You’re using Perplexity Deep Research Wrong

AI Research: Perplexity vs. OpenAI

Overview of AI Research Landscape

• The video discusses the increasing involvement of AI companies in deep research, highlighting Perplexity's entry into this competitive space.

• While Perplexity may not match OpenAI's premium research quality, it is deemed useful for various applications.

Strengths of Perplexity Research

• One major advantage of using Perplexity is its affordability, making it accessible to both free and pro users.

• The speed of response is another strength; typically, deep research returns results within a minute, allowing for follow-up questions.

• Users can switch between different AI models in Perplexity, providing flexibility that enhances project outcomes.

Limitations of Perplexity Research

• A significant limitation is the reliance on Reddit sources when social mode is activated, which can affect trustworthiness due to potential bias.

• The output context window is small, limiting the comprehensiveness of findings despite the ability to source multiple references.

• There are concerns about hallucinations in outputs where insights generated do not align with actual sources used.

Optimizing Deep Research Approach

• To maximize output quality from deep research, a structured planning process should be implemented since Perplexity lacks built-in planning features.

• Utilizing a reasoning model during the planning phase allows for better context provision and more effective use of reasoning capabilities.

Steps for Effective Competitive Analysis

• Begin by creating a custom "Perplexity space" to maintain context and set specific instructions for your research project.

• Generate a detailed research plan outlining goals and steps such as competitor identification and feature analysis while ensuring no external sources are referenced initially.

Structuring Reports and Templates

• After generating a research plan, request a structured template aligned with your objectives to ensure clarity in reporting findings.

• Ask for high-quality reputable sources that provide objective information while also identifying which sources should be excluded to avoid bias.

Finalizing Research Outputs

• Use Pro Search mode to gather recommended guidelines on credible sources relevant to your topic while avoiding marketing content or biased information.

How to Optimize Research Using Perplexity

Recommendations for Effective Research

• It is advised to upload research materials in a chat format and turn off social mode during deep research to avoid biased outputs, particularly from Reddit sources.

• The report template should include clear sections such as market leaders and emerging players, utilizing table formats for pricing information. Sources like Statista are preferred, though some less reliable sources may still be included.

Enhancing Output Quality

• To ensure balanced output that adheres to the research plan and source guidelines, it is crucial to export findings as a PDF report. However, note that Perplexity has limitations compared to OpenAI's Deep Research regarding context limits and potential hallucinations.

Customizing Research Instructions

• When conducting content strategy research for a podcast channel, create a new Perplexity space with custom instructions that indicate whether data is verified or an estimate.

• Generate a research plan focusing on identifying pain points of the target audience (parents) while ensuring reasoning models are utilized.

Source Recommendations and Deep Dive Insights

• Request reputable parenting website sources using standard Pro Search; this step enhances the quality of your research by providing credible references.

• Limit reliance on Reddit sources in custom instructions before starting deep research. Focus on specific topics for deeper insights rather than generating complete reports.

Validating Information and Exporting Findings

• When researching common audience pain points, ensure responses include verified markers indicating direct quotes from sources. This minimizes manual verification but still requires double-checking important figures.

• Conduct additional queries about how different parent segments consume content differently; export all findings into PDF files for comprehensive documentation.

Utilizing Extended Context Windows

• Upload all generated responses into a new chat to unlock extended context windows in Perplexity, enhancing insight synthesis capabilities.

Leveraging Intelligent Models for Content Strategy

• Switch off all search modes and select O3 models for intelligent synthesis of insights. Specific content cluster ideas can emerge from this approach due to tailored responses based on provided research plans.

Structuring Business Cases Effectively

Deep Research Process for Business Case Development

Initiating the Research Process

• Begin by uploading the business case outline and request research on the first section without in-text citations to generate a report incrementally.

• It's beneficial to include specific instructions, such as word limits for each section, to tailor the output effectively. The initial session utilized 54 sources, highlighting challenges in customer support like scalability and efficiency.

Customizing Outputs with Specific Data

• Uploading your own data (e.g., a business case) enhances relevance and specificity of the generated content. Maintain context by using the same chat window for continuous research without re-uploading outlines.

• For subsequent sections, over 20 sources were identified regarding objectives like improving satisfaction metrics and scaling operations through an implementation framework.

Building Sections Incrementally

• Each section is treated as a separate deep research query, allowing for comprehensive development while overcoming output context limits. Once all sections are completed, export them into PDF format for verification.

Verifying Accuracy of Claims

• Conduct manual fact-checking after generating reports; initially use AI tools like Perplexity to verify key statistics and claims within the business case.

• The tool flags verified claims (e.g., Salesforce data in existing challenges), while some claims remain unverified, necessitating further review or replacement with updated data.

Optimizing Fact Checking and Finalization

• This method streamlines manual fact-checking processes; consider cross-verifying with other AI models (like ChatGPT) to identify discrepancies.

• After verifying reports, request an executive summary from AI tools to ensure coherence before final formatting adjustments such as adding covers or tables.

Conclusion on Effective Use of AI Tools