Technical GEO
How AI search retrieves information
From query to cited answer: the retrieval and synthesis steps, and what each means for publishers.
Most AI search experiences combine a language model with a retrieval step. The details differ between systems and are often not published, so treat the following as a general model rather than a description of any one product.
1. Query interpretation
The system rewrites the user's question into one or more search queries. A single question can fan out into several sub-queries.
2. Retrieval
Candidate documents are fetched from a web index, a live search, or a proprietary store. Pages that block relevant crawlers, require client-side rendering, or hide content behind interstitials are less likely to be retrieved.
3. Passage selection
Documents are split into passages and filtered for relevance. Sections with a clear heading and a self-contained answer tend to survive this step better than text that depends on earlier paragraphs.
4. Synthesis
The model writes an answer from the selected passages and its own trained knowledge. This is where omissions and inaccuracies are introduced.
5. Attribution
Some systems attach citations to sentences or list sources. Citation is a design choice of each system and changes over time.
What publishers control
You control steps 2 and 3 most directly: access, rendering, structure, clarity. You influence step 4 through the consistency of information about you across sources. You do not control step 5.
Related reading
- Technical GEOEntity optimization: a working guideMake your organisation unambiguous: one name, one description, consistent identifiers, clear relationships.7 min read
- Technical GEOGEO technical audit checklistTwenty checks across access, rendering, structure and entity data.6 min read
- AI SearchWhat is Generative Engine Optimization?A plain definition of GEO, what it changes about search work, and what it cannot promise.6 min read
Terms used here are defined in the glossary.