ChatGPT’s New Search Index Changes AI Search Optimisation, But Not How You Think

What is Labrador?

Peec AI’s investigation is grounded in server-side events it observed between May and July 2026, when a field in ChatGPT’s own traffic showed results coming from one of four sources: Labrador (OpenAI’s own index) or three external scraping providers. OpenAI’s job postings back this up, with roles describing “indexing systems, retrieval pipelines” and databases run at “exabyte scale”. Testimony from ChatGPT’s head of product during Google’s antitrust trial confirmed OpenAI began building its own index in 2023, initially hoping to answer 80% of queries from it, before conceding under oath that full independence from outside providers could take five years or more to assess.

Crucially, ChatGPT hasn’t dropped Bing or Google. Peec AI found it still pulls from at least eight external providers, including Yelp and TripAdvisor for local listings and Microsoft’s own “Web IQ” grounding platform. Labrador sits alongside those sources, not instead of them, and it also caches pages rather than fetching them live every time, which is why answers come back so quickly.

Why we’re not rewriting your content strategy around it

Here’s the problem with treating any single leak like this as a new set of rules. Labrador is one index, built by one company, described from data captured over a two-month window. Gemini, Copilot, Claude and Perplexity each run their own retrieval stacks, with their own weighting toward video, PDFs, forums, or licensed data partnerships, and none of that is public in the same detail. Optimise a site around what worked for Labrador in August 2026, and you’ve optimised for a system that (by Peec AI’s own admission) is being rebuilt week to week, while doing nothing for the three or four other AI tools your buyers might be using. We covered the broader case for this distinction in GEO vs SEO: what generative search actually changes, and the same logic applies here.

A second problem is specific to your industries. Before asking how to rank inside a particular AI’s index, it’s worth asking whether your buyer is using that AI at all, and on what device.

Who is searching, and how?

Take a facilities manager at a manufacturing plant, researching a compliance requirement on a locked-down work laptop. If their employer has gone the way many mid-size firms have, ChatGPT might be blocked outright and Microsoft Copilot approved instead. This could be because it sits inside the Microsoft 365 licence the company already pays for and has been through IT’s security review. A specifier or architect pulling technical PDFs for a live project is a different case again: much of that research happens on a personal device, at odd hours, and might run through ChatGPT or Perplexity precisely because there’s no IT policy governing it. A property developer scoping a new site is a third pattern again, more likely to start broad on a general chatbot before verifying anything through planning portals, land registry data or a specialist consultant. We looked at how that split plays out day to day in Claude vs ChatGPT: the AI tool that’s actually changing how construction marketing works.

Three buyers, three different tools, three reasons for the difference. None of that is guesswork. Pew Research’s February 2026 survey of over 5,000 US adults found 49% now use an AI chatbot, up from a third in 2024, and 42% use one specifically to search for information, but usage still splits hard by age (around half of under-50s versus 20% of over-65s). Per YouGov’s polling of UK SME decision-makers, hard by sector too, with manufacturing (19%) and real estate (11%) sitting near the bottom of UK sectors for AI adoption, well behind IT and marketing (both over 50%).

Meanwhile, a separate US survey of full-time employees found 78% already use tools like ChatGPT, Gemini or Copilot at work, frequently without their employer’s policy catching up to say which one is sanctioned. Chasing one LLM’s index without knowing which tool, if any, your buyer is even permitted to use at their desk is optimising for a guess. For manufacturers specifically, we’ve gone deeper on where that leaves lead generation in AI tools for manufacturers to generate better leads.

What still works, whoever’s crawling

Strip away the specific index names and the underlying test hasn’t changed. Google, Bing, Labrador and every other retrieval system are trying to answer the same question: is this the most useful, trustworthy source for what the user asked? That’s the case for E-E-A-T (experience, expertise, authority and trust) as the idea worth your time, rather than reverse-engineering one company’s strategy. We set out the practical version of that argument in how to get your brand cited by AI search, and it’s worth checking whether your own site is actually sending you AI-driven enquiries in the first place: see how do you know if AI search is sending you buyers.

In practice, for a construction, property or manufacturing website, that means concise, genuinely useful answers placed near the top of the page rather than buried in the text. It means accreditations, certifications and case studies that prove the “experience” part of E-E-A-T rather than just claiming it. It means FAQ content that mirrors the exact questions your customers type, whichever box they type them into. None of that depends on knowing whether an answer came from Labrador’s PDF vertical or a Bing crawl, and all of it still works if OpenAI rebuilds Labrador again next month, because it was never built for Labrador in the first place.

The practical takeaway

Treat news like the Labrador research as useful context, and observe which platforms your specific audience uses and trusts, since that’s shifting faster than any one company’s index. Keep investing in the content fundamentals (clear answers, credible proof points, genuine expertise) that give you a chance across Google, Bing, ChatGPT, Copilot and whatever comes after Labrador, rather than betting everything on one of them.

If you’d like a second opinion on how your own site reads to both traditional search and AI tools, our construction SEO, property SEO and manufacturing digital marketing teams are happy to take a look. Get in touch.

FAQs

Is ChatGPT’s Labrador index the same as Google’s search index?

No. Labrador is OpenAI’s own retrieval system, separate from Google’s index, though it’s built the same way Google’s was: a general web index plus specialised verticals for content types like video, PDFs and news. ChatGPT still draws on Google and Bing data too, so Labrador sits alongside those sources rather than replacing them.

Should we change our SEO strategy specifically for ChatGPT’s Labrador index?

Not on its own. Nobody outside OpenAI knows what proportion of ChatGPT’s answers Labrador actually powers, and other AI tools run entirely different systems. A stronger AI search optimisation approach is to strengthen the E-E-A-T signals that help you across every search and AI platform, rather than tailoring content to one company’s current, still-changing index.

Which AI tools do construction, property and manufacturing buyers actually use?

It varies by role and by employer. Pew Research and YouGov data both show usage climbing overall, but manufacturing and real estate are among the slower-adopting UK sectors, and many workplaces restrict staff to whichever tool their IT policy has approved (often Microsoft Copilot, bundled with Microsoft 365) rather than leaving the choice open.

Sources

• Peec AI, “ChatGPT built its own search index” 
• Pew Research Center, “Americans and AI 2026: Chatbots, Smart Devices and Views on Impact”, June 2026
• YouGov, “We polled UK SME leaders about AI adoption. Here’s what they said”, August 2025
• HR Dive, “Nearly half of workers say they’ve used banned AI tools at work, survey finds”, August 2025