AI Search Paraphrase Sensitivity

AI search paraphrase sensitivity is the degree to which changing the wording of a query, while preserving intent, changes the brands, domains, citations, or media mix returned by a search system.

Key points

• The paper tests seven paraphrase styles, including justification-required, source-required, quote-required, confidence-score, ranked-order, imperative-list, and keyword-only variants ^[src-028].
• Paraphrases move results less than language changes: brand lists are often fairly stable, while cited domains shift more than the recommendations themselves ^[src-028].
• AI engines are generally more stable across paraphrases than Google, though Perplexity shows stronger category-specific sensitivity in some verticals ^[src-028].
• The earned-heavy source pattern persists across paraphrase styles, even when the exact cited articles change ^[src-028].
• Practically, prompt wording can influence which sources get cited, but language localization and engine choice have larger effects on AI-search visibility ^[src-028].

Related entities

• OpenAI
• Gemini
• Perplexity API

Related concepts

• Generative Engine Optimization
• AI Search
• AI Search Citation Network
• AI Search Language Sensitivity
• Earned Media Bias in AI Search

Source references

• ^[src-028] Mahe Chen, Xiaoxuan Wang, Kaiwen Chen, Nick Koudas — “Generative Engine Optimization: How to Dominate AI Search” (2025-09-10)