Glossary

Plain definitions of AI visibility words: GEO, AEO, mention rate, fan-out, grounding, earned media, llms.txt, citation, hallucination and more.

By Dasha. Updated November 1 2026. https://heardofyou.io/glossary

Plain definitions of the words on this site and in AI visibility work generally. One to three short sentences each, with the origin where it helps. Missing a word? Email hello@heardofyou.io and we will add it.

| Term | What it means |
|---|---|
| AI visibility | How often, and how well, an AI assistant names your product when someone asks a question you should answer. On this site it is always a rate across many answers, never a rank in one. |
| GEO (generative engine optimisation) | Changing what you publish, on and off your site, so AI assistants are more likely to cite or recommend you. The term comes from a 2024 paper (Aggarwal et al., KDD 2024). Citations, quotations and statistics raised visibility. Keyword stuffing did not. Often sold as magic. It is mostly ordinary proof. |
| AEO (answer engine optimisation) | GEO's older cousin: making a page the direct answer to a question, so a search engine or assistant can lift it. In practice people use the two terms interchangeably. |
| LLM (large language model) | The software under ChatGPT, Claude, Gemini, Perplexity and Muse. It predicts text from patterns in its training data. Modern ones also read live web results before answering. |
| Answer engine | A search tool that gives one written answer instead of a list of links. ChatGPT with search, Perplexity, Google's AI Overviews and AI Mode, Claude and Gemini with browsing. |
| Citation | A link the assistant shows as a source for its answer. Cited means your page was read, not that you were recommended. An assistant can cite your page and still suggest a competitor. |
| Mention rate | The share of answers in a fixed test that name your product. Our free check reports it as "named in X of 75 answers", and per assistant. It is our headline number because single answers do not repeat. |
| Share of voice | Your share of all brand mentions across a set of answers. Five products named and you are one: a fifth of the voice in that answer. In one vendor study, shoppers' chosen brands had 24% share of voice, rejected ones 11% (Profound). |
| Fan-out | The web searches an assistant runs on its own before answering one prompt. ChatGPT searched in about 31% of prompts, about 2.17 searches each, often adding "reviews", "comparison" or the year (Nectiv). Missing from those searches usually means missing from the answer. |
| Grounding (live search) | Reading real web pages before answering, instead of answering from training memory alone. Most of today's product recommendations come from grounded answers. So being in live search results matters more than being in the model's memory. |
| Earned media | Pages about you that you did not write or pay for: reviews, "best X" lists, articles, Reddit threads, videos. For US software questions, 72.7% of the sources AI search cited were earned media (Chen et al.). It is the main input to recommendations. |
| Listicle | An article built as a list: "The 10 best invoicing apps for freelancers in 2026". Listicles were 43.8% of the page types ChatGPT cited in one 26,283-URL study (Ahrefs). Getting into the right ones, honestly, is a large part of our project work. |
| llms.txt | A plain-text file at /llms.txt listing a site's key pages with one-line descriptions, proposed as a courtesy for AI agents (llmstxt.org). It does not affect whether assistants recommend you. Studies found no correlation, and most files are never requested. We have one anyway, and say so on it. |
| Structured data (JSON-LD) | A small block of machine-readable facts on a page. This is an article, by this author, on this date. This is an organisation with this email. It helps machines understand a page. It must match what is visible. No special version makes assistants cite you. |
| Crawler | The program a search engine or AI company uses to fetch web pages. Each has a name you can allow or block in robots.txt: OAI-SearchBot (ChatGPT search), GPTBot (OpenAI training), ClaudeBot and Claude-SearchBot (Anthropic), PerplexityBot, Google-Extended (Google AI training). Block a search crawler and you vanish from that assistant's live answers. |
| Hallucination | An assistant stating something false with confidence. A price you never charged, a feature you do not have, a competitor's description under your name. It usually comes from stale or missing sources. That is why branded prompts ("what is [brand]") are part of every check. |
| Re-check | Running the same fixed prompts again, the same way, at a later date, and comparing mention rates. We use day 30, 60 and 90. A re-check with a changed prompt set is not a comparison. |
| Fixed prompt set | The exact list of buyer questions agreed before any work starts, kept unchanged so before-and-after numbers are fair. New prompts get new numbers. Old ones are never edited. |

Related
- How we measure (/how-we-measure): the exact method behind the mention rate.
- How AI decides what to recommend (/guides/how-ai-decides-what-to-recommend): the longer explanation, with sources.
- Prompt library (/prompts): the prompts behind a fixed prompt set.

Sources
- Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 (origin of the term GEO) (https://arxiv.org/html/2311.09735v2)
- llms.txt proposal (https://llmstxt.org/)
- OpenAI, bots and crawlers documentation (https://platform.openai.com/docs/bots)
- Nectiv, ChatGPT search prompt data (fan-out), via Search Engine Land (https://searchengineland.com/chatgpt-search-prompts-data-463407)
- SparkToro, AIs are highly inconsistent when recommending brands (https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/)
- Chen et al., University of Toronto, earned media in AI search, 2025 (https://arxiv.org/html/2509.08919)
- Ahrefs, best lists research: 43.8% of cited page types (vendor) (https://ahrefs.com/blog/best-lists-research/)
- Google, structured data introduction (https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data)
- Profound, The shortlist is the new shelf (share of voice; vendor) (https://www.tryprofound.com/blog/the-shortlist-is-the-new-shelf)