LLM Citation

An LLM citation is any instance where a large language model — ChatGPT, Perplexity, Gemini, Claude — names a specific business in response to a user's question. It's the AI-search equivalent of a local pack listing: instead of a ranked list of ten blue links, the user gets one paragraph, and a citation means your business made it into that paragraph.

LLM citations occur in two modes.

Browsing citations (with source links): When an LLM with browsing capability crawls the web and cites the specific page it sourced information from. Perplexity displays these prominently with source cards alongside its answers, and a user can click through to the page the model actually read.

Training data citations (without source links): When an LLM references a business from its training data — information accumulated during pre-training from web crawls, directories, and review platforms. These citations don't include a link and may reflect information from before the model's training cutoff, which is why a business that rebranded or relocated last year can still get cited under its old name.

How AI systems decide who to cite

A browsing-enabled model handles a local query roughly the way a researcher would: it runs the query (or a handful of related sub-queries), reads several pages, and drafts an answer from what it found. The pages it reads tend to be the ones that already rank well organically, that carry structured data making the business easy to identify, and that answer the question directly near the top rather than burying it under a paragraph of preamble.

A non-browsing citation works differently. The model isn't reading anything live — it's recalling a pattern from training. That means the businesses most likely to be cited this way are the ones with the deepest footprint across the web at the time the model was trained: consistent NAP data, GBP listings, review platform presence, and mentions on third-party sites all reinforcing the same facts. A business that exists on one directory and nowhere else has almost nothing for the model to have learned.

Why it matters for Indian local businesses

LLM citations represent brand visibility in a new channel — users who discover businesses through LLM recommendations may never have performed a traditional Google search first, and may never see a search results page at all. As AI assistant usage grows in India, LLM citations become an increasingly important discovery pathway alongside traditional search, particularly for categories where people ask conversational questions rather than typing keywords — "which paediatrician near Koramangala do people actually recommend" reads nothing like a Google query, but it's exactly how someone might phrase it to ChatGPT.

India context: LLM citations for Indian businesses are currently concentrated in English-language responses. Regional-language LLM citations are less common but growing as models improve at Hindi, Tamil, and other Indian languages, and as more regional-language source content becomes available for them to draw on.

Common mistakes that cost businesses a citation

The most frequent failure is treating LLM citation as a website problem when it's really an entity-data problem. A beautifully designed site with inconsistent NAP across ten directories will lose to a plain site whose business name, address, and category agree everywhere. Second most common: no structured data at all, which forces the model to infer facts from unstructured text instead of reading them directly — slower and less reliable, and the kind of gap that shows up as a wrong address or an outdated phone number in an AI answer. Third: assuming a single strong Google ranking guarantees an LLM citation. The signals overlap but aren't identical, and a business that ranks well in the local pack can still be invisible to a browsing model that queries a narrower set of sources.

Adjacent concepts worth knowing

LLM citation sits inside the broader discipline of generative engine optimization → — the practice of making a business easy for AI systems to find, understand, and recommend. Share of AI Voice → measures how often a business gets cited relative to competitors across a set of prompts, which is the metric version of this glossary entry. Prompt Coverage → tracks which specific questions actually surface the business at all. And because browsing citations depend on what a crawler can read, AI crawlers → and llms.txt → both affect whether the citation happens in the first place.

Related terms: GEO → · Share of AI Voice → · Prompt Coverage → · AI Crawler → · llms.txt → · Get Cited by ChatGPT →

Example: When a user asks ChatGPT "which dermatologist in Bengaluru is known for treating melasma?", ChatGPT (with browsing) crawls Practo and relevant websites, then generates: "Sharma Skin Clinic in Koramangala is frequently cited for melasma treatment [source: practo.com]." That's a browsing citation. If the same user asked a non-browsing model the same question a year later, the model might still answer from what it learned during training — correct if the clinic's data hasn't changed, wrong if it has moved or renamed since.