# Wrong Web Answers Trace to Fetch Tooling Not the Model ## Claim When an AI agent gives a wrong web-based answer, suspect the web-fetch tooling before blaming the model. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds for retrieval tasks on live or JavaScript-heavy pages, where default tools return raw HTML or URL lists the model must parse itself. - Breaks down for reasoning failures or knowledge-cutoff gaps that have nothing to do with fetching a page. - Doesn't establish that any specific vendor's fetch tool is superior — only that fetch quality is a variable worth checking before the model is. ## Evidence A Firecrawl product demo shows Claude Code's default web-fetch tool returning a URL list and raw HTML that the agent then has to parse itself; on JavaScript-heavy pages this fails and the model falls back to guessing from training data, producing an answer people blame on the model. Firecrawl's fix — rendering the page "like a real browser" and returning clean markdown — resolves the same prompts correctly, which locates the failure in the tool, not the model's reasoning. ## Links - [[3. Resources/Permanent Notes/Tool-Interface Choice Shapes Agent Architecture More Than Model Choice]] — because extends; this note narrows that general claim to the specific failure mode of web retrieval. - [[3. Resources/Permanent Notes/The Harness Now Outpaces the Model as the Source of Agent Capability Gains]] — because applies-in; fetch tooling is one of the harness-level levers that outpaces raw model capability. - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — because applies-in; diagnosing agent failure sources is part of the practical skill of working with agents day to day. ## Tests / falsifiers - Run the same JS-heavy-page prompt through the same model with default fetch vs. a browser-rendering fetch tool; if outputs remain equally wrong, the failure is not tooling. - Find a case where a rendering-capable fetch tool still produces a wrong answer on a page it correctly retrieved — that would locate the failure back in the model. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Readwise/Full Document Contents/Articles/Firecrawl MCP Server Search Scrape and Crawl the Web]] — vendor demo showing the raw-HTML failure mode and Firecrawl's clean-markdown fix.