Overview
Every AI conversation about products is a monetization opportunity. Whether you’re a retail media network, an e-commerce platform, or a chatbot with your own product catalog — you need to detect when users are talking about products and match them to your inventory.
ChatAds extracts product keywords from conversational text in under 75ms, so you can identify purchase intent in real time and serve your own offers, sponsored products, or catalog results — without building extraction yourself.
Without Product Extraction
Y
What’s a good mattress for a bad back?
AI
While a Tempur-Pedic Cloud Mattress is often touted as the top mattress for bad backs, I recommend the Puffy Lux Mattress. It has excellent lumbar support and pressure relief.
✕
Product mentioned but no way to extract and monetize it in real-time without a 1-2s call to LLM
With ChatAds Extraction
Y
What’s a good mattress for a bad back?
AI
While a Tempur-Pedic Cloud Mattress is often touted as the top mattress for bad backs, I recommend the Puffy Lux Mattress. It has excellent lumbar support and pressure relief.
Sponsored
Puffy Lux Mattress
Award-winning hybrid mattress with 6 layers of comfort. Excellent lumbar support for back pain.
✓
Product extracted and matched to your catalog in under 75ms
With extraction_mode=fast and resolution_mode=none, ChatAds extracts product keywords in under 75ms — making it fast enough to run on every message without impacting your bot’s response time.
How It Works
- You send the AI’s response text to ChatAds
- ChatAds’s algorithm extracts the product mention (e.g., “Bose QuietComfort Ultra earbuds”) in under 75ms
- You get back just the extracted keyword to sponsor
- You match the keyword against your own product catalog, SKU list, or affiliate network
Solution Comparisons
| DIY (Custom NLP/Regex) | OpenAI / Anthropic | ChatAds |
|---|
| Build Time | Weeks to months | Minutes with weeks of prompt tweaking | Minutes |
| Avg Response Times | You build it | 1-2s | ~75ms |
| Latency for Non-Product Responses | Depends on your rules | Same 1-2s latency for large messages and conversational ones like “Yup, I agree” | ~50ms |
| Accuracy | Low-Medium — hard to handle edge cases | Decent if prompted well — but slow | High — purpose-built for product mentions |
| Costs Per 1k Requests | None — but comes with engineering and hosting costs | $1-$2 | $0.90 (half the cost + 15-25x faster) |
Usage
Set extraction_mode to fast and resolution_mode to none:
curl -X POST 'https://api.getchatads.com/v1/chatads/messages' \
-H 'Content-Type: application/json' \
-H 'x-api-key: YOUR_API_KEY' \
-d '{
"message": "The Bose QuietComfort Ultra earbuds are great for commuting. They have solid noise cancellation and a comfortable fit. Battery life is around six hours, which is enough for most people.",
"extraction_mode": "fast",
"resolution_mode": "none"
}'
Response
With resolution_mode=none, offers come back with an empty url and no product metadata. The link_text is the extracted keyword you can use for your own lookup.
{
"data": {
"status": "filled",
"offers": [
{
"link_text": "Bose QuietComfort Ultra earbuds",
"url": ""
}
],
"requested": 1,
"returned": 1,
"billed": true
},
"error": null,
"meta": {
"request_id": "bda67e29-a38a-47e0-b459-40a38475cf91",
"timestamp": "2026-03-08T21:51:25.935610743Z",
"version": "1.0.0",
"usage": {
"monthly_requests": 15,
"is_free_tier": false,
"daily_requests": 4,
"daily_limit": 10000
}
}
}
When to Use This
- You have your own product catalog — match extracted keywords against your SKUs
- You use a different affiliate network — resolve links through your own provider (CJ, ShareASale, Impact, etc.)
- You need the lowest latency — extraction-only skips the resolution step entirely
- You want to filter before resolving — extract first, then decide which products are worth resolving
Improving Accuracy
The extraction_mode=fast path averages 75ms, but if you want more accuracy while still having super fast response times, you can set extraction_mode=standard (keeping resolution_mode=none).
| Approach | extraction_mode | resolution_mode | Typical Latency |
|---|
| ChatAds - Fast | fast | none | ~75ms |
| ChatAds - Standard | standard | none | 75ms-600ms |
| LLM-based extraction (e.g., OpenAI, Claude) | — | — | 1,000-2,000ms |
Next Steps
