AI Visibility
Track how AI assistants describe your brand.
People increasingly form opinions by asking an AI assistant, not by reading review pages. AI Visibility measures what those assistants actually say about you: it puts a set of prompts to the models you choose, brings the answers back into a Feedback Group, and scores how visible, and how well described, your brand is in them.
AI Visibility is its own source type, alongside surveys, uploads, connectors, and Social Listening agents. It is created through the same agent wizard and runs on the same Web & Social Media Credits, but what it collects and the analytics built on top are distinct, so it appears as its own category across the app.
Access
AI Visibility is an add-on enabled per organisation by the BAI Analytics team, on top of Web & Social listening (it needs the same agent infrastructure). Without it, the AI Visibility entry points show Feature not enabled with a note to contact your BAI Analytics representative. Creating, editing, and running AI Visibility agents requires the add-on; answers already collected stay visible in full even if it is later switched off.
Supported assistants
Claude, ChatGPT, Gemini, Perplexity, Grok, DeepSeek, and Mistral. All seven are pre-selected (and badged Recommended) when you create an agent, and you can trim the list. Each tile carries a one-line description of who uses that assistant. Under each assistant you can also pick the Model versions to query (up to four per assistant); every selected version runs, and is billed, separately.
One agent can query several assistants at once, which is the point: they often disagree, and you want to know which ones tell your story well and which ones don't.
When to use it
You want to know whether an assistant brings you up on its own when a prospect asks "what are the best project-management tools?", without the question ever naming you.
You want to know what an assistant says when asked about you directly: "what is the online perception of Brand X?"
You want your share of voice against competitors inside AI answers, not just in search results.
You want to know which websites shape those answers, and how often your own site is one of them.
You want to catch wrong or outdated claims that assistants repeat about your product.
You want a trend: re-run the same prompts on a schedule and watch your visibility move as your content and coverage change.
Two kinds of questions
Every AI Visibility agent asks each assistant two kinds of questions, and the workspace keeps them apart because they measure different things:
Market questions never name your brand ("Which options should I consider among low-cost airlines in Europe?"). They are the only honest way to measure visibility: if the assistant mentions you here, it chose to. Visibility, prominence, share of voice, and cited sources are all computed on these answers.
The branded question names you ("What is the online brand perception of Brand X?"). Its answers feed the perception side: summary, themes, tone, and the AI answers you can read in the Feed.
Counts differ between the two, by design: the tiles say which scope they measure.
Setting up an AI Visibility agent
Create an agent from the AI Visibility category (or pick one or more AI assistants in the Sources step of a regular agent); the wizard switches to a brand-first flow:
Your brand: the name the prompts should ask about. Answers are scanned for it, including acronyms and common short forms.
Competitors (optional): naming them unlocks share of voice and the Who AI recommends comparison.
Your website (optional): lets BAI Analytics measure how often AI answers cite your own site as a source.
Measure GEO (AI visibility): on by default. This adds the unbiased market questions to every model; turn it off if you only want the branded perception answers.
GEO: your market category (Set Behavior step): BAI Analytics identifies the market you compete in from your brand (say, "project-management software"). Adjust it if the suggestion is off, or Regenerate. The wizard writes the market-question set from it: 3, 4, or 5 phrasings depending on intensity, shown to you before saving, all sharing one answer budget so more phrasings never cost more.
The branded question is built for you as well. Power users can choose Write custom prompts instead and supply their own prompts (one per line); brand tracking then becomes an optional add-on to those prompts.
Everything else works like any other agent: choose Run Once, Scheduled, or Ongoing, and an Answer volume tier. For AI assistants the tiers are 12 / 25 / 50 answers per prompt, per model (Standard / High / Extra high). Runs deduct Web & Social Media Credits, and failed runs refund automatically.
What comes back
Each answer is stored as a response, with metadata you can segment by:
The assistant's answer as the response text.
A verbatim source snippet: text pulled from the web page the assistant relied on, never the assistant's own paraphrase, so every claim is traceable. Citations resolve to the actual page the assistant used, not a search-engine redirect.
The model and version, the prompt that produced the answer, and the cited domain.
Brand Mentioned, Mention Position (early / middle / late in the answer), and Competitors Mentioned, computed for every answer when brand tracking is on.
The raw-data tab of an AI Visibility source is labelled AI answers rather than Feed.
The AI Visibility workspace
The qualitative overview of an AI Visibility source is a dedicated workspace, {Your brand} in AI, built to answer three questions: does AI recommend you, what does it say, and what should you do next.
Visibility (unbiased market questions)
A verdict headline (from Strong and prominent down to Low visibility) sits above four tiles:
Visibility: the share of market answers naming your brand, with the exact count ("Mentioned in 42 of 90 answers to market questions that never name your brand").
Prominence: how early you appear when you are mentioned.
Share of voice: your mentions versus tracked competitors. Locked until you add competitors on the agent.
Your site cited: how often answers with sources cite your own website, as a share of answers that cited anything. A hint shows the same rate on the branded question. Locked until you add your website on the agent; shows a dash while fewer answers cite a source than the metric needs.
On a period page of a tracked group, the scorecard is scoped to that period only and says so.
What AI says (direct questions about your brand)
A perception card carries the summary of the branded answers, the top themes, and an Overall tone bar (positive / neutral / negative). Explore themes opens the full thematic analysis.
What to do next
Three action cards are chosen from your actual numbers, for example Review missing mentions (the models and answers creating the gap), Improve mention position, Strengthen first-party authority, Add your website, or Monitor strong visibility. Each one jumps to the evidence it refers to, or straight to the wizard field it asks you to fill in.
What drives your result
A tabbed evidence panel:
AI models: per-model visibility, mentioned-early rate, and answers with sources, ranked, with Strongest and Biggest opportunity badges when the spread is meaningful. Summary & answers on a model opens a short AI summary of how that model presents your brand, then every market answer it gave, each with its prompt, a Mentioned / Not mentioned badge with position, the competitors it cites, and the domain it sourced. Filter by mention status, competitor, or domain, or open all models together.
Competitors: Who AI recommends, one table with your brand (marked you) and each competitor, with Visibility, Prominence, and Share of voice columns and the raw fractions inline. Click a row to read the answers citing that competitor.
Cited sources: Where AI models source their answers, the top domains as a share of answers with a citation (your own domain marked yours). Click a domain to read the answers sourced from it. These sites shape what AI models say about you; presence there moves your visibility.
Trend: Visibility rate over time and Share of voice over time, one line per model plus Overall. It builds as the agent re-runs; periods follow the Feedback Group's tracking periods when it has them, weekly otherwise.
If the agent has not run yet, or brand tracking or GEO measurement is off, the workspace says so and offers Configure brand tracking or Enable on the agent, which opens the wizard on the right field.
Where it fits in the rest of BAI Analytics
AI Visibility sources are listed under their own category in the feedback hub and on Feedback Group pages, separate from Web & Social sources. Beyond that they behave like any source: answers feed Grouped Themes, monitors with auto-coverage on watch them (Custom Monitoring), and they roll into reports. Put one in an evolution-tracked group and re-run it on a schedule to turn the scorecard into a period-over-period trend.
Best practices
Name your real competitors. Share of voice is the scorecard's most decision-ready number, and it only exists if the wizard knows who to compare you against. Pick brands that are genuinely different from yours; a look-alike name is never counted as you.
Add your website. It is the only way to see whether assistants lean on your own content or on third-party pages when they describe you.
Keep the prompt set stable when you want a trend. Changing the market category or prompts between runs changes what's being measured; add new prompts as a second agent instead.
Start at Standard intensity. 12 answers per prompt is usually enough to see the pattern; scale up when you need statistical comfort, not before.
Read the cited domains. If assistants keep citing a stale comparison page or an old review thread, that page, not the assistant, is what needs fixing.
Last updated