As of 7/9/2025, Siftree is best suited for analyzing perspectives and narratives. The goal of Siftree is to create survey-grade results from observational data on social networks.

What works well:

• "How do people feel about AI in higher education?"

• "What narratives are forming around Iran’s threat to rebuild nuclear infrastructure?"

• "Do people back or oppose President Trump’s choice of 22-year-old Thomas Fugate to head DHS’s CP3 office?"

• "What consumer brands are people talking about positively?"

• "Who do people like better, Zohran Mamdani or Andrew Cuomo?"

These questions dive deep into the hearts and thoughts of people; they're focused on perspectives and opinions.

What doesn't work well:

• "What’s Walmart's engagement rate this week?"

• "Which of our TikToks performed the best by reach and shares?"

• "How many followers did Lebron gain last month?"

• "What's the week over week percent change for #coffee?"

• "What's our share of voice this quarter and did it increase?"

Siftree is not suited to answer these questions - they are best handled by traditional social media management tools.

Why don't we handle those types of questions?

We're focused on analyzing human thought: opinions, beliefs, preferences, and anything that effects collective consciousness and behaviors. That is what Siftree is optimizing for, and will continue to improve upon in the future.

How it works:

When you ask a question like:

"How are people reacting to the movie Sinners from a theatrical perspective?"

Siftree breaks down your query linguistically. What's the domain? What's the subject? What's the intent? What's the scope/qualifiers?

It uses the compartmentalized query to think about the type of response you're looking for, and maps your intent to 5 core strategies:

• Attitude: emotions, feelings, or reactions

• Theme: aspects, topics, or categories

• Entity: people, places, brands, etc.

• Contrast: sides, positions, or comparisons

• Explanation: reasons or causality

The agent uses these strategies to generate the most appropriate labels, after it looks at the data. The goal is to create the most statistically significant response to your query, in the desired format. It attempts to mimic a human analyst, and you're the boss. You ask it a question, and it will think about the best way to satisfy your needs with data analysis.

Once you accept the labels (you can edit, delete, and add labels) the analysis agent will begin a classification process. This process analyzes all of the relevant posts, comments, chats, etc. and classifies them according to the label they fit into best. This is how we're able to take an abstract query, generate specific ways to analyze the data, and then quantify it.

What to expect in the future:

• More public sources

• Longer historical periods

• Better analyses formats

  • Agents can surf the web

  • Agents can analyze more complex and nuanced perspectives

  • More types of data visualizations