Step one: the moment is detected
Large language models are pattern engines trained on how millions of human situations unfold. A sentence about a fender bender and a sore neck matches the accident pattern; jeans that no longer fit matches the weight struggle pattern; “our biggest customer wants SOC 2” matches the compliance crunch pattern. The assistant does not need the user to name the situation. The context is the query.
Step two: needs are inferred
From the detected moment, the model projects what typically comes next. An auto accident implies possible injury, so medical evaluation. An at-fault driver implies liability, so legal counsel. A car in the shop implies transportation, so rental options. These inferences are probabilistic and instant, and they happen whether or not the user asked. The assistant is effectively running the playbook a good friend with expert knowledge would run.
Then the needs get ranked by urgency and the conversation follows the top branches. This is why the same opening sentence reliably produces the same recommendation pathways across ChatGPT, Gemini, Claude and Perplexity, even as the named brands differ by assistant.
Step three: needs become named brands
When a pathway reaches options, retrieval kicks in: the assistant pulls from the sources it trusts, review sites, comparison pages, category articles and brand pages, and names specific providers. This is the step brands can influence. The moment and the inference are fixed human patterns; the shortlist is built from sources you can measure and fix. Which sources feed which answers is exactly what an AI visibility audit maps, and the discipline of doing it moment by moment is Contextual AI Presence Mapping©. If the assistant names competitors at a moment you should own, the fix list runs through the sources behind that answer, as covered in correcting wrong information in ChatGPT.
