The query you never see is the one that matters most

OpenAI's documentation on ChatGPT Search contains a detail that has not received enough attention: ChatGPT does not send your prompt to a search engine. It rewrites it. A user asking "what's the latest on drugs targeting CCR8 for cancer?" triggers an initial query of "CCR8 immunotherapy drug development 2025", followed by a more specific follow-up of "CHS-114 conference 2025" sent to a different provider entirely. The user never sees either query. The brand whose content answers those intermediate queries gets cited. The brand that only optimised for the original conversational prompt does not. This is the intermediate query problem, and it is one of the most underappreciated challenges in AEO today.

<h2>What intermediate queries are</h2>

An intermediate query is the search string an AI system generates internally when it decides to retrieve information from the web. It is not the user's original prompt, and it is not the final answer. It sits between the two — a translation layer that converts conversational intent into something a search index can process. In the CCR8 example, the intermediate queries are progressively more specific: the first is a broad topic query, the second is a named compound query. Each retrieves different content from different sources. The AI synthesises both into a single answer. The user experiences one coherent response. Behind it are two or more separate retrieval events, each targeting different content.

<h2>Why traditional AEO misses this layer</h2>

Most AEO advice focuses on the user-facing prompt: write content that answers the questions your audience asks, structure it clearly, use natural language. That advice is correct as far as it goes. But it addresses the wrong layer of the retrieval stack. The user's prompt is the input to the AI, not the input to the search index. By the time a query reaches a search provider, it has already been transformed. A brand that has optimised its content for conversational prompts but not for the specific, technical, intermediate queries that an AI would generate to answer those prompts may be invisible at the retrieval layer even if it is perfectly positioned at the intent layer.

<h2>How to identify the intermediate queries for your category</h2>

The most direct method is to run representative user prompts through ChatGPT Search and observe the cited sources. The sources that appear consistently are the ones whose content is being retrieved by the intermediate queries. Working backwards from those sources — examining their titles, headings, and specific terminology — reveals the vocabulary and specificity level that the AI's query rewriting is targeting. A second method is to use the Profound ChatGPT Shopping research approach: run a large sample of prompts across a topic domain and identify the cluster of intermediate concepts that appear repeatedly. These clusters represent the intermediate query vocabulary for that domain. A third method is to examine the Alibaba intent-network research, which maps the relationships between user intents and the sub-queries that AI systems generate to resolve them — the "Prompt Domain" concept described in that paper is essentially a formalisation of the intermediate query problem.

<h2>What content optimised for intermediate queries looks like</h2>

Content that performs well at the intermediate query layer tends to share several characteristics. It is specific rather than general: it answers a precise question rather than covering a broad topic. It uses the technical vocabulary of the domain rather than the conversational vocabulary of the user. It is structured so that a single section or paragraph can be extracted and cited without requiring the full context of the surrounding article. It is frequently updated, because intermediate queries for time-sensitive topics often include year or date qualifiers. And it is authoritative on narrow questions rather than comprehensive on broad ones — a 400-word page that definitively answers "CHS-114 phase 3 trial results 2025" will outperform a 3,000-word overview of CCR8 immunotherapy at the intermediate query layer, even if the overview ranks higher for the original user prompt.

<h2>The implication for content architecture</h2>

Optimising for intermediate queries requires a different content architecture than traditional SEO or even first-generation AEO. Instead of organising content around broad topic clusters, brands need to build what might be called a query lattice: a set of specific, interconnected pages that each answer a precise intermediate question, linked together so that an AI retrieving one can navigate to adjacent specifics. This is closer to how a knowledge base or technical documentation site is structured than how a typical marketing content library is organised. The brands that are already well-positioned for intermediate queries tend to be those that have invested in deep technical content, developer documentation, or structured product data — not because they were thinking about AEO, but because that content happens to match the specificity level that AI query rewriting targets.

<h2>Practical steps for AEO practitioners</h2>

Three actions are immediately actionable. First, audit your current AI citations: run your most important user prompts through ChatGPT Search and record which sources are cited. If your content is not appearing, examine what is and identify the intermediate query vocabulary those sources are optimised for. Second, identify the gap between your content vocabulary and the intermediate query vocabulary: if your content uses broad marketing language and the cited sources use specific technical terminology, that gap is where your visibility is being lost. Third, build a set of specific answer pages targeting the intermediate query vocabulary: short, precise, frequently updated pages that answer the exact questions an AI would generate as sub-queries when handling your most important user prompts. These pages do not need to be long or comprehensive — they need to be specific, accurate, and structured for extraction.

<h2>AEO Updates Takeaway</h2>

The gap between what a user asks and what an AI actually searches for is where a significant amount of brand visibility is won or lost. Most brands are optimising for the user-facing prompt. The brands that will win the next phase of AEO are the ones that understand the intermediate query layer — the specific, technical, often invisible queries that AI systems generate to retrieve the content they need to answer conversational prompts. This is not a minor refinement of existing AEO practice. It is a different way of thinking about what content is for and who it is written for. The answer, increasingly, is that the most important audience for a significant portion of your content is not a human reader — it is an AI retrieval system deciding whether your content is specific enough to cite.