AI Search Visibility Tracker (ChatGPT, Perplexity, Gemini)

Check whether ChatGPT, Perplexity and Gemini name your brand when buyers ask them about your market. You write the questions, your brand and competitors; the Actor asks each AI through its official API with web search on and returns one row per question and AI: mentioned or not, in which place, which competitors came up and which websites the answer cited.

Built for brands, SEO and GEO (generative engine optimization) teams. The smallest run is one question, your brand and one API key — usually done in under a minute.

Table of contents

• What it does
• Your API key is required
• Quick start
• Use cases
• Input
• Output
• Pricing
• Limits
• API, export and integrations
• Troubleshooting
• FAQ
• Support

What it does

For every prompt and every AI you pick, the Actor:

1. sends the prompt once through the provider's official API with web search enabled — OpenAI Responses API with the web_search tool, Perplexity Agent API, Gemini API with Google Search grounding;
2. reads the answer and finds your brand, its other spellings and your competitors as whole words, case-insensitive;
3. ranks the names in the order the answer first mentions them, so you see your position next to the competitors;
4. lists the cited sources and tells you whether one of them is on your own domain;
5. saves the answer text, an excerpt around the first mention, tokens and the time it was received.

The run summary adds up each AI: mention rate, average position, how often you came first, how often your site was cited, and the mention rate of each competitor.

Good fit

• Track how often AI assistants recommend you versus competitors for the questions your buyers ask.
• Find which pages AI answers cite for your category, so you know where to earn coverage.
• Run the same prompt set every week with an Apify schedule and compare the mention rate over time.

Not a fit / limits

• API answers are not the same as the ChatGPT, Perplexity or Gemini apps. The apps add personal memory, account settings, location and their own system prompts, and they change often. The API with web search is the closest repeatable signal, not a copy of what one person sees on screen.
• AI answers vary from run to run. One answer is a sample; use several prompts and repeat runs to see a trend.
• The Actor never scrapes the consumer apps and never logs in anywhere. It only uses the public APIs with your key.
• Position is the order of first mention among the names you track, not a place in a numbered list the AI might show.
• Gemini's API has no location setting, so country applies only to OpenAI and Perplexity.

Your API key is required

You pay each AI provider directly for its answers through your own key. Without a key the run still succeeds: you get one free api_key_required row that says which field to fill and where to get the key, and no answer is charged.

AI	Input field	Where to get a key	What the provider bills to your key
ChatGPT models (OpenAI)	openaiApiKey	platform.openai.com/api-keys	Tokens plus a web search tool fee per answer. Needs prepaid API credits; a ChatGPT Plus subscription does not include API use.
Perplexity	perplexityApiKey	perplexity.ai/account/api/keys	Per-request and token fees for the preset; each row shows providerCostUsd as Perplexity reports it (about $0.002 per answer with the fast preset in our tests).
Gemini (Google AI Studio)	geminiApiKey	aistudio.google.com/apikey	Tokens plus Google Search grounding. Grounding needs billing enabled on the key's Google Cloud project: the free tier does not include it for Gemini 3 models. Paid projects get 5,000 free searches a month, then Google charges per 1,000 searches.

Provider prices change; check each provider's pricing page before large runs. Every row reports inputTokens and outputTokens so you can check the bill.

Keys are stored encrypted by Apify (secret input), used only for this run and never written to the dataset, the run summary or the log. The Actor checks each key with a free call before it asks any question, so a wrong key costs nothing.

Quick start

1. Open the Input tab.
2. Keep the prefilled prompt best web scraping platform or write your own under Prompts.
3. Enter your brand under Your brand (prefilled: Apify).
4. Paste at least one key — OpenAI, Perplexity, or Gemini from a Google AI Studio project with billing on — and pick that AI under AI assistants to check.
5. Click Start, then open the Visibility view of the dataset.

Expected result: one row per prompt and AI with mentioned, position, ranking, citedSources and an answerExcerpt.

Use cases

Buyer job	Input	What you get	Key fields	Next step
Am I recommended for my category?	prompts, brand, brandAliases	One row per prompt and AI	mentioned, position, answerExcerpt	Run summary mentionRate per AI
Who wins the answer instead of me?	competitors	Names in order of first mention	ranking, competitorsMentioned	Compare competitor mention rates in the run summary
Which pages do AI answers trust?	brandDomain	Cited links per answer	citedSources, brandCited	Export the dataset and group by domain
Is visibility changing over time?	Same input on a schedule	A new dataset each run	fetchedAt, mentioned, position	Schedule plus webhook or Google Sheets export

Input

Field	What it is	Required	Default	Notes
prompts	Questions buyers ask	yes	prefill: best web scraping platform	Up to 100, each up to 2000 characters. Duplicates are removed.
brand	Your brand name	yes	prefill: Apify	Whole-word, case-insensitive match.
brandAliases	Other spellings of the brand	no	—	Up to 10, e.g. a product name.
brandDomain	Your website domain	no	—	e.g. apify.com. Turns on brandCited; subdomains count.
competitors	Competitor names	no	—	Up to 25. Other spellings after |: Bright Data|BrightData.
providers	AIs to ask	no	every AI you gave a key for	openai, perplexity, gemini.
openaiApiKey, perplexityApiKey, geminiApiKey	Your API keys	at least one for answers	—	Secret, stored encrypted.
country	Two-letter country code	no	—	Approximate searcher location for OpenAI and Perplexity.
openaiModel	OpenAI model	no	gpt-5.4-mini	Must support web search.
perplexityPreset	Perplexity preset	no	fast	Other presets may cost more.
geminiModel	Gemini model	no	gemini-3.8-flash

How the settings combine:

• An AI listed in providers without a key → one free api_key_required row for it; the other AIs still run.
• providers empty → every AI that has a key. No key at all → one free row naming all three key fields.
• AIs run in parallel; prompts for one AI run one after another.

Input example

{
  "prompts": ["best web scraping platform", "best tool to scrape Google Maps reviews"],
  "brand": "Apify",
  "brandDomain": "apify.com",
  "competitors": ["Bright Data|BrightData", "Zyte", "ScrapingBee", "Octoparse"],
  "providers": ["openai", "perplexity"],
  "country": "US",
  "openaiApiKey": "<your OpenAI key>",
  "perplexityApiKey": "<your Perplexity key>"
}

This asks 2 prompts × 2 AIs = 4 answers.

Output

Output fields

Field	Meaning
prompt, provider, model, requestedModel	The question, the AI (openai, perplexity, gemini), the model that answered and the model you asked for.
brand	The brand name checked in this row.
status	ok (answer checked, the only charged status), no_answer (empty answer), api_key_required, error.
mentioned	true / false; null when there is no answer to check — null is not "not mentioned".
position	Place of your brand among the tracked names in order of first mention (1 = first); null when not mentioned.
mentionCount	How many times the brand is named.
competitorsMentioned	Competitors found: name, position, mentionCount.
ranking	Brand and competitors in order of first mention.
brandCited, brandCitedPosition	Whether a cited source is on brandDomain, and its place; null without brandDomain.
citedSources, citedSourcesCount	position, url, title, domain of each cited page, and how many there are. Gemini returns Google redirect links; domain is the real site.
searchQueries	Web searches the AI ran, when the API reports them. Empty means the AI answered from its own knowledge — see below.
answerExcerpt, answer	About 400 characters around the first mention; full answer up to 20,000 characters (answerTruncated).
country, countryApplied	Country sent and whether this AI used it.
inputTokens, outputTokens, providerCostUsd	Usage billed to your key; cost only when the provider reports it (Perplexity).
latencyMs	How long the AI took to answer, in milliseconds.
errorCode, errorMessage	Why a row is not ok, and what to do.
fetchedAt	When the answer was received (ISO 8601).

Web search is switched on for every AI, but the AI decides whether to use it. In our tests Gemini answered general questions such as best web scraping platform from its own knowledge, with no cited sources, and searched the web for questions about fresh facts such as best web scraping platform in 2026 (13 cited sources). Empty citedSources is the AI's choice, not a failed run, and the row is still a checked answer.

Output example

Trimmed row from a real run (Perplexity, US, 22 October 2026):

{
  "prompt": "best web scraping platform",
  "provider": "perplexity",
  "model": "openai/gpt-6-luna",
  "status": "ok",
  "brand": "Apify",
  "mentioned": true,
  "position": 1,
  "mentionCount": 1,
  "competitorsMentioned": [
    { "name": "Octoparse", "position": 2, "mentionCount": 1 },
    { "name": "ScrapingBee", "position": 3, "mentionCount": 1 },
    { "name": "Bright Data", "position": 4, "mentionCount": 1 }
  ],
  "ranking": ["Apify", "Octoparse", "ScrapingBee", "Bright Data"],
  "brandCited": true,
  "brandCitedPosition": 3,
  "citedSources": [
    { "position": 1, "url": "https://webscraper.io/blog/best-web-scraping-platforms", "title": "Best web scraping platforms compared - Web Scraper", "domain": "webscraper.io" },
    { "position": 3, "url": "https://blog.apify.com/best-web-scraping-tools/", "title": "6. Octoparse", "domain": "blog.apify.com" }
  ],
  "citedSourcesCount": 6,
  "answerExcerpt": "**Apify is the best all-around choice for developers**: it combines ready-made scrapers (“Actors”) with custom code, cloud runs, scheduling, and storage…",
  "country": "US",
  "countryApplied": true,
  "inputTokens": 6325,
  "providerCostUsd": 0.00168,
  "errorCode": null,
  "fetchedAt": "2026-10-22T05:46:32.698Z"
}

Run summary

The RUN_SUMMARY record in the key-value store has the run status (COMPLETE, PARTIAL, ERROR, API_KEY_REQUIRED, INVALID_INPUT, CANCELLED), a stop reason, counts, and for each AI: mentionRate, averagePosition, firstPlace, brandCited, each competitor's mentionRate, tokens and provider cost. Key fields only say whether a key was given, never the key.

When something stops a run

Situation	What you get
No key for an AI	One free api_key_required row; the run succeeds.
Wrong key or unknown model	One error row for that AI (API_KEY_INVALID / MODEL_UNAVAILABLE), found by the free key check before any question.
Your provider quota runs out mid-run	Answers so far are kept, the rest of that AI is skipped, status PARTIAL with stop reason QUOTA_LIMIT; the run succeeds and other AIs continue.
Your spending limit on Apify is reached	Answers so far are kept, PARTIAL with CHARGE_LIMIT.
One prompt is refused by the AI	An error row for that prompt (PROVIDER_BAD_REQUEST); the next prompt goes on.
An AI gives 3 empty or refused answers in a row	That AI stops with stop reason SOURCE_ERROR so it does not keep spending your provider quota; rows so far are kept and other AIs continue. The run is marked failed so you notice it.

Pricing

Pricing unit

This Actor uses pay-per-event pricing:

Event	When it is charged	Free plan	Bronze	Silver	Gold and higher
AI answer checked (ai-answer)	Once for each row with status: ok	$0.005	$0.0045	$0.004	$0.0035
Actor start (apify-actor-start)	Once per run, for each GB of run memory (default 256 MB = one event)	$0.002	$0.002	$0.002	$0.002

Rows with api_key_required, error or no_answer are free, so a run without a key, with a wrong key or with an exhausted quota costs only the start event. A run that stops early keeps and charges only the answers already saved. If you raise the run memory, the start fee grows with it (one event per GB); 256 MB is enough for 100 prompts and 3 AIs.

The AI providers bill their own usage to your API keys (see Your API key is required); that cost is not part of the Apify price.

Cost examples

Scenario	Input	Charged answers	Apify price, Free plan	Gold plan
Try it	1 prompt, 1 AI	1	$0.007	$0.0055
Weekly check	10 prompts, 1 AI	10	$0.052	$0.037
Full panel	10 prompts, 3 AIs	30	$0.152	$0.107
Large set	100 prompts, 1 AI	100	$0.502	$0.352
No key or wrong key	any	0	$0.002	$0.002

Set Max cost per run in the Apify run options to cap spending; when it is reached the run keeps the answers so far with stop reason CHARGE_LIMIT.

Limits

• Up to 100 prompts per run, 3 AIs, 25 competitors, 10 brand aliases.
• In our tests an answer took about 3 seconds on Perplexity, about 4 seconds on OpenAI and 15–27 seconds on Gemini, and 10 prompts on OpenAI and Perplexity finished in 44 seconds. Prompts for one AI run one after another and the AIs run side by side, so the run time grows with the number of prompts. A slower model or a rate-limited key takes longer. The default timeout is 1 hour.
• One answer per prompt and AI per run; answers are not cached between runs.
• No Claude, Google AI Overviews, Copilot or other assistants.

API, export and integrations

• API: start a run with the same JSON as the Input tab through the Apify API or client libraries; read rows from the default dataset and totals from the RUN_SUMMARY record.
• Export: the dataset downloads as JSON, CSV, Excel or HTML; the Visibility view keeps the main columns.
• Scheduling: save your input as an Apify task and add a schedule (for example weekly) to track the mention rate over time. Each run creates a new dataset.
• Webhooks / no-code: an Apify webhook on run success can send the dataset to Google Sheets, Make, Zapier or your own endpoint; the next system usually reads prompt, provider, mentioned, position and fetchedAt.

Use with AI agents (MCP)

• MCP server with only this Actor: https://mcp.apify.com?tools=cliqtomedia/ai-search-visibility
• Smallest useful input: {"prompts": ["best web scraping platform"], "brand": "Apify", "providers": ["gemini"], "geminiApiKey": "<key>"}
• Key output fields: mentioned (brand named or not), position (place among tracked names), ranking, citedSources, status.
• A run without a key returns one free api_key_required row; the agent should ask the user for a key.
• A run with 1 prompt and 1 AI usually takes under a minute and costs $0.007 on the Free plan, plus what the provider bills to the key.

Troubleshooting

Gemini rows show API_QUOTA_EXHAUSTED

Google Search grounding is not part of the Gemini API free tier. Enable billing for the Google Cloud project behind your key in Google AI Studio, then run again. On a paid project the error means the monthly grounding allowance or your spending cap is used up.

OpenAI rows show API_QUOTA_EXHAUSTED

Your OpenAI API account has no credits. Add credits under Billing at platform.openai.com; a ChatGPT subscription does not cover the API.

mentioned is null

The row has no answer to check (api_key_required, error or no_answer). Only false means the AI answered without naming your brand.

My brand is a common word

Matching is whole-word and case-insensitive, so a brand like "Notion" also matches the plain word. Add a more specific alias (product name, domain) and read answerExcerpt to confirm.

FAQ

Is this what I see in the ChatGPT app?

No. It is the provider's API with web search, which is close but not identical: the apps add personalization, memory and their own settings. Use it as a repeatable benchmark across prompts, competitors and time.

Why do results change between runs?

AI answers are not deterministic and web results change. Ask several related prompts and repeat runs on a schedule; compare mentionRate in the run summary rather than single answers.

Do you store my key or the answers anywhere else?

The key is a secret input, encrypted by Apify and never written to the dataset, run summary or log. Answers go only to your run's dataset.

Can I check Claude, Google AI Overviews or Copilot?

Not in this version. It covers OpenAI, Perplexity and Gemini APIs.

Support

Open an issue on the Actor's Issues tab. Include the run ID and the error code. Never send your API keys.