Acclaira

Chapter Two

The five engines, and who reads what

"AI search" is not one thing. Five systems dominate, they read different indexes, and being invisible in one tells you nothing about the others. This chapter is the map.

Chapter 2 of 26 · 3 min read

The analogy

Five different newspapers cover your town. They all print stories about local business, but one only reads the council register, one only reads the chamber of commerce newsletter, one sends its own reporter, and two share a wire service. Getting into the council register does nothing for the paper that reads the newsletter. You need to know which door each one uses.

The map

EngineWhere it gets pagesIts crawlerWhat this means for you
ChatGPT
Search + browsing
Primarily Bing's index, supplemented by OpenAI's own crawling OAI-SearchBot (search)
GPTBot (training)
ChatGPT-User (live fetch)
If you are not in Bing, you are largely invisible here regardless of your Google rankings. Bing Webmaster Tools is not optional.
Google AI Overviews
The summary box
Google's main index and ranking systems Googlebot Standard Google SEO is the entry ticket. Google says explicitly there are no additional requirements. TIER A
Google AI Mode
The conversational tab
Same index, but heavy query fan-out and passage-level retrieval Googlebot Same ticket, different game. Ranking #1 for the head term does not guarantee you serve any of the sub-queries.
Perplexity Its own index, rebuilt in-house on Vespa in 2025, plus licensed third-party crawls PerplexityBot An independent front door. You can be strong here and weak elsewhere, and vice versa.
Claude & Copilot Claude uses a search partner; Copilot uses Bing ClaudeBot, Claude-SearchBot Copilot inherits your Bing position. Claude is worth spot-checking because it skews toward documentation and primary sources.

Sources: Google Search Central AI features documentation; Vercel/MERJ crawler telemetry; Perplexity engineering statements; OpenAI crawler documentation. See Appendix B.

The two-word distinction that matters most

Notice that OpenAI runs several bots, and they do different jobs. This trips up almost everyone, including people who should know better, so it is worth being precise:

  • Training crawlersGPTBot, Google-Extended, Applebot-Extended. These gather text that may be used to train future models. Blocking them is a defensible business decision and it does not remove you from AI search results.
  • Search crawlersOAI-SearchBot, PerplexityBot, Claude-SearchBot. These gather pages to cite in answers. Blocking these removes you from those answers.
  • User-triggered fetchersChatGPT-User. Fires when a person or an agent asks the tool to go and look at a specific page, live.
The most expensive mistake in this chapter

In 2023 and 2024 a great many businesses pasted a "block the AI bots" snippet into robots.txt to protect their content from training. A large share of them blocked the search bots at the same time, and then spent 2026 wondering why they never appear in ChatGPT. Go and read your robots.txt today. It takes ninety seconds and it is the highest-return ninety seconds in this book.

Where they overlap, and where they do not

Two practical points follow from the map.

The first is that Bing is now load-bearing infrastructure in a way it has not been for a decade. Not because anyone searches on Bing, but because ChatGPT and Copilot retrieve through it. Verifying your site in Bing Webmaster Tools, submitting a sitemap, and pushing updates through IndexNow costs an hour and unlocks two of the largest answer engines. TIER B

The second is that results diverge sharply between engines, and you should expect that rather than treat it as an error. The same prompt run on ChatGPT, Gemini and Perplexity will often return three substantially different shortlists. Chapter 17 covers what to do about it; for now, simply stop assuming one check is a measurement.

How to check
  • Visit yourdomain.com/robots.txt and read every Disallow line. Confirm no search-class bot is blocked.
  • Search Bing for site:yourdomain.com. If the count is far below your Google equivalent, that is your ChatGPT problem in one number.
  • Ask your developer, or check your server logs, for hits from OAI-SearchBot and PerplexityBot in the last 30 days. Zero hits is a finding.
If you only do one thing

Verify your site in Bing Webmaster Tools and submit your sitemap. It is free, it takes under an hour, and for most businesses it is the difference between existing and not existing inside ChatGPT.

Read the whole book

Answering Machines is 26 chapters on being the answer when your customer stops searching and starts asking. Free, ungated, and every figure is sourced and confidence-rated.