--- language: en title: "Grounding" description: "Grounding connects an AI response to reference information. Learn how to check sources and claims, and why an answer with links still needs review." canonical: "https://tropk.ai/en/resources/glossary/grounding" last_modified: "2026-10-12T16:15:07.788Z" --- # Grounding Reviewed on 2026-10-06. ## What it means Grounding is the link between an [AI response](/en/resources/glossary/ai-response) and information used as a reference, such as retrieved documents or search results. It can be described as anchoring in sources. The aim is to support generation with relevant evidence rather than relying solely on the model’s learned knowledge. The link can involve different mechanisms and levels of transparency. ## Why it matters for marketing This matters in marketing when a system describes prices, features, availability, and [brand attributes](/en/resources/glossary/brand-attribute). Information that changes needs to come from accessible, clear sources. Publishing an accurate description on a website does not guarantee that an assistant will use it immediately, but it provides a verifiable reference for people doing research. Reviews should examine both the response and the page it uses. ## Example in practice An answer states that a particular product offers CSV export and includes a link to the documentation. If the documentation explains the feature and its conditions, it provides relevant support. If the link leads to a generic page or describes a different plan, the reference does not support the entire claim, despite the appearance of a well-grounded answer. ## How to apply or verify Separate the important claims and find evidence for each one. Check the version, date, and restrictions, especially for commercial offers. Record when a source cannot be opened or does not contain the attributed fact. When creating content, make the subject, feature, and its conditions explicit to reduce ambiguity and make human review easier. ## Important distinctions Grounding is not synonymous with guaranteed truth. A source may be wrong, outdated, or misinterpreted. Nor do all consulted sources need to appear as visible citations. A [citation](/en/resources/glossary/ai-citation) is an element shown to the user; grounding describes how generation relates to its references. Evaluating both requires evidence beyond the presence of links. ## Sources and references - [Optimizing your website for generative AI features on Google Search]() — Google Search Central - [Searching the web with ChatGPT]() — OpenAI ## Continue a leitura - [AI citation]() - [Primary source]() - [RAG]() ## About the author [Vinicius Pacheco]() Responsible for Tropk’s technical content, focusing on crawling, indexing, structured data, and agent access to content. --- [All pages in Markdown]()