AI Visibility Tracker

Find out whether AI models mention your brand when people ask about your market. Give it your brand, your competitors and your prompts. It asks each model through the vendor's official API, using your own API keys, and reports whether and how your brand appears, at $20 per 1,000 checks.

It calls the official APIs of OpenAI, Anthropic, Google Gemini and Perplexity. You choose which ones by adding a key for each.

What you get

• One row for each prompt and model, with the full answer
• Whether your brand or an alias was mentioned, how often, and where it first appeared
• Your rank among the brand names you listed, by first appearance
• Which of your competitors were mentioned
• The source links the API returned, and whether any of them point to your website
• Input and output token counts, so you can see what the vendor is likely to bill you for
• One free summary row per run, with visibility share per model and per competitor

Sample output

One check:

{
  "recordType": "check",
  "prompt": "What are the best note-taking apps for small teams?",
  "provider": "openai",
  "model": "gpt-6-luna",
  "brandMentioned": true,
  "mentionCount": 2,
  "firstMentionPosition": { "character": 42, "rank": 2 },
  "competitorsMentioned": ["Evernote"],
  "brandCited": true,
  "citations": ["https://www.notion.so/product", "https://reviews.example.com/apps"],
  "answer": "For small teams, Evernote is popular, and Notion is a strong choice. Many teams pick Notion for docs.",
  "inputTokens": 75,
  "outputTokens": 186,
  "status": "ok",
  "reason": null,
  "checkedAt": "2026-11-02T10:00:00.000Z"
}

The summary row for a run:

{
  "recordType": "summary",
  "overallVisibilityShare": 0.25,
  "checksDelivered": 4,
  "checksFailed": 0,
  "visibilityByModel": [
    { "provider": "openai", "model": "gpt-6-luna", "checks": 2, "failed": 0, "brandMentions": 1, "visibilityShare": 0.5, "brandCitations": 2, "citedShare": 1 },
    { "provider": "gemini", "model": "gemini-3.8-flash", "checks": 2, "failed": 0, "brandMentions": 0, "visibilityShare": 0, "brandCitations": 0, "citedShare": 0 }
  ],
  "visibilityByCompetitor": [
    { "name": "Evernote", "checks": 4, "mentions": 4, "visibilityShare": 1 }
  ],
  "checkedAt": "2026-11-02T10:00:00.000Z"
}

These examples show the format, with made-up answers. Every row also has the other fields set to null, so the columns stay the same. character counts from 0, and a visibility share runs from 0 to 1.

How to use it

1. Enter your brand name, any aliases, your website domain and your competitors.
2. Add your prompts, one per line, such as "What are the best note-taking apps for small teams?"
3. Add an API key for each vendor you want to check, set Maximum checks, and press Start.

Each prompt sent to each model is one check, so 10 prompts and 4 models make 40 checks.

Model names change often. The defaults favour low cost, and you can type any current name from the vendor's documentation. For Perplexity, you can enter an effort level, such as fast, or a model ID with a slash, such as perplexity/sonar.

Pricing

You pay Apify for this Actor, and you pay each AI vendor separately.

Event	Price
Each check delivered	$0.02 ($20 per 1,000)
Each run started	$0.00005

Status rows, the summary row and failed checks are free. A check counts as delivered when a model returned an answer, including an answer the vendor cut short, which is marked truncated. A run with Maximum checks left at 50 costs at most about $1.00 here, plus what your vendors bill.

AI vendor bills: OpenAI, Anthropic, Google and Perplexity bill your own keys for the tokens and web searches you use. Web search has its own fee at some vendors. This Actor has no control over those prices, and a check that fails or gets retried may still appear on your vendor bill. Check your usage limits with each vendor before a big run.

On the free Apify plan, each run delivers up to 100 checks, and a status row tells you when you reach that limit. Any paid Apify plan removes it.

Status rows

The Actor always tells you what happened. Check rows have a status of ok, truncated or failed. Other messages arrive as free rows with recordType set to status:

Status	Meaning
no_keys	No API keys were given, so nothing was asked
no_prompts	The prompt list is empty, so nothing was asked
no_brand	No brand name was given, so nothing was asked
missing_key	One vendor was skipped, because it has no key
bad_key	The vendor refused the key, so no more requests went to it in this run
model_rejected	The vendor refused the model name or its web search tool, so no more requests went to it in this run. Other vendors carry on
quota_exceeded	The vendor says your account is out of credit or over a limit, so no more requests went to it in this run
provider_stopped	3 checks in a row failed on this vendor, so no more requests went to it in this run
max_checks	The run reached your Maximum checks
free_limit	The run reached the free plan limit
spending_limit	The run reached your maximum cost

If you only want answers, filter on recordType equal to check. Temporary errors from a vendor, such as rate limits and server errors, are retried with increasing waits before a check is marked failed.

Your API keys

Keys use Apify's secret input, so they are stored encrypted. Each key is sent only to its own vendor's official API host, in a request header. Redirects are not followed. This Actor doesn't log, store or return your keys, and it removes them from vendor error messages before writing a row. The answers and prompts themselves are stored in your Apify dataset, and each vendor handles what you send it under its own terms. Requests ask OpenAI, Gemini and Perplexity not to store the response for later retrieval.

Good to know

• API answers can differ from what people see in the consumer apps. ChatGPT, Claude, Gemini and Perplexity as apps can use different models, instructions, personalisation and search settings. This Actor shows what the API returns, not what any one person sees.
• Answers vary between runs. The same prompt can give a different answer minutes later, so one run is a sample, not a measurement. Run the same prompts repeatedly and compare the trend before you act on a small change.
• Matching is literal. It ignores case and only counts whole words, so Notion doesn't match notional, and a hyphen or punctuation next to a name still counts. It can't tell meanings apart, so a brand called Box also matches "in a box". It won't find translations or misspellings, so add those as aliases. Names in scripts written without spaces, such as Japanese, can be missed.
• Rank only covers names you listed. A rank of 1 means your brand came before your listed competitors, even if the answer named another brand first. Rank is empty if you list no competitors.
• Overlapping names go to the longest match. If your brand is Apple and a competitor is Apple Music, "Apple Music" counts only for the competitor.
• Citations depend on the vendor. Web search is turned on for every check, and it can bill your key for searches. Answers can still have no citations. For Perplexity, if an answer marks no sources, all of its search results are listed. Google can return its own redirect links, and those are matched to your domain by their title.
• Sentiment isn't included, because a simple rule would be too unreliable to be useful. The full answer is in each row, if you want to score it yourself.
• One model per vendor per run. To compare two models from one vendor, run the Actor twice.
• Your prompt is sent exactly as you wrote it, with no added instructions. Perplexity's effort levels add their own instructions on Perplexity's side.

Using it with AI agents

Every field is optional, and the main inputs are lists. An agent can check many prompts in one call and read status rows to see what went wrong. Keep maxChecks low, because each check also costs money on your vendor keys.

Data and use

This Actor calls only the official APIs you supply keys for. It uses no logins, cookies or scraping of consumer apps. You are responsible for following each vendor's terms and for how you use the results. This Actor is not affiliated with or endorsed by OpenAI, Anthropic, Google, Perplexity or any brand named in its examples.

Support

Open an issue on the Issues tab with your input, leaving out your keys, and what you expected.