AI Brand Visibility Tracker — ChatGPT, Claude, Perplexity, Gemini

37% of consumers now start product research in AI tools instead of Google. This Actor asks the AI assistants the questions your customers ask them — and tells you, per answer, whether your brand was mentioned, where it ranked against competitors, and which sources the AI cited.

What you get

One flat, CSV-ready row per (provider × question):

Field	Description
provider, model	Which AI answered (openai / anthropic / perplexity / gemini)
prompt	The question asked
mentioned	Was your brand in the answer?
mentionCount, firstPosition	How often and how early
rankAmongTracked	1 = your brand was named before every tracked competitor
competitorsMentioned, competitorMentionCounts	Who else the AI recommended
citedUrls, citedDomains	The sources behind the answer — the pages to get featured on
answerText	The full answer, for auditing
askedAt	ISO timestamp

Plus a SUMMARY in the key-value store: overall visibility %, average rank when mentioned, per-provider split, and competitor share of voice — one object a scheduled run can diff week over week.

Sample output

{
  "provider": "perplexity",
  "model": "sonar",
  "prompt": "What are the best note-taking apps?",
  "mentioned": true,
  "mentionCount": 2,
  "rankAmongTracked": 1,
  "entitiesDetected": 3,
  "competitorsMentioned": "Evernote,Obsidian",
  "competitorMentionCounts": "Evernote:1,Obsidian:1",
  "citedDomains": "zapier.com,pcmag.com,reddit.com",
  "answerLength": 1874,
  "askedAt": "2026-09-02T09:00:00.000Z"
}

How it works

1. Give it your brand, your competitors, and either a topic (auto-generates a 5-question panel: best / recommend / compare / choose / list) or your own custom questions.
2. Add your own API keys for any of OpenAI, Anthropic, Perplexity, Gemini. Keys are stored as Apify secrets and used only to run your queries — this Actor does not resell tokens, so you pay the providers their real price, not a markup.
3. Every question is asked to every provider you enabled. Answers are analyzed deterministically — mention detection is whole-word and case-insensitive, rank is order of first appearance among tracked names.

Typical setups

• Agency weekly report — schedule it weekly per client brand; diff SUMMARY to show visibility moving.
• Citation-source mining — citedDomains across Perplexity answers is the list of pages to pitch or optimize for AI search (GEO).
• Competitor watch — leave brandName as your client and load competitors with the category; share of voice comes out per run.

Honest limitations

• Only Perplexity searches the live web and returns citations. ChatGPT, Claude and Gemini (as called here, without browsing) answer from training data — that's still what their hundreds of millions of users see, but it moves slowly. The output labels every row by provider so you never mix the two up.
• LLM answers vary between runs by nature. Visibility % over a panel of questions is meaningful; a single answer is not. For trend data, keep the same inputs and schedule the run.
• Mention detection is exact whole-word matching on the name and your aliases. Add misspellings or old names as aliases if the AI uses them.

Cost control

• You bring the API keys, so LLM cost is yours at provider prices — the defaults (gpt-4o-mini, sonar, gemini-2.0-flash, claude-sonnet-5) are the cheap tiers, roughly a cent or less per answer.
• maxItems is a hard cap on analyzed answers, strictly enforced.
• A rejected API key fails the run immediately with a readable message — not after burning your quota.