LLM-Based Intelligent Notification Composition: From Static Personalization to Context-Aware Persuasive Messaging

Nilesh Agrawal

cs.HCMar 22, 2026

LLM-Based Intelligent Notification Composition: From Static Personalization to Context-Aware Persuasive Messaging

Authors: Nilesh Agrawal

Organizations: Independent Researcher, Seattle, WA, USA

Abstract

Push notifications remain among the most direct channels through which digital platforms engage users, yet existing approaches have invested heavily in who to notify, when to notify, and what to recommend, while leaving how to communicate as the least-optimized stage. This paper argues that message quality is an independent, underinvested lever, and that LLMs create their most differentiated value precisely at this layer. We make three contributions. First, we define notification message quality along six dimensions (contextual relevance, clarity, actionability, novelty handling, linguistic freshness, and persuasive appropriateness) and show how LLM-based composition improves each relative to templates. Across reviewed deployments, reported improvements range from +8% to +14.5% CTR over static templates and +1% to +2.5% over mature slot-filling systems, though these span heterogeneous systems and should not be treated as directly comparable. Second, we provide an architectural attribution analysis disentangling message generation from adjacent components (targeting, ranking, timing), arguing that observed gains are frequently misattributed to text generation alone. Third, we introduce a three-criterion decision framework specifying when LLM generation is and is not the binding constraint. We support these arguments through a PRISMA-guided survey (28 sources from 142 screened), examine domain-specific applications across social media, food delivery, and e-commerce, and propose a unified architectural framework with budget-aware routing, grounded generation, candidate ranking, diversity controls, and online learning.

Figures & tables

Dimension	Definition	Template Capability	LLM Capability
Contextual Relevance	Connects recommendation to user’s current moment: time, location, behavior	Weak: only static slot values	Strong: composes multiple signals naturally
Clarity	Immediately comprehensible on a small notification surface	Variable: slot grammar often awkward	Flexible: optimizes phrasing for brevity
Actionability	Makes the next step obvious and motivating	Formulaic call-to-action	Context-sensitive framing of action
Novelty Handling	Introduces new items in a way that feels appealing, not random	Poor: novelty reads as irrelevance	Better: bridges from known to adjacent
Linguistic Freshness	Avoids structural repetition that trains users to ignore the channel	Low: same structure across exposures	High: semantic variety across exposures
Persuasive Appropriateness	Motivates action without deceptive or manipulative framing	Neutral: limited expressive range	Variable: requires explicit guardrails

Table 1: Notification message quality dimensions and comparative assessment of template-based versus LLM-based composition.

System	Domain	Technology	Primary Dims	Key Result	Tier
PushGen	Social Media	SFT + Pairwise Reward	Freshness, Novelty, Relevance	+14.08% CTR; $3.2\times$ lexical diversity	T1
Instagram	Social Media	Similarity Penalty	Freshness, Novelty	Better CTR with lower volume	T2
DoorDash	Food Delivery	GNN + Contextual Copy	Relevance, Clarity	+1.8% engagement lift (framing only)	T2
E-Commerce	Retail	LLM Content Optimization	Actionability, Persuasion	+12.5% CTR, +8.3% CVR	T1
LinkedIn	Professional	Heterogeneous Treatment	Per-cohort tone	Reduced churn via personalized caps	T1
Duolingo	EdTech	Contextual Bandit	Timing, Freshness	Increased DAU	T1

Table 2: Summary of production evidence linking LLM-based composition to engagement outcomes. CTR = click-through rate; CVR = conversion rate. Full references: PushGen ( Bie et al., 2026 ) ; Instagram ( Sun et al., 2025 ) ; DoorDash ( Sinha, 2024 ) ; E-Commerce ( ACM Digital Library, 2025 ) ; LinkedIn ( Tu et al., 2021 ) ; Duolingo ( Yancey et al., 2020 ) .

Component	Primary Contribution	LLM Relevance
Audience Selection	Ensures budget is spent on users with highest marginal return	None (a targeting problem)
Content Ranking	Ensures the underlying item is relevant to the user	None (a relevance problem)
Send-Time Optimization	Capitalizes on user routines; minimizes interruption cost	None (a timing problem)
Message Generation	Adapts tone, length, hooks, and framing to context	Direct: the message quality problem
Candidate Ranking	Aligns generation with historical CTR; filters poor outputs	Direct: reward model selects best candidate

Table 3: Architectural attribution matrix: which pipeline components contribute to engagement and where LLMs have direct relevance.

Pattern	Main Strength	Main Weakness	Key Failure Mode	Quality Dims Served
Template / Slot	Low latency, high control	Low diversity, weak context	Message fatigue	(Baseline)
RAG	Context-aware grounding	Retrieval latency, hallucination	Faithfulness failure	Relevance, Clarity
PEFT / LoRA	Hot-swappable style control	Requires style curation	Overfitting to tone	Freshness, Persuasion
Pairwise Reward	Robust relative ranking	Exposure bias, delayed reward	Reward hacking	Actionability, Novelty
Diversity Penalty	Reduces fatigue	Can suppress short-term CTR	Under-delivery	Freshness, Novelty
Contextual Bandit	Adapts to user routines	Exploration cost, sparse data	Cold-start timing	(Timing, adjacent)

Table 4: Comparative synthesis of architectural patterns, mapped to the quality dimensions they primarily serve.

Domain	Critical Quality Dimensions	LLM Primary Value	Dominant Risk
Social Media	Freshness, Novelty Handling	Semantic variety; reduced fatigue	Over-frequency; tone mismatch
Food Delivery	Contextual Relevance, Clarity	RAG-grounded temporal framing	False urgency; stale ETA claims
E-Commerce	Actionability, Persuasive Framing	Dynamic frame selection	Dark patterns; hallucinated prices

Table 5: Domain-specific mapping: which quality dimensions matter most and where LLMs create the greatest differentiation.

Panel of Figure 1 — Figure 1: Unified LLM-based notification pipeline. Solid arrows: generation flow. Dashed arrows: online learning feedback loop. The budget-aware router uses the three-criterion framework to decide per-event whether to invoke the LLM path or fall back to templates.

Personalized Text Generation

Personalized text generation adapts language-model outputs to a user's writing style, preferences, history, and context.

See all 23 papers

Oct 17, 2025

POPI: Personalizing LLMs via Optimized Natural Language Preference Inference

Large language models (LLMs) are typically aligned with population-level preferences, despite substantial variation across individual users. We introduce POPI, a user-level personalization framework that separates the problem into two components connected by a natural-language interface: a shared inference model that distills heterogeneous user signals into a concise preference summary, and a shared generator that conditions on this summary to produce personalized responses. Both components are trained under a unified preference-optimization objective, with reinforcement learning handling the non-differentiable inference step. This objective decomposes into generator approximation error and summary informativeness, revealing how a single loss simultaneously drives accurate generation and informative summarization. Because the interface is natural language, learned summaries can be inferred once per user and reused across different generators -- including frozen, black-box commercial APIs. Across four personalization benchmarks, POPI generally improves personalization quality while reducing context overhead by up to an order of magnitude.

Yizhuo Chen, Xin Liu, Ruijie Wang +7

University of Illinois Urbana-Champaign · Amazon · University of Notre Dame

LLM Personalization Personalized Language Models Personalized Language Model Alignment Language Model Post-Training User Preference Modeling

Ask PDF

Mar 4, 2026

When Agents Persuade: Rhetoric Generation and Mitigation in LLMs

Despite their wide-ranging benefits, LLM-based agents deployed in open environments can be exploited to produce manipulative material. In this study, we task LLMs with propaganda objectives and analyze their outputs using two domain-specific models: one that classifies text as propaganda or non-propaganda, and another that detects rhetorical techniques of propaganda (e.g., loaded language, appeals to fear, flag-waving, name-calling). Our findings show that, when prompted, LLMs exhibit propagandistic behaviors and use a variety of rhetorical techniques in doing so. We also explore mitigation via Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and ORPO (Odds Ratio Preference Optimization). We find that fine-tuning significantly reduces their tendency to generate such content, with ORPO proving most effective.

Julia Jose, Ritik Roongta, Rachel Greenstadt

Department of Computer Science and Engineering, New York University, New York, NY, USA

Persuasion in Language Models LLM Agent Safety Language Model Safety Evaluation

Ask PDF

Oct 27, 2026

LCM-Promo: An LLM-Based Uplift Model for Promotion Allocation

We present LCM-Promo, an LLM-based uplift model for promotion allocation. LCM-Promo is a task-specific Large Commerce Model (LCM): a finetuned LLM trained on transactional commerce data to predict user intent and promotion response. It learns monotonic promotion-response curves from production logs by interpreting single-token outputs as conversion-threshold probabilities, while conditioning on session context such as day and delivery shift. In a production A/B test on churn reactivation, LCM-Promo increased conversion by 1% and reduced promotion cost by 6% relative to the Segment-EMA production baseline.

Nesta Midavaine, Floris Fok, Paulo Mendes +4

Prosus, Amsterdam, The Netherlands · iFood, São Paulo, Brazil

Uplift Modeling Conversion Rate Prediction Language Model Post-Training

Ask PDF

Abstract

Personalized Text Generation

Explore similar work

POPI: Personalizing LLMs via Optimized Natural Language Preference Inference

When Agents Persuade: Rhetoric Generation and Mitigation in LLMs

LCM-Promo: An LLM-Based Uplift Model for Promotion Allocation