AI Search Engines - OpenSight

Overview

OpenSight monitors your brand visibility across the three major AI search engines: ChatGPT, Perplexity, and Google AI Overview. Each engine has unique characteristics, response formats, and ranking factors.

Supported Engines

ChatGPT

OpenAI’s conversational AI with search capabilities

Perplexity

AI-powered answer engine with real-time web search

Google AIO

Google’s AI Overview in search results

Engine Architecture

OpenSight uses a unified client architecture to query all engines consistently:

// From: packages/engine-clients/src/types.ts:1-18
export interface EngineQuery {
  prompt: string;
  brandName: string;
  brandUrl: string;
}

export interface EngineResponse {
  engine: 'chatgpt' | 'perplexity' | 'google_aio';
  responseText: string;
  citationUrls: string[];
  rawResponse: unknown;
  queriedAt: Date;
}

export interface EngineClient {
  name: string;
  query(input: EngineQuery): Promise<EngineResponse>;
}

ChatGPT

Overview

ChatGPT uses OpenAI’s GPT-4o-mini model to generate conversational responses with search capabilities.

Implementation

// From: packages/engine-clients/src/chatgpt.ts:1-46
export class ChatGPTClient extends BaseEngineClient {
  name = 'chatgpt';
  private client: OpenAI;

  constructor(apiKey?: string) {
    super();
    const key = apiKey || process.env.OPENAI_API_KEY;
    if (!key) {
      throw new Error('OPENAI_API_KEY is required for ChatGPT client');
    }
    this.client = new OpenAI({ apiKey: key });
  }

  async query(input: EngineQuery): Promise<EngineResponse> {
    const prompt = `${input.prompt}\n\nContext: This query is for ${input.brandName} (${input.brandUrl})`;

    const response = await this.retryWithExponentialBackoff(async () => {
      return await this.client.chat.completions.create({
        model: 'gpt-4o-mini',
        messages: [
          {
            role: 'user',
            content: prompt,
          },
        ],
      });
    });

    const responseText =
      response.choices[0]?.message?.content || 'No response received';
    const citationUrls = this.extractUrlsFromText(responseText);

    return {
      engine: 'chatgpt',
      responseText,
      citationUrls,
      rawResponse: response,
      queriedAt: new Date(),
    };
  }
}

Characteristics

Response Format
Ranking Factors
Configuration

Conversational Style

Natural language responses
Often includes context and explanations
May cite sources inline with numbered references

Citation Handling

URLs extracted from response text
Citations may be embedded in sentences

Model: gpt-4o-miniAPI Key: Set OPENAI_API_KEY environment variableRate Limiting: Handled with exponential backoff retry

ChatGPT queries include brand context to help the model understand the search intent:

const prompt = `${input.prompt}\n\nContext: This query is for ${input.brandName} (${input.brandUrl})`;

Perplexity

Overview

Perplexity is an AI-powered answer engine that performs real-time web searches and synthesizes information from multiple sources.

Implementation

// From: packages/engine-clients/src/perplexity.ts:1-49
export class PerplexityClient extends BaseEngineClient {
  name = 'perplexity';
  private client: OpenAI;

  constructor(apiKey?: string) {
    super();
    const key = apiKey || process.env.PERPLEXITY_API_KEY;
    if (!key) {
      throw new Error('PERPLEXITY_API_KEY is required for Perplexity client');
    }
    this.client = new OpenAI({
      apiKey: key,
      baseURL: 'https://api.perplexity.ai',
    });
  }

  async query(input: EngineQuery): Promise<EngineResponse> {
    const prompt = `${input.prompt}\n\nContext: This query is for ${input.brandName} (${input.brandUrl})`;

    const response = await this.retryWithExponentialBackoff(async () => {
      return await this.client.chat.completions.create({
        model: 'sonar',
        messages: [
          {
            role: 'user',
            content: prompt,
          },
        ],
      });
    });

    const responseText =
      response.choices[0]?.message?.content || 'No response received';
    const citationUrls = this.extractUrlsFromText(responseText);

    return {
      engine: 'perplexity',
      responseText,
      citationUrls,
      rawResponse: response,
      queriedAt: new Date(),
    };
  }
}

Characteristics

Response Format
Ranking Factors
Configuration

Search-Optimized

Direct answers backed by sources
Often includes statistics and data
Strong emphasis on citations

Real-Time Search

Performs live web searches
Synthesizes multiple sources
Favors recent, authoritative content

Model: sonarAPI Key: Set PERPLEXITY_API_KEY environment variableBase URL: https://api.perplexity.aiRate Limiting: Handled with exponential backoff retry

Perplexity uses OpenAI’s SDK with a custom base URL to access their API.

Google AI Overview

Overview

Google AI Overview (formerly SGE - Search Generative Experience) provides AI-generated summaries at the top of Google search results.

Implementation

OpenSight uses the Serper API to access Google search results:

// From: packages/engine-clients/src/google-aio.ts:1-87
export class GoogleAIOClient extends BaseEngineClient {
  name = 'google_aio';
  private apiKey: string;

  constructor(apiKey?: string) {
    super();
    const key = apiKey || process.env.SERPER_API_KEY;
    if (!key) {
      throw new Error('SERPER_API_KEY is required for Google AIO client');
    }
    this.apiKey = key;
  }

  async query(input: EngineQuery): Promise<EngineResponse> {
    const searchQuery = `${input.prompt} ${input.brandName}`;

    const response = await this.retryWithExponentialBackoff(async () => {
      const res = await fetch('https://google.serper.dev/search', {
        method: 'POST',
        headers: {
          'X-API-KEY': this.apiKey,
          'Content-Type': 'application/json',
        },
        body: JSON.stringify({
          q: searchQuery,
          gl: 'us',
          hl: 'en',
        }),
      });

      if (!res.ok) {
        throw new Error(`Serper API error: ${res.statusText}`);
      }

      return (await res.json()) as SerperResponse;
    });

    let responseText = '';
    const citationUrls: string[] = [];

    // Extract answer box if present
    if (response.answerBox?.snippet) {
      responseText = response.answerBox.snippet;
      if (response.answerBox.sources) {
        response.answerBox.sources.forEach((source) => {
          citationUrls.push(source.link);
        });
      }
    }

    // Extract organic results
    if (response.organic) {
      if (!responseText) {
        responseText = response.organic
          .slice(0, 3)
          .map((result) => `${result.title}: ${result.snippet}`)
          .join('\n\n');
      }
      response.organic.forEach((result) => {
        citationUrls.push(result.link);
      });
    }

    return {
      engine: 'google_aio',
      responseText: responseText || 'No results found',
      citationUrls: [...new Set(citationUrls)],
      rawResponse: response,
      queriedAt: new Date(),
    };
  }
}

Characteristics

Response Format
Ranking Factors
Configuration

Answer Box + Organic Results

AI-generated answer box (when available)
Falls back to top 3 organic results
Rich citation data

Serper Response Structure

interface SerperResponse {
  answerBox?: {
    snippet?: string;
    sources?: Array<{ title: string; link: string }>;
  };
  organic?: Array<{
    title: string;
    link: string;
    snippet: string;
  }>;
}

API: Serper API (https://google.serper.dev/search)API Key: Set SERPER_API_KEY environment variableParameters:

q: Search query (prompt + brand name)
gl: Geographic location (default: us)
hl: Language (default: en)

Rate Limiting: Handled with exponential backoff retry

Google AI Overview availability varies by query type and geographic location. Not all searches trigger an AI Overview.

Retry & Error Handling

All engine clients inherit retry logic from BaseEngineClient:

// From: packages/engine-clients/src/base.ts (referenced in other files)
abstract class BaseEngineClient implements EngineClient {
  abstract name: string;
  abstract query(input: EngineQuery): Promise<EngineResponse>;
  
  // Exponential backoff retry for rate limits and transient errors
  protected async retryWithExponentialBackoff<T>(
    fn: () => Promise<T>,
    maxAttempts = 3
  ): Promise<T> {
    // Implementation includes:
    // - Retry on rate limits (429)
    // - Exponential backoff (2s, 4s, 8s)
    // - Max 3 attempts by default
  }
  
  // Extract URLs from response text for citation tracking
  protected extractUrlsFromText(text: string): string[] {
    // URL extraction regex logic
  }
}

Retry Configuration

Retry Behavior

Default Settings:

Max Attempts: 3
Backoff: Exponential (2s → 4s → 8s)
Retry Triggers: Rate limits (429), transient errors

Used in Alert Checker:

// From: apps/api/src/jobs/alert-checker.ts:356-362
runJob('alert-checker', data, alertCheckerProcessor, {
  attempts: 3,
  backoffMs: 2000,
})

Engine-Specific Scores

OpenSight tracks separate visibility scores for each engine:

// From: apps/api/src/services/brand.service.ts:273-277
{
  overallScore: snapshot.overallScore,
  chatgptScore: snapshot.chatgptScore ?? 0,
  perplexityScore: snapshot.perplexityScore ?? 0,
  googleAioScore: snapshot.googleAioScore ?? 0,
  // ...
}

Score Interpretation

ChatGPT Score

Measures visibility in ChatGPT responses0-100 scale

Perplexity Score

Measures visibility in Perplexity answers0-100 scale

Google AIO Score

Measures visibility in Google AI Overview0-100 scale

The overall visibility score is calculated across all three engines, providing a holistic view of your AI search presence.

Best Practices by Engine

ChatGPT Optimization

Focus on:

Conversational, natural language content
Clear explanations with context
Authoritative sources and citations
Comprehensive, well-structured answers

Content Tips:

Write in a helpful, educational tone
Include examples and use cases
Address common follow-up questions
Link to authoritative external sources

Perplexity Optimization

Focus on:

Data-driven, fact-based content
Strong citation practices
Recent, up-to-date information
Domain authority and trust signals

Content Tips:

Include statistics and research findings
Cite reputable sources prominently
Update content regularly with fresh data
Add schema markup for key entities

Google AIO Optimization

Focus on:

Traditional SEO best practices
E-E-A-T signals (expertise, authority, trust)
Fast-loading, mobile-friendly pages
Structured data and schema markup

Content Tips:

Optimize for traditional search ranking
Build high-quality backlinks
Demonstrate expertise with author bios
Implement comprehensive schema markup

Environment Variables

Configure API keys for each engine:

# ChatGPT
OPENAI_API_KEY=sk-...

# Perplexity
PERPLEXITY_API_KEY=pplx-...

# Google AI Overview (via Serper)
SERPER_API_KEY=...

Missing API keys will cause engine clients to throw errors on initialization. Ensure all required keys are set in your environment.

Engine Status & Monitoring

OpenSight tracks query success/failure for each engine in visibility snapshots. If an engine is down or rate-limited, the system:

Retries with exponential backoff (up to 3 attempts)
Logs errors for monitoring
Continues processing other engines
Alerts users if all engines fail

API Reference

Query All Engines

Querying is handled automatically by the brand monitoring system. Each engine is queried with:

{
  prompt: "What are the best project management tools?",
  brandName: "Acme Corp",
  brandUrl: "https://acme.com"
}

Engine Responses

All engines return a standardized response:

{
  engine: 'chatgpt' | 'perplexity' | 'google_aio',
  responseText: string,
  citationUrls: string[],
  rawResponse: unknown,
  queriedAt: Date
}

Next Steps

Brand Monitoring

See how engines contribute to overall visibility

Content Scoring

Optimize content for each engine

Competitor Tracking

Compare performance across engines

Alerts

Get notified of engine-specific changes

​Overview

​Supported Engines

ChatGPT

Perplexity

Google AIO

​Engine Architecture

​ChatGPT

​Overview

​Implementation

​Characteristics

​Perplexity

​Overview

​Implementation

​Characteristics

​Google AI Overview

​Overview

​Implementation

​Characteristics

​Retry & Error Handling

​Retry Configuration

​Engine-Specific Scores

​Score Interpretation

ChatGPT Score

Perplexity Score

Google AIO Score

​Best Practices by Engine

​Environment Variables

​Engine Status & Monitoring

​API Reference

​Query All Engines

​Engine Responses

​Next Steps

Brand Monitoring

Content Scoring

Competitor Tracking

Alerts

Overview

Supported Engines

Engine Architecture

ChatGPT

Overview

Implementation

Characteristics

Perplexity

Overview

Implementation

Characteristics

Google AI Overview

Overview

Implementation

Characteristics

Retry & Error Handling

Retry Configuration

Engine-Specific Scores

Score Interpretation

Best Practices by Engine

Environment Variables

Engine Status & Monitoring

API Reference

Query All Engines

Engine Responses

Next Steps