ChatGPT's Index Doesn't Actually Care If You Have a Deal

2026-08-12 — brand building small business India

I was wrong about this six months ago. So was everyone else, more or less.

The story: Resoneo analyzed ChatGPT's search results and found no difference in how OpenAI's in-house index (called labrador) served licensed publishers and sites with zero deal. Same format. Same snippet length. Same freshness rules. A hundred Italian news outlets with no contract got treated exactly like Reuters or The Guardian.

This should surprise you.

Here's why. Back in June, security researcher Suganthan Mohanadasan read ChatGPT's traffic and called labrador a closed allowlist of established names. Reuters. Guardian. WSJ. Wikipedia. That made sense. OpenAI had signed these deals. Seemed logical they'd get preferential treatment in the index itself.

Then a reader from Italy sent him captures showing that every publisher citation ran through the same pipeline, including tiny sites Mohanadasan had never heard of. He was honest about it. Mohanadasan said he "over-reached". The licensing deals are real. The tier system wasn't.

Resoneo's dataset pushes this further.

They analyzed 1,249 ChatGPT answers from July. What they measured: which infrastructure pipeline fetched each result. In free accounts, labrador handled almost everything. Questions with stable answers. Local businesses. Product searches. Even news results split fairly evenly between labrador and Google scraping. For paid accounts in thinking mode—the version that scrapes more—Google made up 75% of results, labrador about 24%. But both licensed and unlicensed sites came through the same way.

The index itself doesn't discriminate.

Now, before you think the deals don't matter, actually, that's not quite right. They absolutely matter, just differently. Licensed publishers see nearly 7x higher clickthrough rates from ChatGPT according to Tollbit's research. But that's about prominence—how ChatGPT displays the result, not whether it finds it at all. Licensed partners get featured in bigger boxes. Better placement in the response. Links that look more like links. Your small newsletter can be in the index. You're just not getting the 42-point headline treatment.

Resoneo's own documentation of the system shows labrador is mostly "press feeds and open science archives"—public data OpenAI can access without paying anyone. The deals buy you the prominent positioning, real-time news access, and visibility bump. They don't buy you into the index. You're already there.

The implication: if you run a site with useful, stable information, ChatGPT can find and cite you.

The catch, though. The companies that signed deals also shape what ChatGPT thinks is worth citing. OpenAI approaches publishers directly; there's no public application process. Small, non-English language, and independent media have largely been shut out of deals so far. So you can be in the index. But your content might not get the algorithmic weight, the prompt guidance, the fine-tuning that tells ChatGPT "hey, this is a credible answer source." That still comes from the licensed tier.

It's a distinction most consultants miss, including us sometimes.

What matters now: making sure ChatGPT can actually access and understand your content. Allowing OpenAI's crawlers (OAI-SearchBot, ChatGPT-User) through your robots.txt or WAF. Making sure your pages are crawlable, your structured data is clean, your key claims sit in text ChatGPT can extract. The licensing deal would be nice—better visibility, guaranteed citations, traffic multiplier. But you're not locked out of being useful without one.

That wasn't true six months ago, based on what we thought we knew.