Practical Guide to E-commerce Ad Creatives: Real-Time A/B Testing with API Data

2025-05-23

Core Objectives

Automatically generate multiple versions of short video ad copy (titles, hooks, voiceover scripts, etc.)
Quickly push different versions to TikTok/Douyin and monitor their performance
Achieve full automation of the A/B testing process from creation to iteration
Dynamically adjust creative content based on hot comment keywords and product selling points

1. Input Sources: Comment Keywords + Product Detail API

Step 1: Extract Hot Keywords from TikTok Video Comments

Retrieve user comment data for a given TikTok video and perform keyword extraction to identify frequently mentioned topics, which can guide creative copywriting.

import requests
from collections import Counter
def get_tiktok_comments(video_id):
url = "https://luckdata.io/api/tiktok-api/comment_list_by_video"
params = {"video_id": video_id}
res = requests.get(url, params=params)
return res.json()
def extract_keywords(comments):
keywords = []
for c in comments['data']:
text = c.get("text", "")
for word in text.split():  # Replace with jieba or Spacy for better tokenization
if len(word) > 1:
keywords.append(word.lower())
return Counter(keywords).most_common(10)
comment_data = get_tiktok_comments("7349338458284xxxxxx")
hot_keywords = extract_keywords(comment_data)
print(hot_keywords)

Step 2: Fetch Product Details (Lazada Example)

Use API to retrieve product title and key selling points from e-commerce platforms, which will be used for generating ad content.

def get_lazada_product_detail():
url = "https://luckdata.io/api/lazada-online-api/x3fmgkg9arn3"
params = {"site": "vn", "itemId": "2396338609"}
res = requests.get(url, params=params)
return res.json()
product_detail = get_lazada_product_detail()
print(product_detail["data"]["title"])

2. Generating Creative Versions (Prompt + LLM)

Prompt Structure

Combine extracted comment keywords and product titles to design prompts that guide LLMs in generating engaging ad titles and hook lines.

import openai
def generate_hooks(keywords, product_title):
prompt = f"""
You are a short video ad copy expert. Based on the following inputs, generate creative content:
Product Title: {product_title}
Hot Keywords: {', '.join([kw for kw, _ in keywords])}
Please output:
1. Three engaging ad titles suitable for TikTok/Douyin
2. Three hook lines that can be delivered within the first 5 seconds of a video
"""
response = openai.ChatCompletion.create(
model="gpt-4",
messages=[{"role": "user", "content": prompt}]
)
return response['choices'][0]['message']['content']
hooks = generate_hooks(hot_keywords, product_detail["data"]["title"])
print(hooks)

3. A/B Testing Execution: Upload + Performance Monitoring

✅ Auto Publishing Versions with Metadata Tracking

Use third-party tools or TikTok's business API to publish multiple video versions, combining different titles and hooks, such as:

Title A + Hook A
Title A + Hook B
Title B + Hook A
Title B + Hook B

Record metadata like version ID, upload time, and associated creative assets for each variant.

✅ Monitoring Performance via TikTok Video Stats API

Retrieve performance metrics for each video version including plays, likes, comments, and shares.

def get_tiktok_video_stats(video_id):
url = "https://luckdata.io/api/tiktok-api/tiktok_video_info"
params = {"video_id": video_id}
res = requests.get(url, params=params)
return res.json()
video_stats = get_tiktok_video_stats("7349338458284xxxxxx")
print(video_stats)

Key data fields include:

play_count
like_count
share_count
comment_count

4. A/B Test Metrics and Optimization Strategy

KPI Logic and Performance Scoring

Evaluate each version using a combination of metrics such as play count, like rate, completion rate, and order conversions.

Version ID	Play Count	Like Rate	Completion Rate	Orders (E-com)	ROI Estimate
A1	10,000	3.2%	70%	87	High
A2	9,000	2.1%	64%	65	Medium

Composite indicators:

CTR = Like Count / Play Count
Completion Rate = Completed Views / Total Views (if available)
Conversion Rate = Orders / Views (based on tracked links)

5. Automated Iteration Based on Performance

Build an automated loop for creative optimization:

Automatically deactivate underperforming versions (e.g., CTR and conversions < 30% of average)
Extract new keywords from recent comments to generate fresh creatives
Reupload and re-enter the A/B testing cycle

This creates a full feedback loop: Auto-generation → Publishing → Data Feedback → Creative Optimization

✅ Summary: Components Required for Full A/B Testing Loop

Component	Implementation Method
Comment Keyword Extraction	TikTok/Douyin API + NLP Tools
Product Info Collection	E-commerce API (e.g., Lazada)
Creative Generation	LLM (e.g., ChatGPT) with Prompt Design
Data Collection & Analysis	TikTok API for performance metrics
Feedback & Decision Logic	Automated rules based on KPI thresholds