GPT

GPT, short for Generative Pre-trained Transformer, is OpenAI's family of LLMs released since 2018. GPT-2 caught the attention of researchers, GPT-3 brought the technology into mainstream awareness, and ChatGPT — built on GPT-3.5 — launched in late 2022 and effectively kicked off the LLM era. The lineage has since continued through GPT-4, GPT-4o and the GPT-5 line, adding multimodality, real-time voice, search and tool use along the way.

Technically, GPT models combine a decoder-only Transformer, massive pretraining on internet-scale text, instruction tuning and RLHF. The base model learns to predict the next token; the post-training stages teach it to be helpful, honest and safe. A surprising amount of the 'it actually understands me' feeling you get from ChatGPT comes from those alignment steps rather than the raw base model.

From a marketer's point of view, GPT is not just one vendor's model — it is the first surface you measure in any GEO program. ChatGPT dwarfs every other assistant in usage in both the US and Korea, so its answer panel is the primary citation pool for AI brand visibility. With ChatGPT Search now blending live web retrieval into the answer, you also have to think about the live index, not only the training data.

Version-level behaviour matters more than people realise. GPT-4o is fast and natively multimodal, GPT-5 leans into deep reasoning and longer context, and the mini and nano variants exist for cost efficiency. When you track citation rate, splitting it by model version is what makes the data actually useful, because free and paid ChatGPT users hit different models and can see meaningfully different answers about the same brand.

In Korea, GPT dominates both direct ChatGPT usage and API-powered domestic products. The wrinkles are real, though: Korean costs noticeably more per token than English, and several Korean verticals are under-represented in training data, which makes answer quality uneven. A practical Korean GEO workflow has to assume GPT may have a stale or thin picture of your brand, then deliberately fill those gaps with sources GPT actually reads.