AI Brand Visibility Tracker – ChatGPT, Claude, Gemini

AI Brand Visibility Tracker: track brand mentions in ChatGPT, Claude, Gemini & Perplexity

Find out whether AI assistants recommend your brand, or your competitors. Give it the questions your customers ask AI ("What's the best CRM for a small business?"), your brand, and your competitors. The Actor asks each AI model, detects every brand in each answer, and builds a visibility report for each model:

• Visibility score (0–100), mention rate, and share of voice against your competitors
• Rank (which brand is mentioned first) and list position in "top 10" style answers
• Sentiment around each brand, and cited domains (does the AI link to your site?)
• The prompts where your brand is missing, which is your GEO to-do list
• Other names in lists: products the AI recommends that you aren't tracking yet
• Change vs the previous run, plus a 30-run history, for weekly or daily tracking

Bring your own API keys, with no markup on AI costs. You pay OpenAI, Anthropic, Google, Perplexity, OpenRouter or xAI directly, at their normal API prices, usually under $0.001 per answer for small models. This Actor charges only for the analysis. Most AI visibility tools bundle the AI cost and charge $0.08–$0.30 per check. Here it's $0.005 per answer.

What can this AI visibility tracker do?

• 🔑 Official APIs, no scraping of chat UIs. No proxies, captchas, or broken runs when a website changes.
• 🎲 Statistically honest. AI answers change from one run to the next. Ask each prompt 3–5 times (samplesPerPrompt) to get a real mention rate instead of a one-off screenshot.
• 🌐 Many models in one run: compare ChatGPT, Claude, Gemini, Perplexity, Grok, Llama, Mistral, DeepSeek and others side by side. One OpenRouter key covers most of them.
• 🔎 Optional web search and grounding, to see what AI search cites: the OpenAI web search tool, Anthropic web search, Gemini Google Search grounding, the OpenRouter web plugin, and Perplexity.
• 📈 Tracking built in: reports compare against the last run with the same trackingId. Schedule it weekly and watch your score move.
• 🔒 Keys are secret inputs: encrypted by Apify, never logged, never written to results.
• 💸 Failed AI calls are free. A bad key, an unknown model, or a rate limit never costs you an event.

How to track your brand in AI answers (GEO / AEO)

1. Add your brand in the form Name | alias | domain.com (e.g. HubSpot | HubSpot CRM | hubspot.com).
2. Add competitors in the same format, one per line.
3. Add prompts: real buyer questions. Keep them the same between runs so the trend stays meaningful.
4. Add models as provider:model-id, using any model id your provider currently lists:
   • openrouter:<vendor>/<model>: one OpenRouter key covers them all. Cheap current ids we tested (Nov 2026): openrouter:openai/gpt-6-luna, openrouter:anthropic/claude-haiku-5.5, openrouter:google/gemini-3.5-flash-lite, openrouter:x-ai/grok-4.3, openrouter:perplexity/sonar
   • openai:<model>, anthropic:<model>, gemini:<model>, perplexity:<model>, xai:<model> with that provider's own key
   • custom:<model> with customBaseUrl for any OpenAI-compatible endpoint (Groq, Together, Azure-compatible gateways, your own proxy)
5. Paste the API key for each provider you use. Run it.

Example input

{
  "brand": "HubSpot | HubSpot CRM | hubspot.com",
  "competitors": ["Salesforce | salesforce.com", "Zoho CRM | Zoho | zoho.com", "Pipedrive | pipedrive.com"],
  "prompts": ["What is the best CRM for a small business?", "Which CRM should a 10-person startup use?"],
  "models": ["openrouter:openai/gpt-6-luna", "openrouter:anthropic/claude-haiku-5.5", "openrouter:google/gemini-3.5-flash-lite", "openrouter:x-ai/grok-4.3"],
  "openrouterApiKey": "sk-or-...",
  "samplesPerPrompt": 3,
  "webSearch": false,
  "trackingId": "hubspot-crm-us"
}

Output

Dataset items have an itemType. The Console views are Visibility reports, AI answers and Answer text.

report: one per model, plus all-models when you track several (trimmed, illustrative values):

{
  "itemType": "report",
  "model": "openrouter:openai/gpt-6-luna",
  "brand": "HubSpot",
  "answers": 6,
  "brandVisibilityScore": 72.5,
  "brandMentionRate": 0.833,
  "brandShareOfVoice": 0.31,
  "brandAvgRank": 1.4,
  "brandCitationRate": 0.0,
  "brandLeaderboardPosition": 1,
  "leaderboard": [{ "name": "HubSpot", "visibilityScore": 72.5, "mentionRate": 0.833, "shareOfVoice": 0.31 }],
  "promptsWhereBrandMissing": ["Which CRM should a 10-person startup use?"],
  "otherNamesInLists": [{ "name": "Freshsales", "answers": 3 }],
  "topCitedDomains": [{ "domain": "g2.com", "answers": 2 }],
  "previousBrandVisibilityScore": 64.0,
  "brandVisibilityScoreChange": 8.5,
  "providerCostUsd": 0.0021
}

answer: one per AI answer. It includes prompt, sample, brandMentioned, brandRank, brandListPosition, brandSentiment and brandCited. It also includes brandsMentionedInOrder, a per-competitor breakdown (competitors), citedDomains and citationUrls. Finally there's otherNamesInLists, the full answerText, token usage (with costUsd when the provider reports it, e.g. OpenRouter, and truncated when the answer hit maxAnswerTokens) and latencyMs.

info: written only when no API key is given. The run still succeeds, charges nothing, and explains which key field to fill in.

error: a failed AI call with its reason (auth_error, model_not_found, rate_limited, and so on). These are never charged.

How the visibility score works: 60% mention rate + 25% rank score (1 / rank, 0 when not mentioned) + 15% citation rate, scaled to 0–100. Citations only exist when answers carry sources (web search on, or Perplexity), so without them the top possible score is 85. brandRank is the order of first mention among your tracked brands; brandListPosition is the position in the answer's main recommendation list (bullets, numbered headings or a comparison table), empty when the brand is only mentioned outside it. Sentiment is a fast word-list heuristic over the brand's own list section and the sentences naming it, so treat it as a signal, not a verdict. Reports with webSearch on keep a separate history from plain runs, so the change vs last run compares like with like.

How much does AI brand visibility tracking cost?

Pay per event. AI usage goes on your own provider bill, at cost.

Event	Price
AI answer analyzed	$0.005
Visibility report (per model, plus a combined one)	$0.02
Failed AI call	free

Examples:

• 10 prompts × 3 samples × 1 model = 30 answers + 1 report = $0.17, plus roughly $0.01–0.05 at your AI provider for small models
• 20 prompts × 3 samples × 4 models = 240 answers + 5 reports = $1.30 a run. Weekly, that's about $5.60 a month.

If you set a maximum cost per run, leave room for the reports: answers × $0.005 + reports × $0.02.

Limits and good to know

• These are API answers, not the consumer chat apps. Consumer apps add memory, personalization and their own system prompts. Use systemPrompt, locationHint and webSearch to get closer to what users see. API answers are the reproducible baseline most GEO teams track.
• Brand detection matches names, aliases and domains you give it, with whole-word, case-insensitive matching. Add aliases for common spellings. Very generic brand names (e.g. "Notion", "Apple") can produce false positives.
• Model ids change often. Use the ids your provider lists today, or OpenRouter's catalog.
• Some newer models accept only the default temperature, so leave temperature empty unless you need it.
• Web search makes answers slower and costs more at your provider.

Use cases

• GEO / AEO (generative engine optimization): find the buyer questions where AI assistants skip your brand, then fix your content and check again.
• Competitive intelligence: a share-of-voice leaderboard of you vs competitors, for each model.
• Agency reporting: a weekly AI visibility report per client, with change vs last week.
• Launch monitoring: see when a new product starts showing up in LLM answers.
• Model comparison: compare how ChatGPT, Claude, Gemini, Perplexity and Grok talk about your category.

FAQ

Do I need an API key? Yes, for the providers you want to test. One OpenRouter key covers most models. Without a key, the run finishes successfully, charges nothing, and tells you which key field to fill in.

Is this an LLM visibility checker or a rank tracker? Both. It checks whether you're mentioned (mention rate), where (rank and list position), how (sentiment), and whether your site is cited. Run it on a schedule and it becomes an AI rank tracker with history.

Why are results different every run? AI answers are probabilistic. Use samplesPerPrompt of 3–5 for stable rates, and keep prompts the same between runs.

Does it check Google AI Overviews? No. It uses AI model APIs. For Google AI Overviews, use a dedicated AI Overviews scraper.

Legal and data

The Actor sends your prompts to the AI providers you choose, under your own API accounts and their terms. It doesn't scrape any website or chat interface. Don't put confidential data in prompts.

Related signalwharf Actors

• Telegram Channel Scraper: scrape public Telegram channel posts, views, reactions and media without login.
• Substack Scraper: Substack posts with full text, comments, subscriber estimates and paid plan prices.
• AI Brand Visibility Tracker: track how often ChatGPT, Claude, Gemini and Perplexity mention your brand vs competitors.