Live Prompt Check

Live Prompt Check is Rankion's tool for ad-hoc multi-LLM comparisons. Unlike AI Visibility Tracking (continuous monitoring on a schedule), Live Prompt Check is a single run at the push of a button: input → 8-12 LLM responses in parallel → live card view → save + export.

What It Can Do

• Multi-LLM fan-out — a single prompt goes to ChatGPT, Perplexity, Claude, Gemini, Copilot, Grok (whichever are available).
• With/without web search — a separate toggle per platform. You see side by side how knowledge changes with live browsing.
• Live display — cards fill in progressively. Polling every 3s, no F5 needed. If one LLM takes 30s, the rest are already long done.
• Sentiment + brand extraction — every response is automatically classified (positive/negative/neutral/not mentioned) and all mentioned brands are extracted. Multi-brand lists per response.
• Persistent runs — all inputs + responses land in a searchable list with sorting + filters.
• Full-text search — search across your prompts AND the LLM responses. Filter by platform, sentiment, status, project.
• Sharing — token-protected public link with an expiration date + optional password. Recipients see the results with Rankion branding, no login required.
• Export — CSV (for Excel/Sheets) or JSON (for devs/scripts).
• Re-run — restart with the same settings in one click (e.g., two days later, to see whether ChatGPT "remembers").

When to Use

• You want to know: "What does ChatGPT vs. Claude vs. Perplexity say about my brand?"
• You're testing a prompt for a new campaign or a blog brief and need multiple perspectives.
• You're comparing: how do "with search" and "without search" responses differ?
• You're documenting LLM responses as of a specific date (candidate research, client briefing, audit).
• You need a quick sentiment snapshot on a current question.

When NOT to Use

• For continuous monitoring — that's AI Visibility Tracking (TrackingProject + frequency). Live Prompt Check is for one-off runs.
• For bulk (50+ prompts at once) — no bulk endpoint. Use tracking runs with configured prompts instead.
• If you only need one LLM response — the Agentic Chat itself has an LLM, which is faster.

Getting Started

1. Open Sidebar → Analysis → AI Visibility (GEO) → Live Prompt Check.
2. Click "New Prompt Check".
3. Select a project (for brand context in sentiment analysis — brand_name must be set on the project).
4. Enter a prompt (5-5000 characters, e.g., "What are the best mobile phone plans for freelancers?").
5. Select platforms + modes — by default all are active with web search. Toggle per platform.
6. "Start Prompt Check" — the server dispatches 8-12 worker jobs in parallel.
7. Follow along live — cards fill in. No F5 needed.

Detailed Usage

Card Accordion Layout

Every card's header shows: brand icon + platform + mode (with/without web search) + sentiment badge + latency. By default, all cards are expanded — you see all responses directly. Clicking the header collapses individual cards (status stays visible).

In the body: the markdown-rendered LLM response (with bold text, lists, links), below that brand pills with mentioned brands, and expandable citations (URL list in web search mode).

Sticky Bar at the Top

Sentiment distribution across all responses + top brands (most frequently mentioned) for a quick scan.

Compare Mode

Toggle in the top right: "↔ Compare Mode" switches from the card accordion to a side-by-side table (Platform | Sentiment | Brands | Response excerpt) — good for quickly scrolling through when you want to scan 8 responses in parallel.

List + Filters

On the index page /prompt-check:

• Full-text search 1: searches prompts + titles.
• Full-text search 2: searches the LLM responses themselves (slower, since it's a full-text scan).
• Filters for project, status, sentiment, platform — combinable.
• Sorting on all columns.
• 3-dot action menu per row (open, CSV/JSON export, delete).

Sharing

In the detail view: the "Share" button → a modal with an expiration date (1d/7d/30d/never), an optional password, and an optional max number of views. The generated link is copied to your clipboard. Recipients see the results at /share/prompt-check/{token} with Rankion branding (no login required).

Credits

1 credit per platform×mode call. Examples:

• ChatGPT blind + grounded + Claude blind = 3 credits
• All 6 platforms, both modes each = 12 credits

The exact total cost is shown to you in the estimator BEFORE you start.

API Access

All functionality is also available via the REST API — see the prompt-check.* skills.

• GET /api/v1/prompt-check — list with all filters (search, sentiment, platform, status, project_id)
• POST /api/v1/prompt-check — new run, async (202 Accepted + run_id)
• GET /api/v1/prompt-check/{id} — run + all responses
• GET /api/v1/prompt-check/{id}/export?format=csv|json — download
• POST /api/v1/prompt-check/{id}/rerun — re-run with the same settings
• DELETE /api/v1/prompt-check/{id} — soft delete

Chatbot Access

The Agentic Chat knows Live Prompt Check as a skill. There you can ask, for example:

• "Show my recent prompt checks about mobile phone plans"
• "Which runs contain Vodafone in the responses?"
• "Open run 42 — what did Claude answer?"
• "Start a new prompt check for '{prompt}'"

The bot then calls the right endpoints (scoped to your team) and gives you the results directly in the conversation, with deep links to the UI.

Background: Why Live and No Cache?

Every run makes REAL live API calls — NO caching by prompt hash. This is intentional:

• Knowledge changes — an LLM response today is not the same as tomorrow. Re-runs reveal drift.
• Live character — the module name promises it. Caching would mislead users.
• Brand sentiment can shift — if you have a funding round this week, you want to see that reflected in tomorrow's responses.

For reproducible snapshots at fixed dates: AI Visibility Tracking with configured prompts and a weekly frequency.