Back to the tool
Frequently asked questions

FAQ

Everything you need to know about what data the tool accepts and how it handles them.

Data security and privacy

Uploaded CSV data stay in your browser. No file is uploaded to a server, stored in a database, or logged. Processing runs locally via DuckDB-wasm — an engine that runs directly in your browser tab.

Reload the page = completely clean. We don't use cookies, localStorage, IndexedDB or any persistent storage. Data only live for the duration of your session.

We (JIRKONT s.r.o.) don't have access to your data. The app is stateless — the server only forwards requests and returns responses.

What is sent to language models

To generate prompts, we send the model only aggregated snippets from your data:

  • Header and 5 sample rows of each file (for auto-detection of type)
  • Top 300 most searched queries + impression counts (lowercased)
  • Top 20 URLs of the most cited / most visited pages (as context)
  • Personas (if you uploaded them)

These snippets travel through OpenRouter to the selected language model (Claude, GPT, Gemini — depending on configuration). Route:

Your browser → Vercel (proxy) → OpenRouter → language model

Neither Vercel nor we log request contents. The standard data policy of OpenRouter and model providers (Anthropic, OpenAI…) doesn't store prompts for training, but specifics depend on the model — see their data policy.

What data does the tool need?

The tool processes CSV exports from marketing tools. Auto-detection recognizes the type from the header — you don't need to rename files.

SourceWhat's insideUsage
Bing AI Performance — queriesquery textUsed for ranking
Google Search Console — queriesquery textUsed for ranking
Keyword researchkeywordUsed for ranking
Bing AI Performance — pagesURLContext only
Google Search Console — pagesURLContext only
Google Analytics 4 — landing pagesURLContext only
Personasname + descriptionContext only

What the columns mean: "Used for ranking" = top prompts are picked directly from this source. "Context only" = used as supporting information for the LLM (site topic, popular pages, target audience).

Other frequent questions

Do I have to upload all file types?

No, anything you have works. The minimum requirement is at least one source of queries or keywords (Bing AI queries, Google Search Console queries, or keyword research). Without it the tool has nothing to build prompts from — URLs and personas are only context, not what people actually typed.

What about files with diacritics or special formats (Bing with BOM, GA4 with NBSP)?

No problem. The tool automatically removes BOM, normalizes non-breaking spaces, and other invisible characters. It also handles GA4 exports that have metadata before the header (lines starting with #) — it finds the header itself.

What is a 'Discovery prompt'?

Concrete prompts come directly from real queries in your data — typically asking about specific products or categories ("Which sheet-metal cabinet for a workshop?"). Discovery prompts are hypothetical questions of a customer at the very beginning, who doesn't know about your offer yet — they ask based on life situation ("How do I tidy up my garage?"). They serve as inspiration for top-of-funnel content.

Which model does the tool use?

The tool calls a selected model through OpenRouter — default is openrouter/owl-alpha, but it can be switched in configuration (Anthropic Claude, OpenAI GPT, Google Gemini and others). If the primary model fails or returns an invalid response, a fallback model is tried automatically.

Can I use the tool on my own account / install locally?

Yes, the project is open-source. You can clone it from GitHub, add your own OpenRouter API key, and run it locally (npm run dev) or deploy to your own Vercel account.