Pattern analysis

Pattern analysis is the engine under most of buzzabout's research. It's a technique, not a button: you point a natural-language question at a dataset and the assistant discovers an emergent taxonomy — it reads the posts, finds the recurring distinctions, and groups them into a short, ranked set of clusters along one axis you define per run. Ask a B2B SaaS project-management dataset "what pain points do people raise?" and you get clusters like onboarding friction, integrations, pricing objections, and missing reporting — each backed by the exact posts that say so. Ask a different question over the same posts and you get a different map.

This page explains what pattern analysis is and what you can get from it. Running it is the easy part — it's a single question in a chat, covered at the bottom.

What it can extract

The axis is open-ended: you choose it by how you phrase the question. The same technique surfaces, among others:

• pain points
• feature requests
• objections
• topics & themes
• frequently asked questions
• CTA types
• hooks
• tone of voice
• communication / narrative style
• emotions
• intent
• content categories

Each run reads across text and visual signals in the posts, scores the recurring ones, and clusters them into a handful of clearly distinct, named categories — then assigns each post to the cluster it best matches, with the citations to prove it.

Why it's useful

A dataset of thousands of posts is unreadable by hand. Pattern analysis collapses it into the handful of distinctions that actually recur — and, crucially, ranks them by how much they matter (reach × intensity × recurrence, not raw count), so a loud-but-rare complaint doesn't outweigh a quiet-but-common one. The output is a map you can act on: the themes to address, the language your audience uses, the objections to pre-empt, the gaps to fill — each traceable back to real mentions.

How to run it

There's no setup, no "Patterns" page, and no button — you just ask. Point it at a dataset that already has collected mentions (aim for ~100+ for meaningful clusters), open its chat, and ask a question that defines the axis you care about:

• "What pain points do people raise about project-management tools?"
• "What kinds of calls to action show up in these posts?"
• "What tones of voice do people use here?"

The assistant interprets it as a pattern question, runs a Pattern detection block in the conversation (asynchronous, ~4 minutes), and then summarises the discovered clusters — citing the underlying posts for any of them. Per-post assignments also show up as an optional Pattern Detections column on the dataset's canvas, so you can see exactly which cluster each mention landed in. Reading and re-reading results is free; only the original run charges credits.

Now analyse it

The built-in Skills turn the technique into ready-made reads, each pointing pattern analysis at a specific axis and shaping the result into a chart + table + takeaways. Open the skills picker from the chat input (Open skills book icon → Browse all skills), then Use this skill:

• Analyse pain points — the friction clusters, ranked, with handles to act on.
• Analyse feature requests — the product-directed asks buried in your mentions.
• Spot trends — which clusters are accelerating over time.

Gotchas

If you use the MCP tools directly, the chat trigger maps to buzzabout__create_pattern_detection_run (the run that charges credits), with buzzabout__get_pattern_detection_run to poll status and buzzabout__get_pattern / buzzabout__get_pattern_item to read the results for free.

Next steps