AI visibility glossary
The words used when measuring how AI assistants talk about brands, in plain English. Each entry links to where the docs or our method go further.
- AI crawler
- Also called AI bot
- A program that fetches web pages for an AI company, to train models or to find pages for answers. Examples include GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic and PerplexityBot from Perplexity. Sites can allow or block each one in robots.txt.
- Free AI crawler checkerThe crawler check in Ocita
- AI referral
- A visit to a website from a link in an AI assistant's answer, such as from chatgpt.com or perplexity.ai. It shows up in web analytics as a referral from that assistant's site.
- Engine guides
- AI visibility
- How often, and how prominently, a brand appears in the answers AI assistants give to the questions its buyers ask. It's usually measured by asking an assistant a set of prompts and counting the answers that mention the brand.
- Visibility in OcitaHow Ocita measures
- Alias
- A name variant that counts as a brand when answers are scored: a short name, a former name, a product name or a domain.
- Competitors and aliases
- Answer engine
- Also called AI search engine
- A service that replies to a question with a written answer rather than a list of links, often citing the web pages it drew on. ChatGPT with search, Gemini and Google AI Mode are examples.
- Engine guides
- Bootstrap
- A way to estimate a range by resampling the data many times and recalculating the result each time. Ocita resamples prompts first, then answers within each prompt, so the range reflects which prompts were tracked as well as how answers vary.
- How ranges are calculated
- Branded prompt
- A prompt that names the brand being measured. Answers to it nearly always mention the brand, so a set with many branded prompts overstates visibility.
- Branded and non-branded prompts
- Citation
- Also called Source
- A link an AI answer gives to a web page it drew on. Citations show which sites shape an engine's answers in a category.
- SourcesHow Google AI Overviews cite
- Citation rate
- The share of answers that link to at least one page on a brand's own site. Ocita works it out over answers that carry citations and counts answers with no sources separately.
- Citation rate in Ocita
- Confidence interval
- Also called Range, Margin of error
- A range around a measured value that shows how far it could reasonably sit, given how much the sample varies. A 90% interval is built so that, over many repeated measurements, nine in ten such ranges would contain the true value. A narrow range means a steadier number.
- How ranges workHow ranges are calculated
- Gap analysis
- A list of the sites cited in answers that name competitors but never in answers that name you. They already write about the category, so they're the pages most worth approaching.
- Gap analysis in OcitaSources
- Generative engine optimization (GEO)
- Also called Answer engine optimization, AEO, LLM optimization
- Work aimed at making a brand appear, and be described accurately, in AI-generated answers. It overlaps with SEO: crawlable pages, clear facts about what you offer, and being covered by the sites answers cite.
- Engine guidesFinding sources to approach
- Grounding
- Also called Search grounding, Retrieval
- Letting a language model look things up, usually with a web search, before it answers, and base the answer on what it found. Grounded answers can cite sources and reflect recent pages.
- How Gemini grounds answersWhere Ocita's answers come from
- llms.txt
- A proposed convention: a Markdown file at /llms.txt that gives language models a short summary of a site and links to its most useful pages. It's a proposal rather than an agreed standard, so whether a given AI service reads it is up to that service.
- Ocita's own llms.txt
- Mention
- An answer that names a brand. Trackers decide this by matching the brand's name and its variants, which is why aliases matter.
- AliasesHow answers are scored
- Position
- Also called Average position, Rank in answer
- Where a brand appears when an answer lists several options, such as third in a numbered list. Averaged across answers that place the brand in a list; lower is better.
- Average position in Ocita
- Prompt
- The question sent to an AI assistant. In visibility tracking, a prompt stands for something real buyers ask, such as "best payroll software for a small business". The set of prompts you track is your sample, so the numbers describe those prompts and not every possible question.
- Choosing prompts
- Real gain and Real drop
- Ocita's terms for a change that clears the noise: this week's range doesn't overlap the range from the previous four weeks. If the ranges overlap, the move is reported as no real change.
- Real gain, Real drop and Directional
- robots.txt
- A text file at the root of a site that tells crawlers which paths they may fetch, crawler by crawler. Well-behaved crawlers follow it, but it's a request, not a lock. Google-Extended is a robots.txt token rather than a crawler: it controls whether Google may use pages for Gemini, without affecting Google Search.
- Check a site's robots.txtThe OpenAI crawlersWhat Google-Extended controls
- Sampling
- Also called Repeats
- Asking the same prompt several times and measuring across the answers. AI assistants don't give the same answer twice, so one check can mislead; repeated samples show the pattern and how much it varies.
- How answers are sampled
- Stable and Directional
- Ocita's labels for how much to trust a number. Stable needs enough prompts, a narrow range and a finished first weekly run; anything else is Directional, worth watching but not acting on yet.
- Stable and DirectionalThe exact rules