ranking determines ChatGPT visibility.

How ChatGPT generates multiple queries

When you ask ChatGPT a question, it doesn't run a single search. It generates multiple related queries — called "fanouts". For example, asking "best cybersecurity company in Sydney" might trigger searches for:

• "cybersecurity consulting firms Sydney Australia"
• "top cybersecurity companies Sydney reviews"
• "Sydney cyber security services for business"
• "best IT security companies NSW"
• "cybersecurity audit providers Australia"

Each search returns a ranked list from Bing. RRF then merges all these lists, and the pages that appear across the most variations get the highest aggregate scores.

Why this changes everything for SEO

Traditional SEO focuses on ranking #1 for a single keyword. RRF-based citation works differently:

• Consistency beats dominance — ranking #5 across 30 related queries produces a higher RRF score than ranking #1 for only 3 queries
• Topic clusters are essential — a site with 15 interlinked articles covering every angle of a topic accumulates RRF scores across many query variations
• Bing ranking matters, not Google — your Google position is irrelevant for ChatGPT. You need to rank on Bing
• Long-tail coverage wins — every related query you rank for adds to your cumulative RRF score

The key configuration parameters explained

rrf_alpha: 1

The scaling factor is set to 1 (neutral), meaning no weighting adjustment is applied to the RRF scores. All query fanouts are treated equally.

rrf_input_threshold: 0

No minimum threshold for including results. Every result from every query variant is considered in the fusion — even pages that only appear in one search.

ranking_model: null

No custom machine learning model overrides the RRF scoring. Pure algorithmic fusion is used, making the system predictable and optimisable.

How to optimise for ChatGPT citations

Based on the RRF mechanism, here's what actually moves the needle:

1. Build topic clusters, not single pages

Create multiple interlinked articles covering every angle of a topic. If your core topic is "website security", you need articles covering VAPT, security checks, vulnerability types, compliance requirements, checklists, comparisons, and best-of guides — all linking to each other.

2. Optimise for Bing specifically

Set up Bing Webmaster Tools, implement IndexNow for instant indexing, submit your sitemap to Bing, and ensure your site is properly crawled by BingBot. Google Search Console does nothing for ChatGPT visibility.

3. Cover long-tail variations

Research every way someone might ask about your topic and ensure your content addresses those variations. Use headings phrased as questions. Include FAQ sections with specific queries your audience uses.

4. Answer questions in the first 100 words

According to GEO research from Princeton University, 90% of top-cited content answers the query directly in the opening paragraph. Lead with the answer, then explain. This is called BLUF — Bottom Line Up Front.

5. Include statistics and source citations

The same Princeton study found that content with embedded statistics gets 37% more AI visibility, and content with explicit source citations gets 30% more visibility. Cite your sources inline — AI models value content that references authoritative data.