ICP & Personas

Set up your ICPs and brand keywords so AEKO persona analysis is sharper and mentions are caught accurately.

7 min read

Brand Keywords — Domain Level

The list of keywords used to detect your brand in AI responses lives at the domain level. Go to Settings → Domain Management, expand a domain, and edit the brand_keywords array.

Example

["grafen", "그라펜", "グラフェン"]

These keywords are ILIKE-matched against AI response text to drive mention counts. If you're a Korean seller monitoring overseas markets, register Korean, English, Japanese, and Chinese spellings so mentions across every market are caught.

Caveat

Registering generic words ("beauty", "skincare") explodes noise matches and distorts your mention metrics. Stick to brand names, sub-brands, and common misspellings.

What Is an ICP

An ICP (Ideal Customer Profile) is one concrete description of your core target buyer. You can create multiple ICPs per domain to separate analysis by market or segment.

Every tracked prompt is automatically classified to its closest ICP (gpt-4o-mini runs the classification at save time). The Analyze → Persona page uses that classification to compare which personas remember your brand vs. competitors.

Defining an ICP

From Improve → Brand Settings, click + Add ICP. You'll see these fields:

Field	Example
Name	"LA Gen-Z K-beauty newcomer"
Persona Type	One of four (see below)
Target Country	US, KR, JP — ISO code
Age Range	"20–29", "late 30s" — free text
Interests	"sensitive skin, vegan cosmetics, Korean trends"
Purchase Motivation	"Wants to try the K-beauty trends they saw on TikTok"
Preferred Channels	"Instagram, TikTok, Sephora in-store"
Notes	Free text — Sonnet references this when generating PDP and content

The four persona types

Price Sensitive (price_sensitive) — price and discounts drive the decision
Premium Oriented (premium_oriented) — quality and scarcity outweigh price
Brand Loyal (brand_loyal) — repeat buyers of trusted brands
Feature Focused (feature_focused) — specs, performance, comparison-driven

Status — Draft vs Active

ICPs can be saved as Draft for review, then promoted to Active. Only Active ICPs surface in the Persona analysis page and feed into content generation.

How Persona Analysis Works

The Analyze → Persona page lays out one row per ICP and shows your brand's exposure vs. competitors in the prompts mapped to that ICP.

Data flow

At prompt-tracking registration, gpt-4o-mini compares ICP text fields against the prompt and maps it to the closest ICP (user_prompts.icp_id)
The Persona page aggregates response data per ICP
Keyword-graph nodes come from comma-splitting the ICP's interests field
Persona-type distribution groups ICPs by persona_type_id

Overriding the mapping

If you don't like an auto-mapping, change the ICP via the dropdown in the prompt list at /dashboard/improve/brand. Changes apply from the next analysis cycle.

Persona Opportunity Analysis

The real question on the Persona page is not "which personas like us" but "which personas know our competitors but don't know us." That gap is your next-quarter content priority.

Signal	Meaning	Recommended action
You ↑ / Competitors ↑	Red ocean — actively contested	Differentiation content / PDP polish
You ↑ / Competitors ↓	Monopoly — defend it	Messaging consistency, reinforce citation sources
You ↓ / Competitors ↑	Opportunity — addressable market	Create new content targeting that ICP
You ↓ / Competitors ↓	Cold market — small or immature	Deprioritize, monitor only

Practical Tips

Recommended setup order

1. In Domain Management, register multilingual brand_keywords (5 min)
2. Define 2–3 ICPs for your primary market (30 min)
3. Promote them to Active and wait one analysis cycle (~3 days)
4. Check the Persona page for gaps → generate content variation guides
5. Revisit ICPs and keywords each quarter as markets shift

Patterns to avoid

Making 10+ ICPs — data gets diluted and insights blur
Registering generic words as brand_keywords — noise matches explode

Reading Your Results

Measure — AI Readiness