How to Use Web Search for Research on magicdoor.ai
Use current search-enabled chat models for sourced research on Magicdoor; older Perplexity Reasoning and Deep Research picker choices are retired.
Current access: Magicdoor's previous Perplexity Reasoning and Deep Research (pplx) chat picker models are retired. This URL remains for readers of the earlier research workflow; those models cannot be newly selected. Magicdoor now uses an Exa-backed web search tool with supported current chat models. Check the live model directory for availability before starting.
Research with current models and web search
- Choose a current search-enabled chat model. Use the model selection guide to check its tool and file support.
- Enable web search when the task needs current information. Ask a precise question, a date range, and for original URLs, for example: “Which official sources describe this policy as of today? Link each one.”
- Open and verify the original sources. Search results are leads, not proof; check publication dates and the exact supporting passage. Do not assume paywalled or private sources are accessible.
- Ask follow-up questions with the same model, or switch to another supported model for independent synthesis. A model change may affect tools and attachment support.
- Check the live cost indicator. Web search costs are reported by the search provider and add to the selected model's token charges; long answers and repeated searches may increase the total.
For a fact check, ask for evidence for and against the claim, then inspect primary sources. For a broader topic, split it into focused questions and compare the cited documents rather than expecting the retired Deep Research picker workflow. For a writing task, collect and verify URLs first, then ask Claude Sonnet 5.5 or another currently available model to synthesize with explicit citations.
Historical context: the older Perplexity models
Earlier versions of this guide recommended selecting Perplexity Reasoning for focused web-backed answers and Deep Research (pplx) for multi-step briefs from the Magicdoor model dropdown. Those were distinct chat picker entries with different token and request costs. Neither is a current Magicdoor chat model. Existing chat history is retained through runtime compatibility; the current web search tool is not a promise of the same output, depth, pricing, or routing. Perplexity's own products are separate from Magicdoor's model lineup.
See current search workflows and current model costs for the appropriate next step.
FAQs
Can I select Perplexity Reasoning or Deep Research on magicdoor.ai?
No. Perplexity Reasoning and Deep Research were older Magicdoor picker models and are no longer selectable. Choose a current search-enabled chat model and enable web search when the task needs fresh sources.
Do I need a separate Perplexity subscription for web search on magicdoor.ai?
No. Web search is a tool on supported Magicdoor chat models, not a separate selectable chat model. Web search costs are reported per request by the search provider, in addition to the selected model's usage.
Can I continue researching with Claude in the same conversation?
Yes. Choose a supported search-enabled model, gather source URLs, verify them, and switch to a supported Claude model for synthesis if it fits your task. Tool and file capabilities can differ between models.
Related Resources
Web Search on Magicdoor: Fact-Checking Guide
Use a current search-enabled chat model to research and verify sources. Magicdoor's older Perplexity Reasoning and Deep Research chat picker entries are retired.
Reasoning Models Comparison: Current Magicdoor Choices
Compare current Magicdoor reasoning-capable chat models using their standard rates and supported tools, without retired research picker entries.
What Happened to Perplexity Research on Magicdoor?
The older Perplexity Reasoning and Deep Research picker entries are retired. Learn how to research with current search-enabled chat models.
Claude Sonnet 5.5 with Image Generation and Web Search
Discover the limitations of Claude Sonnet 5.5 and learn how Magicdoor overcomes these challenges by combining Claude, web search, and image generation.