AI Visibility — your own API keys

AI Visibility queries each AI model on your behalf, which needs an API key for that model. You have two options. With Managed AI, CrunchJunkie supplies the keys — we run the models on our own platform keys, metered by check under a monthly Managed AI plan (from €25/mo, €0.05/check overage), with nothing to set up. Or you can bring your own API key for each model and pay the provider directly. You can mix the two and switch anytime — for example, use your own OpenAI key but Managed AI for the rest. Either way, keys are stored encrypted (AES-256-GCM). They run your scans, and an OpenAI, Claude or Gemini key also powers Crunch's AI features (report summaries, Ask Crunch, content briefs) unless you add a separate assistant key; Grok and DeepSeek keys run scans only. On Managed AI, those features run on Claude with CrunchJunkie's Anthropic key. The sections below show how to get your own key from each provider.

Create an account at platform.openai.com. IMPORTANT: OpenAI API access is pay-as-you-go and is billed SEPARATELY from any ChatGPT subscription — a ChatGPT Plus, Team or Business plan does NOT include API access, and its login/keys won't work here. This is the single most common mistake. So first, under Settings → Billing at platform.openai.com, add a payment method and a little credit. Then open platform.openai.com/api-keys, click "Create new secret key", name it "CrunchJunkie", and copy it — it's shown only once. Paste it into CrunchJunkie. When you connect a key, CrunchJunkie runs a quick real test of each feature the key powers and rejects a key that has no credits or billing on the spot — so you're never left thinking you're set up when you're not. The settings a key needs, and what the test costs, are in the next section. A valid key powers both the ChatGPT visibility model (currently GPT-5.6 Luna with web search) and CrunchJunkie's own AI (report summaries and the Crunch assistant, on GPT-5.6 Terra). If you'd rather not set up OpenAI billing at all, a Claude or Gemini key works just as well for CrunchJunkie AI, or use Managed AI.

An OpenAI key runs up to three things here, each on its own endpoint and model: the AI Visibility scans (the Responses API, model gpt-5.6-luna), CrunchJunkie AI (Chat Completions, model gpt-5.6-terra) and voice transcription (the audio endpoint, model whisper-1). When you connect the key, CrunchJunkie tests each of them and shows the result per feature. The key is saved for the features that work, and a feature OpenAI refuses shows OpenAI's reason and where to fix it. Four settings decide the result. Permissions. A key whose permissions are set to All works. A Restricted key needs Model capabilities: Request, which covers the three endpoints the check and the features call: Responses (Write) for the scans, Chat completions for CrunchJunkie AI and Audio for voice transcription. Set Model capabilities to Request first and then switch off what you don't need, because ticking only a sub-item has been known to leave out the request permission itself. You can change the permissions of an existing key at platform.openai.com → API keys → edit the key → Permissions; no new key is needed. When OpenAI refuses for a missing permission, CrunchJunkie names the one OpenAI reports, for example Responses: Write. If the key is set to All and OpenAI still reports a missing permission, the key's owner needs a project role that may make requests (Member or Owner). Credit and billing. API use is billed by OpenAI separately from ChatGPT (see above). The organization needs credit or a payment method under Settings → Billing, and a spend limit you set for the organization or the project stops the key once it is reached. Allowed models. If the key's project limits which models it may use (Settings → your project → Limits → Model usage), allow gpt-5.6-luna for the scans and gpt-5.6-terra for CrunchJunkie AI. OpenAI lets some models be used only by a verified organization; if it says so, verify under Settings → Organization → General → Verify Organization. OpenAI says access can take up to 15 minutes to arrive after that. No IP allowlist. Our servers' addresses change, so a project or organization that accepts requests only from listed IP addresses refuses every request we send. Use a key from a project without an IP allowlist. What the check costs. One tiny request per feature, billed by OpenAI to the key's account: a Responses request to gpt-5.6-luna capped at 16 output tokens and without web search, a Chat Completions request to gpt-5.6-terra, also capped at 16 output tokens, and, when the key also runs CrunchJunkie AI, half a second of silent audio for whisper-1. Together that is well under a tenth of a US cent. Keys from the other providers are checked with one tiny real generation as well, and a refusal names the provider's reason. Gemini's causes and fixes are under Google — Gemini. An Anthropic key needs credit under Plan & Billing at console.anthropic.com, and a monthly spend limit you set there stops it once reached. Perplexity gives the same answer for an invalid key and for an account without credit, so check both when it refuses a key.

Go to aistudio.google.com and sign in with a Google account. Open aistudio.google.com/apikey ("Get API key") and click "Create API key". Gemini has a free tier you can start with; for higher rate limits and reliable web-search grounding, link a Google Cloud project that has billing enabled. Copy the key and paste it into CrunchJunkie. This powers the Gemini (web search) model. When you connect a Gemini key, CrunchJunkie runs one tiny real generation with the model it scans with, and if Google refuses it, it tells you Google's actual reason rather than a guess. A key can be refused for reasons that have nothing to do with the key itself, so a brand-new key is no guarantee. The usual ones: the key was cut off when copying; the key's Google Cloud project has a billing account that is disabled, closed or not fully set up (BILLING_DISABLED); the Prepay credit on that billing account is at $0 (HTTP 402, which stops every key on the account); the Generative Language API is switched off in the project (SERVICE_DISABLED); the key has an application restriction (websites, IP addresses or apps) that blocks requests from our servers, or API restrictions that leave out the Generative Language API; Google refuses the request with FAILED_PRECONDITION, which Google ties to billing; or Google has blocked the whole project ("Your project has been denied access"). For billing problems, either finish the billing account in Google AI Studio → Billing, or unlink the project from it (Google says the project then returns to the free tier), or create the key in a project without billing. For a blocked project, open it in Google AI Studio and follow its banner; a new key in the same Google account usually gets the same answer. For a key restriction, remove or relax only the application restriction (IP addresses, websites or apps), since our servers' addresses aren't fixed, and keep the key's API restriction (“Restrict to Gemini API only” in Google AI Studio → API Keys): Google rejects unrestricted standard keys. A key Google is only rate-limiting is saved, with a note. One cause is about the key's age: since September 2026 Google rejects unrestricted "standard" keys, and keys created in Google AI Studio since 28 May 2026 are the newer type. So if Google says a key is "not valid" and it was made before that date, either restrict it with “Restrict to Gemini API only” in Google AI Studio → API Keys (Google keeps accepting restricted standard keys), or create a new key at aistudio.google.com/apikey and replace the old one in Settings → AI & API Keys. Nothing else changes, and your history stays. The same reason shows under "Needs attention" on Home when a scan fails on your key. Managed AI is not affected.

Sign in at perplexity.ai and open Settings → API (perplexity.ai/settings/api). Add a payment method or buy credits — API access is billed separately from a Perplexity Pro subscription — then generate an API key and copy it. Paste it into CrunchJunkie. It powers the Perplexity visibility scans, which run through Perplexity's Agent API with its web search and citations.

Create an account at console.anthropic.com and add billing/credits under Plan & Billing. Then go to Settings → API Keys (console.anthropic.com/settings/keys), create a key, and copy it. Paste it into CrunchJunkie. This powers the Claude (web search) model.

Sign in at console.x.ai with your X / xAI account, add billing, then open the API Keys section, create a key and copy it. Paste it into CrunchJunkie. This powers the Grok (web search) model.

Create an account at platform.deepseek.com and top up your balance under Billing / Top up. Open the API keys section, create a key and copy it. Paste it into CrunchJunkie. This powers the DeepSeek model.

Google's AI Overviews and AI Mode answers are read through SerpAPI rather than a direct model API. Sign up at serpapi.com and find your key on the dashboard (serpapi.com/manage-api-key) — there's a free monthly search allowance, with paid plans for higher volume. Paste the key into CrunchJunkie. This powers the Google AI Overviews and Google AI Mode models. Each check uses one SerpAPI search, except an AI Overview that Google loads in a second step: SerpAPI fetches it with a second request and bills that as a second search, so those checks use two. Which overviews load that way is only known when they run. The AI usage export (Settings → Billing) shows the searches each check used, in the serpapi_searches column, so the count can be matched against your SerpAPI dashboard.

All API keys live in one place: Settings → AI & API Keys. Paste each provider's key once, then choose what it powers with the toggles beside it — "Use for AI-Visibility tracking" (scan that model) and, for the providers that also run CrunchJunkie's own AI (OpenAI, Anthropic, Gemini), "Power CrunchJunkie AI" (report summaries, anomaly detection and the Crunch assistant). So a single OpenAI key can drive both your ChatGPT visibility scans and your report summaries — no need to enter it twice. Keys are encrypted at rest (AES-256-GCM) and used only for the purposes you enable; you can remove one at any time. When you connect a key, CrunchJunkie runs one tiny real generation and, if the provider refuses it, says why in your language: the key isn't valid, the account has no credit or billing, the key lacks a permission, its project may not use the model, the organization must be verified first, the provider is only rate-limiting it (the key is then saved, with a note), or the provider couldn't be reached. When the provider's answer fits none of these, you see its own words. An OpenAI key is tested per feature, as described in Using your own OpenAI key. If a key later expires or runs out of credit, it's flagged in "Needs attention" on Home, so a scan never fails silently. Connecting a key does not add its model to any project: each project's Engines to track decides what its scans ask. After you connect a key for AI visibility, the page offers to add its engine to the projects that don't track it yet; nothing changes until you click, and nothing runs until each project's next scan, on your key. Prefer not to manage keys at all? Turn on Managed AI for any model and we'll run it on our key for a flat monthly price.