BYOK AI Visibility Tracker: ChatGPT, Perplexity, Gemini

Bring your own API keys (BYOK). This Actor does not include any AI usage. You supply your own OpenAI, Perplexity and/or Google Gemini API key, and each provider bills you directly for the calls made with it. Without a key, the run is a free preview: a sample report with fictional brands plus a real AI-readiness check of your own website, $0 charged. Google AI Overviews are optional and also need your own Apify API token.

Your buyers now ask ChatGPT, Perplexity, Gemini and Google's AI Overviews which product to choose. With your keys, this Actor asks those engines your buyers' real questions and shows you, in one run:

• whether each engine names you, and in what position next to your competitors
• your share of voice and a 0–100 AI visibility score for each engine
• the exact pages AI cites instead of you: review sites, Reddit threads, blogs and competitor pages
• a prioritised fix list built from that evidence, plus a free AI-readiness check of your own site
• a white-label HTML report your client can open on any phone, and a shareable "Does ChatGPT recommend you?" score card (PNG)

Every answer comes from each provider's official API: OpenAI's Responses API with web search, Perplexity's Agent API with web search, and the Gemini API with Grounding with Google Search. Google AI Overviews (optional) come through Apify's own maintained Google AI Overviews Scraper, started with your Apify token so it bills to your account. Nothing logs into or scrapes the ChatGPT, Perplexity or Gemini apps.

This is an independent tool. It is not affiliated with, sponsored or endorsed by OpenAI, Perplexity, Google or any other provider; product names are used only to say which API is called.

Try it free first: run it with no keys (or turn on Demo mode) to get a full sample report for fictional brands plus a real AI-readiness check of your site, with no charge.

What you get

Output	Where	What it is
REPORT.html	Key-value store	The client-ready report: verdict, score per engine, competitor leaderboard, question-by-question matrix, pages cited instead of you, AI-readiness of your site, ranked fixes, change since last run, method notes, answer excerpts. Self-contained (no external files), responsive, prints cleanly. Dark or light theme, your agency name, logo and colour.
SCORECARD.png	Key-value store	1200×675 score card sized for X and LinkedIn.
Dataset	One row per engine × question × run	Answer text, whether you were named, your position, whether your site was cited, tone, competitors named, every cited URL and domain, search queries the engine ran, and the estimated API cost of that answer.
OUTPUT	Key-value store	All scores, the leaderboard, fixes, citation gaps, discovered brands and links to the report and card, for Zapier, Make, n8n or your own dashboard.

How it compares

	BYOK AI Visibility Tracker	Typical Store tracker	SaaS (Otterly, Semrush AI toolkit, Ahrefs Brand Radar)
Engines	ChatGPT (OpenAI web search), Perplexity, Gemini (Google Search grounding), Google AI Overviews	2–5, often through a model router or scrapers	Varies by plan
How answers are collected	Official provider APIs with your own keys; AI Overviews via Apify's own actor on your token	Varies; some scrape the consumer apps	Proprietary
Pages AI cites instead of you, by type	Yes: review site, forum, video, publisher, competitor	Cited domains list	Yes, on higher plans
Prioritised fix list with evidence	Yes, each fix shows why and how	Rare	Some
Your site's AI-readiness	Yes: AI crawler access in robots.txt, bot-challenge detection, structured data, meta, sitemap	No	Partial
White-label HTML report + share card	Yes, both, every run	JSON/CSV; one offers a Markdown report	Yes, on agency plans
Run-to-run consistency	1–3 runs per question, shown as k of n	Rare	Rare
Price	$0.02 per analysed answer + $0.79 per report; you pay the AI providers directly, usually about a cent per answer	$0.05–$0.35 per answer	$29–$299 a month

Quick start

1. Add at least one of your own API keys: OpenAI, Perplexity or Gemini. You need an account with that provider; this Actor cannot run engines without your key. Each key is a secret input: Apify stores it encrypted, and the Actor never writes it to the dataset, the report or the log.
2. Enter your brand name and website, up to 10 competitors (Name | domain.com), and either your buyers' questions or a category (the Actor writes buyer-intent questions for you).
3. Set Max API spend on your keys. The Actor stops asking new questions before it would pass it.
4. Run. Open REPORT.html from the Storage tab or OUTPUT.reportUrl.

{
  "brandName": "Acme CRM",
  "brandDomain": "acmecrm.com",
  "competitors": ["HubSpot | hubspot.com", "Pipedrive | pipedrive.com", "Zoho CRM | zoho.com"],
  "category": "CRM",
  "audience": "small agencies",
  "market": "UK",
  "maxPrompts": 10,
  "openaiApiKey": "<secret>",
  "perplexityApiKey": "<secret>",
  "geminiApiKey": "<secret>",
  "includeGoogleAiOverviews": true,
  "apifyTokenForAio": "<secret, your own Apify token>",
  "samplesPerPrompt": 2,
  "maxApiSpendUsd": 1,
  "trackingName": "acme-weekly",
  "agencyName": "Brightline Growth",
  "reportTheme": "light"
}

Track it weekly: set the same trackingName on a weekly schedule. Each report then shows the change in score and share of voice since the previous run.

How the score works

• AI visibility score (0–100) = 50% how often you are named + 25% how early you are named (1 ÷ position among tracked brands) + 25% how often your own site is cited. It is calculated per engine, and the overall score is the average across engines.
• Share of voice = answers naming you ÷ all answers naming any tracked brand.
• Tone is a transparent keyword score of the sentences that name each brand, including negation ("not expensive"). It is deterministic and costs nothing.
• Brands you did not list that AI keeps recommending are surfaced so you can add them as competitors.

AI answers vary. This audit tells you by how much.

AI answers change between runs, days, locations and accounts, and API answers can differ from the consumer apps. This audit is a measured sample, not a promise of what every user sees. Set Runs per question to 2 or 3 to see how consistent each engine is (shown as "named in k of n runs"), and track the trend with weekly runs rather than judging any single answer.

Costs

This Actor (pay per event):

• $0.02 per engine answer that is analysed
• $0.79 per audit report (HTML report, score card and fix list), charged once per run
• Demo mode, failed calls, skipped questions and empty AI Overviews are free
• The Actor keeps room for the report inside your run's maximum charge, so a capped run still delivers a report

Your own API keys are billed by each provider. The run's estimated spend is shown in OUTPUT.estimatedApiSpendUsd, and maxApiSpendUsd caps it.

• OpenAI: web search is $10 per 1,000 calls, plus tokens. With the default gpt-6-luna that is about $0.011 per answer.
• Perplexity: the Agent API at fast effort, billed per token plus search.
• Gemini: Grounding with Google Search has 5,000 free requests a month on a paid-tier key, then $14 per 1,000.
• Google AI Overviews (optional): runs Apify's actor with the Apify API token you paste, so it is billed to your own Apify account at that actor's Store price, capped by aioMaxChargeUsd. Without the token it is skipped.

A typical 10-question audit on 3 engines costs about $1.39 here plus about $0.35 on your keys.

FAQ

Is this scraping ChatGPT? No. It calls OpenAI's, Perplexity's and Google's official APIs with your keys. Google AI Overviews are optional and come from Apify's own maintained actor.

Why bring my own keys? You pay the providers' list price with no markup, your usage stays on your accounts, and you can pick the model. This Actor never pays for or resells AI usage.

Do I need keys for all engines? No. One key is enough; engines without a key are skipped and listed in OUTPUT.missingKeys.

Are my keys safe? They are secret inputs, encrypted by Apify. The Actor sends each key only to its own provider's API and never writes it to the dataset, report, score card or log (this is covered by an automated test).

Does it work for local businesses or other countries? Yes. Set market ("Pune", "London") and country (IN, GB).

Can I white-label it for clients? Yes. Set agencyName, logoUrl and accentColor. hideOarsmanBranding removes the "Built with" footer.

What if a key is wrong? That engine stops after its first rejected call, the other engines finish, and OUTPUT.engineErrors says why. Rate limits and server errors are retried with backoff.

Limits

• Consumer apps can personalise answers, so API answers are a close, repeatable proxy for them, not an identical copy.
• Gemini returns cited sources as Google redirect links. The Actor reads the source domain from each link's title.
• The site check reads public pages only and respects bot challenges: if your site challenges our checker, the report says so instead of guessing.

Built by Oarsman AI with the help of AI coding agents. Every engine's parser is tested against fixtures in each API's documented response shape; results come only from the APIs you call with your keys.