ChatGPT's Search Index Serves Small Sites Too: 2026 Data
New Resoneo data from 1,249 ChatGPT answers shows OpenAI's in-house index serves small sites exactly like its licensed partners. Here's what that means for earning AI citations in 2026.
Key Takeaways
- The mix depends on the account. In Resoneo’s free-account data, the in-house index handled most ChatGPT search results, with…
- The study also measured what the model stores and quotes. Resoneo reviewed 534 cited pages and found:
- 1. Lead with a tight H1 and a direct answer. Keep your H1 under about 50 characters and open the body with the answer itself.…
ChatGPT’s Search Index Serves Small Sites Too: 2026 Data
New research from French SEO consultancy Resoneo shows that OpenAI’s in-house search index serves small, unlicensed websites exactly the same way it serves licensed publishing partners — same format, same snippet length, same freshness. The finding means generative engine optimization is not reserved for big brands with OpenAI content deals.
Resoneo analyzed 1,249 ChatGPT answers captured in July 2026 and reported the results in a Search Engine Journal story on August 11, 2026. The consultancy reverse-engineered ChatGPT’s network traffic and identified four pipelines that fetch web results, including one tagged “labrador” — OpenAI’s own in-house index.
When Resoneo compared pages served through that index, a licensing deal changed nothing. Partner and non-partner sites were returned with the same format, length, and freshness. Small Italian sites, local businesses, and product pages appeared through the index just as readily as the big news outlets.
How the Pipelines Actually Split
The mix depends on the account. In Resoneo’s free-account data, the in-house index handled most ChatGPT search results, with news results split fairly evenly between the index and content scraped from Google. In paid “thinking mode,” the balance flipped: across 16,407 recorded results, about 75% came from Google scraping and roughly 24% from OpenAI’s in-house index.
The finding corrects an earlier reading. In June, SEO Suganthan Mohanadasan described the index as an “allowlist” of established publishers — an interpretation he retracted on July 14 after readers sent captures showing small sites flowing through the same pipeline. Licensing deals are real, but the index itself is not a walled garden.
What ChatGPT Actually Sees on Your Page
The study also measured what the model stores and quotes. Resoneo reviewed 534 cited pages and found:
- Of the 463 pages with an H1 heading, 387 had that H1 reflected in the snippet — 83.6%.
- Snippets are cut off just after 200 characters, and they pull from the beginning of the page content — not the meta description, which the Google-scrape pipeline only captures about one time in three.
- The median H1 was 51 characters, leaving roughly 150 characters of body copy to make the cut.
In other words, ChatGPT’s citation surface is your H1 plus the first sentence of your page — not the polished meta description you wrote for Google.
What SEOs Should Do Now
1. Lead with a tight H1 and a direct answer. Keep your H1 under about 50 characters and open the body with the answer itself. If the first 200 characters of your content clearly state the point, you maximize the chance ChatGPT quotes you accurately and cites you. For the full methodology, see our guide to ranking in ChatGPT and Gemini AI overviews.
2. Stop treating meta descriptions as your AI ticket. ChatGPT’s in-house index favors the opening of your page content. Write for the model the way you write for the skimming reader — answer first, fluff later — and track your presence with the same discipline as your LLM visibility KPIs.
The headline takeaway for small publishers: you don’t need an OpenAI deal to get cited. You need a page a machine can read in 200 characters.
