---
id: "concept-prompt-based-optimization"
type: "concept"
source_timestamps: ["§ [ Stage 3 ] Make Other AI Agents Choose Your Brand"]
tags: ["prompt-engineering", "llm-optimization", "marketing-operations"]
related: ["claim-prompt-wording-alters-recommendations", "action-test-prompt-variations", "concept-share-of-model", "entity-perplexity"]
definition: "The ongoing marketing practice of testing and tuning how brand/product information is phrased so LLMs are more likely to surface and recommend it."
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-ext-18-preparing-brand-agentic-ai"
sourceUrl: "https://hbr.org/2026/03/preparing-your-brand-for-agentic-ai"
sourceTitle: "Preparing Your Brand for Agentic AI"
---
# Prompt-Based Optimization

**Prompt-Based Optimization** is the continuous marketing discipline of testing and tuning how brand and product information is phrased — and anticipating the exact phrasings consumers use — so that LLMs are more likely to surface and recommend the brand. It is the practitioner engine behind [[concept-share-of-model]].

Its empirical basis is [[claim-prompt-wording-alters-recommendations]] (Carnegie Mellon research showing synonym-level rewording can shift brand choice by up to 78.3%), and it is executed through [[action-test-prompt-variations]] (mining search logs and support transcripts for real phrasings, then testing product information across synonym variants). [[entity-perplexity-d18]]'s transparent, reasoning-style output offers a blueprint for reverse-engineering how models weigh price, compatibility, and reviews.

**Enrichment / verification.** This is consistent with well-documented LLM prompt sensitivity, which makes any brand-ranking strategy *probabilistic rather than deterministic*. A more rigorous methodological backbone than anecdotal prompting is **algorithmic auditing of recommender systems**: sample prompts, compare outputs across models and over time, and track mention rate, position, sentiment, and citation behavior. (This note is a synthesis hub added to resolve the extraction's dangling 'prompt-based optimization' references.)