Generative AI to Industrialize Phonetic Naming and Copywriting

Why AI is the natural partner of sound symbolism

Sound symbolism relies on two massive operations:

1. Generation: produce numerous and varied candidates that respect a precise sonic profile.
2. Scoring: evaluate each candidate on objectifiable phonosymbolic dimensions.

Before AI, both operations were slow, expensive and fragile. A handful of names per day, validated by gut feel. Today, an AI-equipped copywriter can generate and score 500 candidates in an hour, then iterate several cycles within the same day.

AI does not replace the marketer's trained ear. It gives that ear a thousand-times speed boost, capable of hearing a thousand variants where it used to hear ten.

Five-step naming pipeline

graph LR
    A[1. Phonetic brief] --> B[2. Candidate generation]
    B --> C[3. Legal & linguistic filtering]
    C --> D[4. Phonosymbolic scoring]
    D --> E[5. Final consumer test]

Step 1 — The structured phonetic brief

A solid AI naming brief contains seven elements:

• Product category and main competitors.
• One-sentence business promise.
• Emotional positioning.
• Target phonetic profile on four axes (shape, size, speed, texture).
• Desired length and structure (syllable count, ending type).
• Target languages (FR, EN, ES, other).
• Constraints (must contain X letter, avoid Y sound, etc.).

Step 2 — Name generation prompt

You are a naming specialist trained in sound symbolism.

BRIEF
- Category: [category]
- Main competitors: [brands]
- Promise: [sentence]
- Emotional positioning: [soft/edgy/elegant/fun/...]

TARGET PHONETIC PROFILE (1 to 5 scale)
- Shape (1 = very round, 5 = very angular): [X]
- Size (1 = large/heavy, 5 = small/light): [X]
- Speed (1 = slow, 5 = fast): [X]
- Texture (1 = warm/organic, 5 = cold/techno): [X]

CONSTRAINTS
- Syllables: [2 to 3]
- Target languages: [FR, EN]
- Must avoid: [forbidden sounds]
- Should include if possible: [desired sounds]

TASK
Generate 50 original, non-existing candidate names, pronounceable in all
target languages. For each name, give its phonetic structure
(C = consonant, V = vowel) and explain in one line why it fits the
target profile.

Output: a table with columns Name | Structure | Rationale.

Step 3 — Automatic filtering

AI can pre-filter:

• Existing trademarks: cross-check against INPI/USPTO/EUIPO via API, or have the AI flag obvious hits.
• Available domains: integrate with a registrar script.
• Negative connotations in other languages: dedicated prompt that scans the 10 main languages.

Linguistic filtering prompt:

For each name in the list below, check whether the word or a near homophone
evokes anything NEGATIVE, RIDICULOUS or TABOO in these languages:
French, English, Spanish, German, Italian, Portuguese, Dutch, Polish,
Mandarin (pinyin), Japanese (rōmaji).

List: [names]

Output: table Name | Status (OK / Avoid) | Reason if Avoid.

Step 4 — Phonosymbolic scoring with AI

This is where the magic happens. Ask the AI to score each name on the four dimensions:

You are an expert in sound symbolism with deep articulatory linguistics
background.

For each name in the list, assign a score from 0 to 10 on the 4 dimensions
below, then compute a COHERENCE score against the target profile.

DIMENSIONS
- Shape: 0 = very round/smooth, 10 = very angular/spiked
- Size: 0 = large/heavy, 10 = small/light
- Speed: 0 = slow/poised, 10 = fast/staccato
- Texture: 0 = warm/organic, 10 = cold/techno

TARGET PROFILE
- Shape: [X]
- Size: [X]
- Speed: [X]
- Texture: [X]

CALCULATION
Coherence score = 100 - sum of absolute differences * 2.5
(Max score = 100 if perfectly aligned.)

List: [names]

Output: table Name | Shape | Size | Speed | Texture | Coherence.
Sort by coherence descending.

Step 5 — Final consumer test

AI does not replace testing on real targets, but it prepares and accelerates it. Take your top 10 names, build a Typeform/Tally survey, ask for perception on the four axes, and compare against the AI scores. Adjust the scoring model if needed.

Phonetic copywriting pipeline

Use case: rewriting a landing page

You are a copywriter and an expert in sound symbolism.

CONTEXT
- Product: [description]
- Main promise: [sentence]
- Desired phonetic profile: [Bouba dominant / Kiki dominant / Mixed]
- Audience: [persona]

ORIGINAL COPY
[paste headline and 3 main blocks]

TASK
1. Audit each section on the phonetic profile (0-10 score per axis).
2. Identify the words that contradict the target profile.
3. Propose 3 rewrite variants per section, progressively increasing
   phonetic alignment.
4. For each variant, indicate the expected phonetic score and the
   projected emotional effect on the reader.

Output: table Section | Original | Variant 1 | Variant 2 | Variant 3
| Variant 3 score.

Use case: generating A/B test variants

For the following headline, generate 8 variants to A/B test on Meta Ads.
Each variant must explore a different phonetic combination per this plan:

V1: Maximum Bouba (round, soft, long)
V2: Maximum Kiki (dry, fast, short)
V3: Bouba + number
V4: Kiki + number
V5: Bouba + action verb
V6: Kiki + action verb
V7: Mixed Bouba (75%) + Kiki (25%)
V8: Mixed Kiki (75%) + Bouba (25%)

Source headline: [sentence]

Output: table Variant | Text | Dominant profile | Hypothesis to test.

Common pitfalls when using AI

Building a brand-wide phonetic charter

Once the name is validated, extend the phonetic profile to the entire brand:

Prompt — Brand phonetic coherence audit

Below is a list of verbal assets for the brand [name]:
- Brand name
- Tagline
- 10 feature names
- 5 recent email subject lines
- 3 main CTAs

For each item, assign a phonetic profile on 4 axes.
Compute a global coherence index (0 to 100).
Identify the 3 most incoherent items and propose rewrites aligned with
the brand's dominant profile.

Declared dominant profile: [Bouba / Kiki / Mixed]

KPIs of an AI-driven naming program

To measure the ROI of an AI-sound-symbolism pipeline:

• Useful generation rate: % of names shortlisted / total generated.
• Average coherence score on shortlisted names.
• Consumer test pass rate: % of names scoring 70/100 on target.
• Time-to-name: duration from brief to final validated name.
• Cost-per-validated-name: AI cost + copywriter time / number validated.

Benchmark observed across 12 bilingual naming missions: time-to-name down from 6 weeks to 5 days, cost-per-validated-name divided by 4.2.

Takeaway

Generative AI does not replace phonosymbolic expertise — it gives it industrial speed. A five-step pipeline (brief, generation, filtering, scoring, testing) lets you produce dozens of validated names per cycle. The same principle applies to copywriting and A/B variants. The key remains a structured brief on the four phonosymbolic axes and a scoring cross-checked across several models. In the next chapter, we move to the entrepreneurial plane: phonetic architecture of an entire brand, pricing, packaging and product voice.