What We Know About ChatGPT Search Ranking Factors

ChatGPT Search isn't Google Search

ChatGPT Search launched in October 2024 and reached general availability in early 2025. By 2026 it's the second-most-used answer engine after Google's AI Overviews. For agencies advising clients on SEO, "ranking" is no longer a single-engine optimization problem — it's a multi-engine one.

ChatGPT Search uses Bing's index as its primary retrieval source, but the ranking signals it applies on top of that index are substantially different from both Bing's traditional ranking and Google's. This post covers what's confirmed by OpenAI documentation, what's inferred from observed citation patterns over the last 12 months, and what to optimize for if you want to be cited.

Confirmed: how the system works

From OpenAI's published documentation and developer talks:

1. Index source is Bing. ChatGPT Search retrieves from Bing's web index for live results. Pages not indexed by Bing cannot be cited.
2. Citations are required for "Search" mode answers. When the user invokes search (via the search button or by triggering OpenAI's automatic search heuristic), the response must include source citations.
3. Crawler is OAI-SearchBot. Distinct from GPTBot (used for training) and ChatGPT-User (used when a ChatGPT plugin or browse tool fetches a URL on demand). Sites can allow or block each independently in robots.txt.
4. Re-ranking is happening. OpenAI re-ranks Bing's results using a model that emphasizes citation-worthiness, not click-likelihood.

Inferred from observed citation patterns

The patterns that consistently produce citations in our testing across 200+ queries over 12 months:

1. Specific numerical claims with sources

ChatGPT preferentially cites pages that make verifiable numerical claims. A page that says "average page-load time across the top 100 e-commerce sites is 3.2s" gets cited far more often than a page that says "page load times have gotten faster recently."

The grader at /tools/ai-visibility-grader checks for what it calls "extractable claims" — sentences that follow the pattern [entity] [is/was/has] [specific value/metric]. Pages with high extractable-claim density are more often cited.

2. Definitional structure near the start

ChatGPT favors pages that define their topic in the first 200 words. A blog post that buries the definition under three paragraphs of context loses citation rank to a page that opens with "Cumulative Layout Shift (CLS) is a Core Web Vital that measures..."