AI Brand Visibility Tracker: ChatGPT, Gemini & Perplexity (GEO / AEO)

Find out how often AI assistants recommend your brand when buyers ask questions, compared with your competitors, and which sources the AI cites. Each prompt is asked several times per engine, because AI answers change on every run. You get a mention rate with a confidence interval instead of a fake "AI rank".

Why bring your own key? You pay the AI provider directly at cost (typically a fraction of a cent per answer) instead of a bundled markup, and your prompts go only to the provider you choose. Fastest start: a free Google AI Studio key (aistudio.google.com/apikey, about a minute, no card) runs the Gemini engine. Add OpenAI or Perplexity keys for live web-search answers.

First, check AI bots can read your site at all. If ChatGPT's or Perplexity's crawler is blocked by robots.txt or your CDN, they can't cite you no matter what you publish. Run AI Crawler Access Audit (no key, $0.02 per page) before tracking mentions.

Sample output

One record per prompt × engine, plus one summary record. This is real output, recorded on 2026-10-17 with Gemini (free key, so no web search). It's the same sample the Actor returns in demo mode when no API key is given.

{
  "type": "check",
  "prompt": "What is the best note-taking app for a small team?",
  "engine": "gemini",
  "model": "gemini-3.1-flash-lite",
  "webSearch": false,
  "runsRequested": 2,
  "runsSucceeded": 2,
  "brand": "Notion",
  "brandMentioned": 2,
  "mentionRate": 1,
  "mentionRate95": {
    "low": 0.342,
    "high": 1
  },
  "avgRankWhenMentioned": 1,
  "mentionSnippets": [
    "op four recommendations based on how your team operates: --- ### 1. Notion: The \"All-in-One\" Workspace **Best for:** Teams that want to combine"
  ],
  "domainCitedRate": 0,
  "citedUrls": [],
  "topCitedDomains": [],
  "competitors": [
    {
      "name": "Obsidian",
      "mentions": 2,
      "mentionRate": 1,
      "avgRank": 2,
      "citedRate": 0
    },
    {
      "name": "Evernote",
      "mentions": 0,
      "mentionRate": 0,
      "avgRank": null,
      "citedRate": 0
    },
    {
      "name": "Coda",
      "mentions": 2,
      "mentionRate": 1,
      "avgRank": 3,
      "citedRate": 0
    }
  ],
  "shareOfVoice": {
    "Notion": 0.333,
    "Obsidian": 0.333,
    "Evernote": 0,
    "Coda": 0.333
  },
  "sampleAnswer": "There is no single \"best\" app, because the right choice depends on your team’s **workflow culture**. Do you need a structured database, a free-flowing collaborative canvas, or a project management hub?\n\nHere are the top …",
  "checkedAt": "2026-10-17T18:46:27.161Z"
}

Summary record:

{
  "type": "summary",
  "brand": "Notion",
  "domain": "notion.so",
  "prompts": 2,
  "checks": 2,
  "checksWithResults": 2,
  "answersAnalyzed": 4,
  "mentionRate": 1,
  "mentionRate95": {
    "low": 0.51,
    "high": 1
  },
  "domainCitedRate": 0.25,
  "engines": [
    {
      "engine": "gemini",
      "model": "gemini-3.1-flash-lite",
      "webSearch": false,
      "checks": 2,
      "runs": 4,
      "mentionRate": 1,
      "mentionRate95": {
        "low": 0.51,
        "high": 1
      },
      "domainCitedRate": 0.25
    }
  ],
  "shareOfVoice": {
    "Notion": 0.333,
    "Obsidian": 0.333,
    "Evernote": 0.167,
    "Coda": 0.167
  },
  "topCitedDomains": [
    {
      "domain": "getupnote.com",
      "count": 1
    },
    {
      "domain": "onenote.com",
      "count": 1
    },
    {
      "domain": "obsidian.md",
      "count": 1
    },
    {
      "domain": "logseq.com",
      "count": 1
    }
  ],
  "promptsWhereBrandIsAbsent": [],
  "howToRead": "(explains each metric)",
  "errors": [],
  "checkedAt": "2026-10-17T18:46:42.703Z"
}

What it measures

Field	Meaning
mentionRate, mentionRate95	Share of answers that name your brand (whole-word match, plus aliases), with a 95% Wilson interval
avgRankWhenMentioned	Order of first mention among your brand and the tracked competitors. Noisy, so read it as a hint only
mentionSnippets	The text around each match, so you can audit it
domainCitedRate, citedUrls, topCitedDomains	How often your domain is cited or linked, and which sites the AI used as sources
competitors, shareOfVoice	Mention rate per competitor, and each brand's share of all tracked mentions
webSearch	true when the engine searched the web (ChatGPT search, Google Search grounding, Perplexity), false when it answered from model memory
promptsWhereBrandIsAbsent (summary)	The prompts where you are missing and which competitors show up instead

How to use

1. Enter your brand, domain and 2–5 competitors.
2. Add the buyer prompts your customers would type, e.g. "best price tracker for Wallapop", or just describe what you sell and let the Actor generate prompts in your language.
3. Paste at least one API key (Gemini, OpenAI or Perplexity) in the API keys (bring your own) section. Keys are encrypted secret inputs, sent only to that provider, and never written to the output.
4. Set runs per prompt (3 is a good start, 5–10 for tighter intervals) and run. Schedule it weekly to track the trend.

Without any key, the Actor returns the clearly labelled demo sample ("demo": true, not charged) for the sample brand. For any other brand it stops with a message explaining how to add a key.

Engines

• OpenAI: Responses API with the web_search tool (default model gpt-4.1-mini, the closest to ChatGPT Search). Optional country sets the search location.
• Google Gemini: Google Search grounding (default gemini-flash-latest). Note that free AI Studio keys usually can't use grounding on current models. The Actor then falls back to a no-search answer and marks webSearch: false. Turn this off with allowUngroundedFallback.
• Perplexity: sonar with its citations and search results.

Pricing

$0.04 per prompt × engine check, and all repeated runs of that prompt on that engine are included. You pay the AI provider for the calls with your own key, usually fractions of a cent each. Failed checks (bad key, provider errors) and demo output are not charged. Example: 10 prompts × 3 engines × 3 runs = 30 checks = $1.20 on Apify.

Tips

• Visibility is a rate, not a rank. SparkToro's 2026 study found AI tools almost never return the same list twice, so measure many prompts, several runs each.
• Add your city or country to some prompts if you sell locally. Local prompts often surface smaller brands.
• topCitedDomains shows where AI looks. Being mentioned on those sites (reviews, directories, comparisons, forums) usually moves visibility more than on-page tweaks.
• If your brand is also a common word, turn on case-sensitive match and check mentionSnippets.

Limits

API answers can differ from the consumer apps (personalization, ads, model versions), so treat results as a consistent benchmark, not a screenshot of what every user sees. There are at most 50 prompts and 10 runs per prompt per run.