Tracking Pixel
The Classify Pixel is a lightweight JavaScript tag that fires on every page load where it's embedded. It captures the real URL, domain, and delivery context — giving you ground-truth data independent of what any ad platform reports.
The pixel serves two distinct audiences:
- Advertisers — track where your ads actually served and analyze campaign delivery
- Publishers — detect bot, agent, and non-human traffic across your properties
For advertisers
What it does
The pixel embeds alongside your ad creatives as an impression tracker. When the creative loads, the script captures the actual URL where the ad appeared and records it against your campaign. This gives you:
- Delivery transparency — see exactly where your ads ran, not just what your DSP claims
- Mis-represented inventory detection — catch inventory that was labeled or sold as something it isn't
- Post-hoc contextual reporting — analyze what contextual topics your campaign actually reached, even if you didn't use Classify segments to target
- Campaign optimization — identify strong and poor-performing URLs, publishers, and domains
Use cases
| Use case | Description |
|---|---|
| Campaign optimization | Monitor impression delivery of a curation deal or segment. Identify what's working and what isn't. |
| Campaign reporting | Track delivery of any campaign — even ones where you don't control the targeting — for post-delivery performance analysis. |
| Contextual analysis | Classify can analyze the URLs where your ads served and produce contextual reports covering IAB categories, keywords, entities, sentiment, product categories. |
For publishers
What it does
Embedded on your pages, the pixel captures every visit — human and non-human. Classify analyzes this traffic to identify bots, scrapers, search crawlers, and AI agents, and builds a detailed picture of who is accessing your content and how.
This data powers AgentGraph — Classify's intelligence layer for understanding non-human traffic.
AgentGraph
AgentGraph identifies and categorizes every non-human visitor to your site. It answers:
- What percentage of your traffic is non-human? Broken down by category: search bots, LLM agents, scrapers, and other automated traffic.
- Which agents are visiting? Individual agent identification (e.g., Googlebot, GPTBot, ClaudeBot, Bytespider) with traffic volume and trends.
- How are they accessing your content? Crawling patterns, retrieval frequency, and which pages attract the most non-human attention.
- What's changing? Weekly trends showing whether bot and agent activity is increasing, decreasing, or shifting between categories.
AgentGraph categorizes non-human traffic into four types:
| Type | API value | Description | Examples |
|---|---|---|---|
| Search bots | search | Search engine crawlers indexing your content | Googlebot, Bingbot |
| LLM training | llm_training | AI model training crawlers | GPTBot, CCBot |
| LLM retrieval | llm_retrieval | AI agents fetching content for real-time answers | ClaudeBot, PerplexityBot |
| Scrapers | scraper | General-purpose web scrapers | Bytespider, custom scrapers |
Use cases
| Use case | Description |
|---|---|
| Bot detection | Understand what percentage of your traffic is non-human and which agents are responsible. |
| Agent monetization | Identify AI retrieval traffic to evaluate monetization strategies for agent-accessed content. |
| Content protection | Detect high-velocity scraping and take action to protect your content. |
| Traffic quality | Separate human engagement metrics from bot inflation for accurate analytics. |
How it works
Regardless of use case, the setup flow is the same:
- Register — create a pixel campaign via the API, specifying your use case
- Generate pixel code — produce a
<script>tag scoped to your campaign ID - Embed — attach the script to your ad creatives (advertisers) or embed it on your pages (publishers)
- Pull reports — retrieve campaign delivery data or publisher traffic analysis via the API
What the pixel captures
| Data point | Description |
|---|---|
| Domain | The domain where the pixel fired |
| Full URL | The complete URL of the page |
| Impression count | Number of pixel loads (impression volume) |
| Timestamps | When each impression occurred |
| DSP macros | Deal ID, publisher, inventory domain, inventory source (when configured) |
| Device category | Derived device type (desktop, mobile, tablet, etc.) |
| Geography | Derived country/region |
Classify does not expose raw IP addresses or user agent strings. Device and geography data are derived internally and returned only as aggregated categories.
See the Tracking Pixel API for full setup instructions.