AI visibility: a working definition

AI visibility describes how a brand shows up inside AI-generated answers. It breaks into four parts:

• Mention: the answer names the brand, with or without a link.
• Citation: the answer links to or attributes a source page, either yours or a third party's that talks about you. See citations vs mentions for why the two behave differently.
• Accuracy: the facts stated about the brand (price, products, location, ownership) are correct.
• Sentiment: the tone of the description, positive, neutral or negative.

A brand can score well on one part and badly on another. Being named often with the wrong pricing is a visibility problem, not a win.

An AI answer passes through four stages, and each one affects visibility differently.

1. Parametric: what the model learned in training data, frozen at a cutoff date. Nothing you publish this quarter reaches these answers.
2. Access: whether a crawler may fetch your page. robots.txt rules are per bot. OpenAI states that GPTBot covers training while OAI-SearchBot controls ChatGPT search, and that each setting is independent.
3. Retrieval: whether your page enters the model's context for a given query. As of October 2026, retrieval is per query and discretionary. OpenAI says ChatGPT search rewrites a prompt into one or more targeted queries before searching.
4. Generation: which retrieved sources the model uses and cites.

The practical upshot: you control stage 2, you influence stage 3 indirectly, and you don't control stages 1 or 4.

Engines don't share an index or a model. Google says AI Overviews and AI Mode "may use different models and techniques" even inside Google Search. To appear as a supporting link in either, a page must be indexed and eligible to show with a snippet.

The differences can be large. In OtterlyAI's 2026 experiment with AI-generated sites, Microsoft Copilot produced 2,718 of the 3,934 citations tracked for one site, while ChatGPT and Google's AI surfaces almost never cited it. Google had deindexed the site and Bing had not.

A blended number would have hidden that split. That's the case for per-platform scores over one blended score. It's also why answers from an API can differ from what the consumer app shows, covered in API answers vs consumer app answers.

• "There's an AI rank to track." Google's own AI reporting shows impressions only, with no position data. No ground truth exists for a rank number.
• "Cited means used." OpenAI's tools return more retrieved sources than cited ones, and mechanistic research shows citation output can be changed without changing the sources. Citation counts are a noisy proxy for influence.
• "Schema or llms.txt gets you cited." Google says structured data isn't required for its generative AI features and that Google Search ignores llms.txt. Neither is a lever for AI answers.
• "Only your own site matters." When engines cite other sites, being named on those sites is associated with appearing in the answer. See your own site vs third-party mentions.
• "Tools can stop wrong answers." They can find wrong facts. They can't edit the model. Read whether tools prevent hallucinations.

LLM visibility and AI search visibility are near-synonyms. LLM visibility often stresses what the model knows from training; AI search visibility stresses answers built from live retrieval. For measurement across models, prompts and engines, read the full guide to LLM visibility tracking.

Is AI visibility the same as SEO? No, but they overlap. Google's AI features use the same index and snippet eligibility as Search. Other engines use their own crawlers and indexes, so a page that ranks well in Google can still be blocked from ChatGPT search by a robots.txt rule or a firewall. SEO covers ranked links. AI visibility covers whether a generated answer names, cites and correctly describes you.

How do I check my brand's AI visibility? Ask the questions your buyers ask, on each engine separately, logged out, and record mentions, citations and any wrong facts. The free checker on this site runs up to three buyer questions and scores the answers. Treat any result as a snapshot: retrieval fires per query, and answers change as models and indexes update.

What can I actually control? Mostly access. Check that robots.txt allows each retrieval bot, such as OAI-SearchBot, PerplexityBot and Claude-SearchBot. Confirm your CDN or firewall doesn't return 403 to them. Put key facts in the HTML before JavaScript runs. Avoid accidental noindex or nosnippet tags. You can't control what a model says or whether it cites you.