Prompt runs

Overview

A prompt run (/prompts/[id]) is a working session where you ask AI how it answers your monitored buyer questions. Each run snapshots prompts from company prompts at creation time. AI responses are stored as corpus observations linked to those prompts (with the run id in observation metadata), including mention/citation flags.

On the run detail page you pick a platform and use Ask to fetch responses one at a time or in bulk. On Pro, you can enable web search for retrieval-aware answers that reflect what models find on the live web. Scheduled automations write into the same observation stream, so periodic runs build a timeline in the Obsurfable engine.

Who it's for

• Anyone running their first visibility check after setting up company prompts
• Pro users who re-run prompts with web search to track changes after content updates
• Teams reviewing individual AI answers before generating insights or comparing runs

Prerequisites

1. At least one company prompt in your library.
2. A prompt run—create one from Prompts with New run.
3. For web search: a Pro (or Agency) plan.

Step-by-step

Open a prompt run

1. From /prompts, click a run or create New run.
2. You'll land on /prompts/[id] with the run overview: question count and last updated time.

Ask a platform

1. Browse the list of questions in the run.
2. Use the platform picker next to the Ask button to choose ChatGPT or Gemini. The button label follows your choice.
3. Click Ask on an individual prompt, or use bulk actions to run multiple prompts.
4. Wait for each response—Obsurfable stores the full AI answer and evaluates mention/citation flags.
5. Switch platform and ask again to capture a second answer to the same question. Results are stored separately per platform, so nothing is overwritten.

See Supported AI platforms for what each platform covers.

Enable web search (Pro)

1. On the run page, find the Use web search toggle.
2. Turn it on before asking prompts—responses will use live web retrieval: OpenAI web search on ChatGPT, Google Search grounding on Gemini.
3. Free plan users see an upgrade prompt instead of the toggle.

Review results

1. Once results are saved, the run becomes read-only to preserve the snapshot.
2. Each result shows whether your product was mentioned, your domain was cited, and which competitors appeared. When a run covers more than one platform, each result carries a platform badge.
3. Read the full AI response text for context on framing and sources.
4. Expand Why this answer? on any result to see the recommendation trace — prompt, entities, sources, brands considered, and final answer.

Check score impact and insights

1. After results are saved, scroll to Score impact to see how this run affects your visibility score.
2. Click View insights to generate or open Insights for this run (Pro).
3. Visit Visibility once you've run at least 3 prompts.

Edit an incomplete run

1. If no results are saved yet, you can add, remove, or edit prompts in the run.
2. Once any result is saved, the prompt list locks—create a new run to test again.

Tips

• Run at least 3 prompts to unlock your full AI Visibility Score report.
• Use web search on Pro for answers closer to what users see in ChatGPT with browsing or Gemini with grounding enabled.
• Ask your most important prompts on both platforms. ChatGPT and Gemini read different search indexes, so the answers often diverge.
• Ask discovery prompts first—they reveal category gaps where competitors may appear instead of you.
• After site changes, create a new run and use Compare runs to see what moved.

FAQ

What's the difference between asking with and without web search?

Without web search, the model answers from its training knowledge. With web search (Pro), it retrieves live web content—closer to how ChatGPT answers with browsing enabled, or Gemini with Google Search grounding.

Claude and Perplexity retrieval are coming soon — the app shows these platforms as "Coming soon" on the AI Brand Health dashboard until integrations ship.

Why is my run read-only?

Runs with saved results are locked snapshots. This ensures compare runs and visibility scoring reference consistent data.

Can I re-run a prompt on Free?

Each company prompt can be run once per platform on the Free plan, so you can ask the same question on ChatGPT and on Gemini. Upgrade to Pro to re-run prompts and use web search.

How long do runs take?

Each prompt is one AI call per platform. A run with 20 prompts on both ChatGPT and Gemini means up to 40 sequential requests—run a subset first if you're testing.

Where do I see mention and citation details?

On the run page in each result card. For a side-by-side view across two dates, use Compare runs.

How do I get recommendations from a run?

On Pro, open Insights from the run page or go to /insights/for-run/[promptRunId] to generate AEO/GEO recommendations.