---
id: "action-test-prompt-variations"
type: "action-item"
source_timestamps: ["§ [ Stage 3 ] Make Other AI Agents Choose Your Brand"]
tags: ["prompt-engineering", "marketing-research"]
related: ["claim-prompt-wording-alters-recommendations", "concept-prompt-based-optimization"]
action: "Continuously test product information performance against subtle synonym variations in consumer prompts."
outcome: "Maximized likelihood of being the top recommended brand across diverse user queries."
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"
---
# Test Prompt Variations with Synonyms

**Action.** Do not rely on a single phrasing for SEO/LLM optimization. Analyze search logs and customer-service transcripts to find the exact phrasing consumers use, and test how your product information performs across subtle synonym variations — which can alter recommendation likelihood by over 70% (see [[claim-prompt-wording-alters-recommendations]]). This is the core loop of [[concept-prompt-based-optimization]].

**Outcome.** Maximized likelihood of being the top recommended brand across diverse user queries.

**Enrichment note.** Because LLM output varies with wording, context, and ordering, treat results as probabilistic; use algorithmic auditing to detect stable versus noisy shifts.