AI Visibility Tracker: ChatGPT, Perplexity & AI Overviews Rank

AI Visibility Tracker: ChatGPT, Perplexity, Gemini, Claude, Grok & Google AI Overviews

Watch the 25-second film

See what AI assistants and Google's AI Overview answer when your buyers ask a question, which websites they cite, and whether your brand is named or cited, by country. One run asks the same questions of ChatGPT's models (OpenAI GPT), Perplexity, Google Gemini, Anthropic Claude and xAI Grok, each through its official API with live web search, and reads Google's AI Overview for the same question from Google's own results page. You get every answer, every cited source, and per brand and engine a visibility score (0-100), share of voice, sentiment (positive, neutral, negative or mixed, with the sentence it comes from), mention rate, mention rank and citation share for GEO / AEO tracking. Turn on AI search volume to see how often each question is asked in AI search every month.

Always the newest models. Each run uses each company's newest model at that moment, so your tracking follows what people actually get: on 12 October 2026 that was GPT-6.1 Sol, Perplexity Sonar Pro Search, Gemini 3.8 Flash, Claude Sonnet 5.5 and Grok 4.7. You can pin a model under Advanced if you want a fixed one.

Try it in a minute: the form is filled in. Press Start and you get the answers, the sources they cite, a brand report and a monitoring baseline. A first run of one question across three engines costs about $0.18, plus about $0.10 of AI usage at cost.

Then run it every week. Keep the project name and add a schedule (Actor page, then Schedules). Each run shows the mentions you gained and lost since the last one, per question and engine, and which competitors moved. An agency tracking 50 questions on 4 engines every week pays about $20 a week in answers, plus AI usage at cost.

What it does

For every question, engine and run, it returns:

• The full answer text (for Google, the whole AI Overview, including the part behind "Show more").
• Every source the answer cites, in order (position, URL, domain), with redirect links resolved to the real pages.
• For each brand you track: whether the answer names it, its rank among your brands by first mention, whether its site is cited, and at what position.
• The sentiment of each brand mention: positive, neutral, negative or mixed, with the sentence from the answer it is based on and the words that decided it.
• The model that answered, the country used for the search, the time, and the AI cost of that answer.
• With AI search volume on: how often the question is asked in AI search per month in your country.

When the run finishes, the key-value store record SUMMARY starts with the visibility score per brand, overall and per engine, best first, and gives per brand, overall and per engine:

• Visibility score (0-100): one number from mention rate, mention rank and citation share (formula below).
• Share of voice: the brand's share of all mentions of the brands you track.
• Sentiment: how many answers speak of the brand positively, neutrally, negatively or mixed, and a net sentiment from -1 to 1.
• Mention rate: share of answers that name the brand.
• Citation share: share of answers that cite the brand's site.
• Average mention rank and average cited position.
• The most-cited sites overall and per engine, and the source mix (Reddit, Wikipedia, YouTube, marketplaces).
• With AI search volume on: each question's monthly AI search volume, its 12-month history and trend.

Visibility score, share of voice and sentiment

• Visibility score = 100 × (0.60 × mention rate + 0.30 × citation share + 0.10 × first-mention rate), per brand, overall and per engine. Mention rate: share of answers that name the brand. Citation share: share of answers that cite its site. First-mention rate: share of answers that name it first among your brands (its mention rank is 1). Only answers that came back count; failed and free rows are left out. Grok doesn't say which pages it used, so its answers are left out of citation share and its score uses the other two parts in the same proportions.
• Share of voice = the brand's mentions ÷ all mentions of the brands you track, in the same answers (one per answer that names it). Together your brands make 100%.
• Sentiment is read from the answer's own words around each mention, the same way every time, with no extra AI call and no extra charge: the sentences and table rows that name the brand, the lines under a heading or list item that names it (pros and cons), and positive and negative words such as "best", "intuitive", "expensive", "limited", with "not", "no" and "less" turning them round. Each brand gets positive, neutral, negative or mixed per answer, the sentence it is based on (copied from the answer) and the words that decided it. Net sentiment = (positive − negative answers) ÷ answers that name the brand. It reads tone, not facts.

AI search volume (optional)

Turn on AI search volume per question (includeSearchVolume) and each question gets its monthly AI search volume in your country: how often people ask it in ChatGPT, Google AI Overviews and other AI tools, with 12 months of history, the trend and the peak month. It is the same data as our AI Search Volume Actor, from a keyword database's estimate. Use it to see which questions matter most before you work on them. One lookup covers up to 50 questions in one country and costs $0.05; it is made only when your spending limit also covers every answer, so the answers always come first.

Who uses it

• Brands and agencies tracking GEO / AEO: are we recommended when buyers ask AI assistants or see Google's AI Overview, and who is recommended instead?
• SEO teams: which pages Google's AI Overview and the assistants cite, so you know where to be mentioned.
• Local markets: the same question asked as a searcher in the US, the UK, Germany, India, South Africa or anywhere else.

Example

Two questions buyers ask about project management software, asked as a searcher in the US of all six engines, one run each, 12 October 2026. 11 answers: Google showed an AI Overview for one of the two questions (the other was free).

Brand	Named in the answer	Its site cited	Average rank when named
ClickUp	11 of 11	4 of 9	2.1
Asana	11 of 11	4 of 9	2.2
Trello	11 of 11	4 of 9	2.7
monday.com	9 of 11	2 of 9	2.8
Notion	5 of 11	0 of 9	5.0

The engines disagree on who comes first: ChatGPT named Asana first in both answers, Perplexity and Google's AI Overview led with ClickUp, Claude with monday.com. The products' own sites were cited most: Asana's 5 times, ClickUp's and Trello's 4 each, ahead of any review site. Notion was named in under half the answers and never cited. Grok's two answers aren't in the cited column, because Grok doesn't say which pages it used.

Input

{
  "prompts": ["What is the best project management software for a small team?"],
  "brands": ["Asana | asana.com", "Trello | trello.com", "monday.com"],
  "engines": ["openai", "perplexity", "gemini", "claude", "grok", "google-ai-overview"],
  "country": "US",
  "runsPerPrompt": 3,
  "includeSearchVolume": true
}

• prompts: the questions, one per line, as buyers would type them into an assistant or Google.
• brands: a name, a domain, or both as Name | domain.com. With a domain, citations are matched on the site; with only a name, on the name in the address.
• engines: any of openai (ChatGPT's models), perplexity, gemini, claude, grok, google-ai-overview.
• country: ChatGPT, Perplexity, Claude and Grok search as from that country, and Google's AI Overview is read from that country's Google. Gemini's API takes no location. language sets Google's results language (optional).
• runsPerPrompt: assistants' answers change from run to run; use 3 or more for a stable rate. Google's AI Overview is read once per question.
• includeSearchVolume (off by default): each question's monthly AI search volume in your country, $0.05 per lookup of up to 50 questions.
• Advanced: a fixed model for any engine (an OpenRouter model id), answer length and concurrency.

Output

One row per question, engine and run (answer text and citations shortened):

{
  "prompt": "best project management tool for small business",
  "engine": "google-ai-overview",
  "model": "google-ai-overview",
  "run": 1,
  "country": "US",
  "channel": "serp",
  "status": "ok",
  "answer": "...",
  "citations": [{ "position": 1, "url": "https://clickup.com/...", "domain": "clickup.com", "source": "ClickUp",
                  "kind": "other" }],
  "citedDomains": ["clickup.com", "..."],
  "brands": [{ "brand": "ClickUp", "domain": "clickup.com", "mentioned": true, "mentionRank": 1,
               "cited": true, "citedPosition": 1, "sentiment": "positive",
               "sentimentSentence": "ClickUp – Best for All Methodologies: Offers an expansive feature set (...) and affordable plans, ...",
               "sentimentWords": ["best for", "affordable", "ideal"] }],
  "mentionedBrands": ["ClickUp", "monday.com", "Asana", "Trello"],
  "brandSentiment": { "ClickUp": "positive", "monday.com": "positive", "Asana": "positive", "Trello": "positive" },
  "aiSearchVolume": 1,
  "aiCostUsd": null,
  "charged": true
}

SUMMARY starts with the score per brand, best first (a run of 12 November 2026, 2 questions on 5 engines, numbers shortened):

"visibilityScores": { "brands": [
  { "brand": "Asana", "score": 41.0, "byEngine": { "openai": 50.0, "perplexity": 50.0, "claude": 35.0, "gemini": 35.0,
    "google-ai-overview": 35.0 }, "mentionRate": 0.5, "citationRate": 0.2, "firstMentionRate": 0.5,
    "shareOfVoice": 0.179, "sentiment": { "positive": 5, "neutral": 0, "negative": 0, "mixed": 0 }, "netSentiment": 1.0 },
  { "brand": "ClickUp", "score": 36.0, "shareOfVoice": 0.179,
    "sentiment": { "positive": 1, "neutral": 1, "negative": 1, "mixed": 2 }, "netSentiment": 0.0 } ] }

With AI search volume on, SUMMARY.searchVolume lists each question (US, 12 November 2026; long, specific questions are asked less than short ones):

"searchVolume": { "country": "US", "language": "en", "found": 3, "questions": [
  { "question": "best crm for small business", "aiSearchVolume": 273, "aiSearchVolumeAvg": 299,
    "aiTrendDirection": "rising", "dataMonth": "2026-10", "aiMonthlySearches": [ { "year": 2025, "month": 11, "aiSearchVolume": 494 }, "..." ] },
  { "question": "What is the best project management software for a small team?", "aiSearchVolume": 5 },
  { "question": "best project management tool for small business", "aiSearchVolume": 1 } ] }

status is ok (answer with sources), zero-cite (answer without sources, still data), no-ai-overview (Google showed none for that question, free) or error (free).

Perplexity lists every page it pulled while answering: citations holds the ones its answer marks as used, and pagesRead the rest. Grok lists every page it came across and doesn't say which it used, so its rows have them in pagesRead, citationsUnknown: true and cited: null per brand, and its answers are left out of citation share.

Pricing

• $0.06 per AI answer analysed ($0.054 on Silver, $0.048 on Gold and higher plans): one engine's answer to one question, with its citations and brand matches.
• Free: failed calls, and questions for which Google shows no AI Overview. Visibility score, share of voice and sentiment add nothing to the price.
• $0.05 per AI search volume lookup, only when you turn AI search volume on: up to 50 questions in one country per lookup. Not charged when the lookup fails.
• AI usage for the assistants is billed to your Apify account by Apify's own OpenRouter gateway, at OpenRouter's rates for paying Apify plans (Apify charges free-plan accounts more for it). Google's AI Overview has no AI usage. Measured per answer on 12 October 2026 with the newest models: Gemini 3.8 Flash about $0.007, Perplexity Sonar Pro Search about $0.026, Claude Sonnet 5.5 about $0.048, GPT-6.1 Sol about $0.085, Grok 4.7 $0.15-0.23 (it reads 60 or more pages per answer).
• Plus Apify's standard run start fee of $0.00005. Your spending limit for the run is respected.

Use it from code, automations and AI agents

• API: run it and get the results in one call: POST https://api.apify.com/v2/acts/zaiq~ai-answer-tracker/run-sync-get-dataset-items?token=YOUR_TOKEN with the input as JSON. The API tab on this page has ready-made Python, JavaScript and cURL examples.
• Python (pip install apify-client):

from apify_client import ApifyClient

client = ApifyClient("YOUR_TOKEN")
run = client.actor("zaiq/ai-answer-tracker").call(run_input={
    "prompts": ["What is the best project management software for a small team?"],
    "brands": ["Asana | asana.com", "Trello | trello.com"],
    "engines": ["openai", "perplexity", "google-ai-overview"],
})
for item in client.dataset(run.default_dataset_id).iterate_items():
    print(item["engine"], item["mentionedBrands"])

• No code: connect it to Google Sheets, Zapier, Make or n8n from the Integrations tab, schedule weekly runs in Apify Console, and get a webhook when a run finishes.
• AI agents (MCP): add it as a tool to Claude, Cursor or any MCP client through Apify's MCP server: https://mcp.apify.com?tools=zaiq/ai-answer-tracker.

Weekly monitoring, competitor gaps and a report

On when a project name is set (the form starts with one), at no extra charge (no new events):

• Project name (projectName): keeps every run's results under that name and compares each run with the last one: mentions gained and lost per question and engine, citation-rate moves, cited sites gained and lost, and competitors that appeared (record CHANGES). Schedule the Actor weekly to follow the movement.
• Report (includeReport): a one-page REPORT.html with the visibility score and its formula, the score of every brand per engine with share of voice and sentiment, the sentences behind negative and mixed mentions, AI search volume beside each question when it is on, per-engine tables, competitor gaps (the questions where a competitor is named or cited and you are not, with the sites cited there) and a fix list drawn from the pages the engines cite.
• Cited pages to read (maxCitedPages, default 8): how many of the cited pages are read for the fix list.

FAQ

Is this what people see in the ChatGPT, Gemini, Claude or Grok apps? These are the vendors' official APIs with web search: the same models and search the vendors offer to developers. The consumer apps can differ (memory, personalisation, their own interface). This Actor does not log in to or scrape any AI app. Google's AI Overview is read from Google's public results page, as a signed-out searcher in the country you choose.

Why do answers change between runs? AI answers are sampled. One run is a sample, not a verdict: track the same questions weekly with 3 or more runs and watch the mention rate.

Does every question get a Google AI Overview? No. Google shows one for many questions but not all, and it varies by country and day. Questions without one are free and are marked no-ai-overview.

Which model does each engine use? The newest in the company's main family at the time of the run (ChatGPT: GPT Sol; Gemini: Flash; Claude: Sonnet; Grok: the newest Grok; Perplexity: Sonar Pro Search). Each row records the exact model.

Does it invent citations? No. What the engine returned is what you get; an answer that cites nothing is recorded as zero-cite.

How is sentiment measured? From the answer's own words around each mention of your brand, by fixed rules, so the same answer always gets the same result and there is no extra AI cost. Every label comes with the sentence it is based on, copied from the answer, so you can check it. It reads tone ("the best choice", "can get expensive"), not whether a claim is true, and like any rule it can miss sarcasm or an unusual turn of phrase.

What is AI search volume? A keyword database's monthly estimate of how often a question is asked in AI search (ChatGPT, Google AI Overviews and other AI tools), modelled from the questions people search for. Short, common questions have more volume than long, specific ones. Use it to rank your questions, not as an exact count.

Limits

• Up to 10 runs per question and 10 answers at a time.
• Grok's answers take about a minute each; the others 5-25 seconds.
• Google's AI Overview is read in the country's usual language unless you set language.