5/9/2026
This article is part of a short series drawn from Leveraging Patterns part of the my workshop. Here we focus on general pattern leverage: what you can ask models to do when you think in terms of patterns in text, not magic answers.
Language models are built on statistical patterns in language. That is useful when you frame work as pattern tasks: notice something, classify it, score it, predict from history, diff two versions, mine themes, generate in a style, or apply an edit when a pattern signals friction.
Below are practical verbs and examples—adapted from the AI Brainstorming Kit—so you can reuse them in prompts and workflows.
Detect
Flag passive voice, weak CTAs, or inconsistent tone across a landing page.
Identify
Classify a page as awareness vs. conversion intent from its copy structure.
Estimate
Infer reading age, persuasion score, or message clarity from sentence patterns.
Forecast
Predict which email subject lines will underperform based on phrasing patterns from past sends.
Compare
Diff two landing page variants to surface which copy elements changed and how.
Discover
Find which objection words appear before churn in support transcripts or exit surveys.
Generate
Produce headline variants, meta descriptions, or CTAs matching a brand’s proven tone patterns.
Act
Rewrite a weak CTA, shorten a sentence, or restructure a paragraph when patterns suggest friction.
References
The pattern verbs above (Detect, Identify, Estimate, Forecast, Compare, Discover, Generate, Act) are adapted from the AI Brainstorming Kit v1 (PDF) by the AI Design Kit project — credit to the original authors for the framing.
Further reading on why models are strong at text-side pattern tasks (structure in language, attention over context, scaling):
- Vaswani et al., Attention Is All You Need — transformer architecture and self-attention as the mechanism that weights patterns across the sequence you give the model.
- Brown et al., Language Models are Few-Shot Learners (GPT-3) — large LMs performing many tasks without weight updates, by conditioning on prompts and demonstrations (the empirical backbone behind “pattern leverage” in practice).
- Wei et al., Emergent Abilities of Large Language Models — survey-style framing of capabilities that appear mainly at scale; useful context for what “general” verbal skills can and cannot mean.
- Dong et al., A Survey on In-context Learning — overview of in-context learning: how examples and instructions steer behaviour at inference time.
- Jurafsky & Martin, Speech and Language Processing (3rd ed. draft) — textbook coverage of statistical NLP, representations, and linguistic structure that justify treating LLM work as pattern-based.
- Baroni, On the proper role of linguistically-oriented deep net analysis in linguistic theorizing — connects probing / analysis of neural LMs to linguistic theory (what it means for nets to encode grammatical patterns).
Also in this series
