How LLMs Work
Knowledge Cutoff
A knowledge cutoff is the date after which a language model has no trained knowledge, the point its training data ends. Without web search, a model cannot know about products, rebrands, or news after its cutoff. This explains why AI chatbots describe outdated versions of brands and why retrieval features matter.
What the cutoff means in practice
Every model release carries a cutoff date reflecting when its training data was collected. Ask about anything later, a launch, an acquisition, new pricing, and the model either admits ignorance or, worse, confidently answers from stale knowledge. Cutoffs typically trail a model's release by several months to a year.
Cutoffs differ per model, so the engines users actually talk to hold different vintages of the web. A rebrand from last quarter may exist in one assistant's answers and be completely absent from another's.
The brand consequences of frozen knowledge
From-memory AI answers describe the brand you were at training time: old positioning, discontinued plans, former pricing, missing products. For fast-moving companies this gap is a real revenue risk, prospects hear obsolete information delivered with full confidence, a close cousin of hallucination.
Platforms compensate with web search: when retrieval fires, the answer reflects live pages instead of frozen weights. Whether a given prompt triggers retrieval varies by platform and phrasing, so your visibility is a blend of both knowledge ages.
Working around the cutoff
Two moves close the gap. First, win the retrieval layer so current facts override stale memory: keep updated pages indexed, crawlable, and citable, since grounded answers pull from them. Second, prepare for the next training cycle by propagating changes everywhere, your site, directories, press, so the next cutoff captures the updated you.
After major model releases, answers can shift overnight as new cutoffs land. Geonimo's daily prompt tracking makes those shifts visible immediately, showing whether a new model version learned your latest story or is still telling the old one.
Frequently asked questions
Why does ChatGPT not know about my new product?
If the product launched after the model's knowledge cutoff, it simply is not in the model's trained memory. The assistant can only learn about it through web search at answer time. Ensure launch pages are indexed and citable so retrieval surfaces them, and the next model generation will absorb them permanently.
Do AI search engines have knowledge cutoffs?
The underlying models do, but retrieval masks it: engines like Perplexity and ChatGPT Search fetch live pages, so answers include post-cutoff information with citations. Cutoffs still matter for the model's background framing of brands and for prompts that do not trigger a search.
How do I find out a model's knowledge cutoff?
Providers publish cutoff dates in model documentation, and you can probe a model by asking about datable events. Remember that platforms swap models frequently and may blend retrieval, so behavior matters more than the stated date: test your actual brand prompts to see which era of facts comes back.
Related terms
Training Data
Training data is the text corpus, web pages, books, code, forums, and licensed content, used to teach a language model during training. It determines what the model knows and believes, including how it describes brands. A brand's presence in training data shapes AI answers for years, since models retrain infrequently.
Retrieval-Augmented Generation (RAG)
Retrieval-augmented generation is an AI architecture where a language model retrieves relevant documents, typically via web or database search, before generating its answer, grounding the response in fetched content. RAG powers AI search engines like Perplexity and ChatGPT Search, and it is the mechanism through which web pages earn citations in AI answers.
AI Hallucination
An AI hallucination is a confident but false statement generated by a language model, invented facts, wrong attributions, or nonexistent details, produced because models predict plausible text rather than verify truth. For brands, hallucinations matter doubly: AI can state false things about your company, and false claims erode user trust in uncited answers.
Content Freshness
Content freshness is how recently a page was published or meaningfully updated, and how current its information is. Search engines boost fresh content for time-sensitive queries, and AI engines with live retrieval favor recent sources — while models without browsing are limited by their training cutoff, making maintained content visible across both modes.
Last updated: 2026-06-11
Track this for your brand
Geonimo monitors how ChatGPT, Perplexity, Claude, Gemini and Google AI talk about your brand — and generates the content that gets you cited.
Get your free audit