Built for the AI era of developer tools

AI is rewriting how developers find, read, and build. Carbon’s model—human attention, counted honestly, on hand-picked sites—was already the answer. The shift makes it stronger, not weaker.

What changed

Developers moved. The web’s numbers stopped meaning what they did.

Two things are happening at once. Developers are doing more of their work inside AI assistants and agents, and away from the open pages that used to carry an ad. At the same time, automated agents now make up a large and fast-growing share of raw web traffic.

Cloudflare found that only about 2% of AI crawling corresponds to a real user action; the rest is automated training and search extraction. One AI crawler fetched roughly 38,000 pages for every visitor it referred.1 And these fetchers read raw HTML: OpenAI’s GPTBot generated 569M+ requests in a month and executed zero JavaScript.2

An impression counted from an agent that never saw a page is a number with nothing behind it. Across the web, those numbers are climbing, quietly inflating what advertisers pay for and what publishers report.

  • 1Cloudflare, “The crawl-to-click gap,” Aug 2025. AI crawling by purpose: ~80% training, ~18% search, ~2% user actions.
  • 2Vercel / MERJ, “The rise of the AI crawler,” Dec 2024. GPTBot, ClaudeBot, PerplexityBot, Meta and Bytespider fetch but don’t execute JavaScript (Google’s Gemini and AppleBot are exceptions).

By construction

A Carbon impression only counts when the ad’s script runs and the unit renders.

Carbon’s web ad unit is script-based. It registers a served impression only when a real browser runs the JavaScript and the unit renders, the “served and renders” rule Carbon has always billed on.

Raw-HTML fetchers do neither. They pull the markup and stop. No script runs, nothing is ever rendered, so nothing is there to count. The raw-HTML agent traffic that inflates impression counts elsewhere on the web can’t inflate Carbon’s. The number stays clean by construction, not by policing.

AI crawler · raw HTML

01Fetches the page markup
02Does not execute the ad script
03Nothing renders

Nothing to trigger. Not counted.

Human · in a browser

01Loads the page and runs JS
02Runs the ad script
03The unit renders, impression served

A real browser that ran the script.


The bar for one impression

One impression. One real render.

That’s the only way a Carbon impression fires. The ad’s script runs and the unit renders. A raw-HTML fetch never runs that script, so it can’t manufacture a Carbon impression.


The developers still reading are the ones worth reaching.

Human page traffic is under real pressure. That’s the honest read of the shift. But the pressure is selective. As routine lookups move into AI tools, the people who still open the docs, the changelog, the deep-dive are a narrower and higher-intent segment: developers evaluating, deciding, and choosing what to build with.

Carbon reaches them directly, across 350+ hand-picked tech sites, in a single native unit they do not resent. The pitch was never volume. It is the right attention, from the right people, on pages they chose to be on.


Both sides of the shift

As developers move into AI tools, Carbon moves with them.

The same SDK and API that place native ads on the web let AI-first developer tools become Carbon inventory. In these contexts the ad counts on a successful request—no viewport required—so a terminal, an agent, or an IDE assistant can carry a respectful text ad the same way a website carries a native one.

$ npm install @carbonads/sdk
GET https://srv.buysellads.com/ads/{zonekey}.json

This is not hypothetical. Freebuff, a free AI-coding platform, is funded by Carbon text ads served through the API. And it sits in one of the fastest-growing corners of developer tooling. Carbon captures both sides of the transition: the humans still on the web, and the tools they are moving into.

See the placement in Freebuff’s CLI and web app →

The record

Not a pivot. The same discipline, running since 2010.

Since 2010Advertising for developers
$120M+Paid to publishers
350+Active hand-picked sites
600+Active advertisers

Case study · CircleCI

65% lower cost-per-lead

CircleCI ran native performance ads on developer sites, decreasing cost-per-lead while raising lead quality.

Case study · Rollbar

2.4× more clicks

Rollbar reached thousands of developers on high-quality placements, more than doubling its click volume.

Two ways in

One network, built for what comes next.

For advertisers

Reach the developers who still decide.

Put your product in front of a high-intent technical audience, counted honestly, on sites they trust. No raw-HTML bot inflation, no $40 CPMs, no discovery-call gauntlet.

Plan a Campaign

For publishers & tool builders

Turn your tool into inventory.

A website, a CLI, an agent, an IDE extension: if developers use it, it can carry a respectful Carbon ad. One SDK, one API, and a network that has paid out $120M+ since 2010.

Monetize your tool