GlossaryAI search
What is grounding in AI?
Also called: grounded generation, retrieval-augmented generation, RAG
Definition
Grounding is giving an AI model real source content, such as search results or documents, at the moment it answers, so the answer is based on that content instead of only what the model memorized in training.
Grounding, explained
A language model on its own answers from patterns learned during training. That knowledge has a cutoff date, and the model can confidently produce things that aren't true. Grounding fixes part of this by fetching relevant content when a question arrives and putting it in front of the model, with instructions to answer from it. Google's definition, in its crawler documentation, is providing content from the Google Search index to the model at prompt time to improve factuality and relevancy.
The general technique is often called retrieval-augmented generation, or RAG. For web answers, the retrieval step is a search: the system turns the question into queries, gets results from an index, reads the top pages, and writes an answer, usually citing the pages it used. ChatGPT search, Perplexity, Claude with web search, Gemini and Google's AI Overviews all ground answers in web content this way.
For site owners, grounding is the mechanism behind citations. Your page can only be used if it's in the index the system searches, which is why AI search crawlers and robots.txt rules matter. And it can only be chosen if it matches the sub-question and states things clearly enough to be used.
Grounding is also why freshness and accuracy matter more than before. An assistant grounded in your pricing page will repeat what it says. If the page is out of date, or if an old directory listing says something different, the answer may be wrong about you.
Controls exist for some systems. Google-Extended governs whether Google may use your content to ground Gemini. For other assistants, the search crawler rules in robots.txt decide whether your pages can be retrieved.
Why it matters for founders
If your pages are what an AI grounds its answer on, you shape what it says about your category and product. If they're not reachable, it grounds on someone else's description of you.
Example
A user asks Claude what a product costs. Claude searches, retrieves the product's pricing page and a review, and answers with the current price, citing both. An outdated review would have produced a wrong answer.
Common mistakes
- Blocking AI search crawlers and expecting accurate AI answers about your product.
- Leaving outdated pricing or feature pages live.
- Inconsistent product facts across your site and directory listings.
- Assuming a model's training data is the only thing that shapes answers.
Sources
Checked
Related terms
- LLM citationAn LLM citation is a link or named source that an AI assistant shows alongside its answer to say where information came from. Being cited sends referral traffic and signals that your page was trusted.
- Google-ExtendedGoogle-Extended is a robots.txt token that controls whether content Google crawls from your site may be used to train Gemini models and for grounding. It doesn't affect Google Search, including AI Overviews.
- AI crawlersAI crawlers are bots run by AI companies that fetch web pages, either to collect training data, to build an index for AI search answers, or to read a page a user asked about. Each has its own robots.txt name.
- Generative engine optimization (GEO)Generative engine optimization (GEO) is the practice of making your content more likely to be found, used and cited in answers from AI systems like ChatGPT, Perplexity, Claude, Gemini and Google's AI Overviews.