Unlike traditional scrapers relying on CSS selectors, ScrapeGraphAI uses large language models to understand page content semantically. When a site redesigns, the AI adapts to the new structure automatically because it reads content contextually rather than following rigid selector paths. This dramatically reduces maintenance overhead.

SmartScraper extracts data from a single URL using a natural language prompt. SmartCrawler follows links across multiple pages on a site with configurable depth and page limits, extracting data from each page it visits. Use SmartScraper for individual pages and SmartCrawler when you need data from an entire section or site.

Yes. Both SmartScraper and SmartCrawler accept an output_schema parameter where you define a JSON schema for the extracted data. ScrapeGraphAI maps the page content to your schema, returning consistently structured results. This is especially useful when feeding extracted data into databases or downstream systems.

No. Extracted data is fetched in real-time during conversations and used only to respond to the current request. Tars does not maintain a persistent copy of scraped content. Each extraction is independent, and results are discarded after the agent delivers its response.

ScrapeGraphAI charges 10 credits per page in extraction mode, which uses AI to pull structured data. Markdown conversion mode costs just 2 credits per page, making it 80% cheaper. The agent can switch between modes based on whether you need structured extraction or plain text conversion.

Yes. For large crawl operations, ScrapeGraphAI returns a task ID immediately and processes the job asynchronously. The agent polls the status endpoint and notifies you when results are ready. This prevents conversation timeouts on multi-page crawls that may take several minutes to complete.

Traditional scraping APIs return raw HTML that still requires parsing. ScrapeGraphAI uses AI to understand and extract specific data points from pages using natural language prompts. You describe what you want in English instead of writing CSS selectors or XPath queries, and get structured JSON back directly.

The agent receives an empty or partial result set. You can submit feedback through the API's feedback endpoint to help improve future extractions. The agent handles this gracefully by informing the user and suggesting alternative approaches or manual verification for the missing data points.