Skip to main content

​
What are prompts?

Prompts are questions or queries that OneGlanse automatically submits to AI providers to test how they mention your brand. Each prompt represents a real user question that might trigger a response containing brand mentions. When you create a prompt, OneGlanse:
  1. Submits it to all enabled providers in your workspace
  2. Waits for complete responses using real browser automation
  3. Extracts the response text and citation sources
  4. Analyzes brand visibility, sentiment, and competitive positioning
  5. Stores results for historical tracking and trend analysis

​
Why prompts matter

Prompts are the foundation of your AI brand monitoring strategy. Well-crafted prompts help you: Discover brand visibility - Test if AI providers mention your brand for relevant queries Track competitive positioning - See how you rank against competitors in AI responses Monitor sentiment changes - Detect shifts in how AI describes your brand over time Identify content gaps - Find topics where AI providers lack current information about your brand Validate marketing claims - Check if AI responses reflect your positioning and messaging

​
How prompts work

​
Prompt structure

Each prompt contains: 
interface UserPrompt {
  id: string;           // Unique identifier
  user_id: string;      // Creator
  workspace_id: string; // Associated workspace
  prompt: string;       // The actual question
  created_at: string;   // Creation timestamp
}

​
Execution flow

When OneGlanse runs prompts, it follows this process:
1

Browser initialization

Launch a real browser session for each AI provider using Playwright automation:
// Provider-specific configuration
const config = {
  url: "https://chatgpt.com/",
  warmupDelayMs: 5000,
  requiresWarmup: true
};
2

Prompt submission

Type and submit the prompt through the provider’s interface:
async function executePrompt(
  page: Page,
  prompt: string,
  provider: Provider
): Promise<{ response: string; sources: Source[] }> {
  await askPrompt(page, prompt, provider);
  // Wait for response to complete
  const response = await fetchPromptResponses(page, provider);
  const sources = await checkAndExtractSources(page, provider);
  return { response, sources };
}
3

Response extraction

Wait for the AI to finish generating its response, then extract the complete text:
// Wait for streaming to complete
await waitForResponse(page);

// Extract markdown-formatted response
const response = await extractResponse(page);
4

Source collection

Extract citation sources if the provider includes them:
interface Source {
  title: string;      // Source article title
  cited_text: string; // Excerpt from the source
  url: string;        // Full URL
  domain: string;     // Domain (e.g., "example.com")
  favicon?: string;   // Favicon URL
}
5

Analysis

Analyze the response for brand mentions, sentiment, and competitive positioning. See Analysis for details.

​
Prompt payload

When prompts are scheduled to run, they’re bundled into a payload: 
interface PromptPayload {
  user_id: string;
  workspace_id: string;
  prompts: {
    id: string;
    prompt: string;
  }[];
  created_at: string;
}
This allows efficient batch processing across multiple prompts and providers.

​
Crafting effective prompts

​
Prompt categories

Different prompt types reveal different insights:
Test if AI providers recognize your brand by name:
Examples
"What is [Your Brand]?"
"Tell me about [Your Brand]"
"How does [Your Brand] work?"
Insight: Measures basic brand awareness and accuracy of AI knowledge
Test if your brand appears for relevant problem-solving questions:
Examples
"What are the best tools for [your category]?"
"How can I [solve specific problem]?"
"What should I use for [use case]?"
Insight: Reveals competitive positioning and market perception
See how you’re positioned against competitors:
Examples
"[Your Brand] vs [Competitor]"
"Compare [Your Brand] and [Competitor]"
"Which is better: [Your Brand] or [Competitor]?"
Insight: Shows head-to-head competitive narratives
Test if AI providers know about your key features:
Examples
"Tools with [specific feature]"
"Best [category] for [feature]"
"Does [Your Brand] support [feature]?"
Insight: Identifies knowledge gaps about product capabilities

​
Prompt best practices

Mirror real user language

Write prompts as real users would ask questions, not marketing copy. Use natural language and common search patterns.

Test multiple angles

Create prompts for different stages of the buyer journey: awareness, consideration, and decision-making.

Include competitors

Explicitly mention competitors in some prompts to understand comparative positioning.

Vary specificity

Balance broad category queries with specific feature or use case questions.

​
Prompt response data

Each prompt execution generates a response record: 
interface PromptResponse {
  // Identifiers
  id: string;
  prompt_id: string;
  prompt: string;
  
  // Context
  user_id: string;
  workspace_id: string;
  model_provider: string;
  
  // Response data
  response: string;        // Full AI response text
  sources: Source[];       // Citation sources
  
  // Analysis
  is_analysed: boolean;    // Whether brand analysis completed
  
  // Timing
  prompt_run_at: string;   // When the prompt was executed
  created_at: string;      // When the response was stored
}

​
Understanding prompt runs

Each time your scheduled job executes, it creates a new prompt run. The same prompt can have multiple runs over time: 
// Prompt runs are organized by timestamp
type PromptRunMap<T> = Record<
  string,              // prompt_id
  Record<
    string,            // prompt_run_at timestamp
    T[]                // responses from different models
  >
>;
This structure enables tracking how responses change over time: 
{
  "prompt_abc123": {
    "2026-03-01T09:00:00Z": [
      { model: "chatgpt", response: "...", ... },
      { model: "perplexity", response: "...", ... }
    ],
    "2026-03-01T15:00:00Z": [
      { model: "chatgpt", response: "...", ... },
      { model: "perplexity", response: "...", ... }
    ]
  }
}

​
Response validation

OneGlanse validates responses before storing them to ensure quality: 
// Responses must meet these criteria
if (!response || response.trim().length === 0) {
  throw new ExternalServiceError(
    provider,
    "Empty response extracted"
  );
}

const validation = validateResponse(response, provider);
if (!validation.valid) {
  throw new ValidationError(
    `Invalid response: ${validation.reason}`
  );
}
Invalid responses trigger automatic retries with the provider’s retry policy.

​
Monitoring prompt performance

​
Success metrics

Track these indicators to evaluate prompt effectiveness:
  • Mention rate - Percentage of responses that mention your brand
  • Visibility score - How prominently your brand appears (0-100)
  • Sentiment trend - Changes in positive/negative mentions over time
  • Position ranking - Where you rank in recommendation lists
  • Source diversity - Variety of sources AI providers cite about your brand

​
Example analysis

For the prompt “What are the best project management tools?”, you might see: 
{
  "model_provider": "chatgpt",
  "response": "Here are some popular project management tools:\n\n1. **Asana** - Great for team collaboration...\n2. **Monday.com** - Visual workflow management...\n3. **Jira** - Best for software development teams...",
  "sources": [
    {
      "title": "Best Project Management Software 2026",
      "domain": "techcrunch.com",
      "url": "https://techcrunch.com/..."
    }
  ]
}
From these responses, you can analyze:
  • Whether your brand appears in the list
  • Your ranking position (1st, 2nd, 3rd, etc.)
  • How you’re described compared to competitors
  • Which sources AI providers cite about your brand

Managing Prompts

Step-by-step guide to creating and organizing prompts

Providers

Learn how prompts are executed across different AI platforms

Analysis

Understand how prompt responses are analyzed for brand metrics

Interpreting Metrics

Make sense of visibility, sentiment, and position data