TL;DR
- Each AI engine runs on its own map of the web: ChatGPT leans on Bing and OAI-SearchBot, AI Overviews pulls Google’s index, Perplexity runs a live search per prompt.
- Only 11% of domains get cited by both ChatGPT and Perplexity for the same query, and 71% of cited sources appear on only one platform.
- AI Overviews and AI Mode both run on Gemini 3 but cite different sources — URL overlap is about 10.7%, citation overlap 13.7%.
- Only about 12% of URLs cited by ChatGPT, Perplexity, or Copilot also rank in Google’s top 10 — ranking strong on Google does not guarantee AI visibility.
- Pick one customer question, ask it in all four engines (ChatGPT, AI Overviews, AI Mode, Perplexity), and log which engines cite you — that maps which doors are open.
Two small business owners ask the same question. One types it into ChatGPT. The other types it into Perplexity. The answers come back completely different.
Same question. Same week. Not even close to the same list of cited sources.
That is not a glitch. It is how the tools are built.
Every AI engine keeps its own address book of the web. ChatGPT has one. Google AI Overviews has another. Perplexity has a third. Google’s own AI Mode, on a separate tab from AI Overviews, keeps a fourth. When someone asks a question, the engine opens its own address book first.
The books overlap a little. Not much.
So if you are trying to figure out how to be seen by AI, the first honest answer is simple. There is no single AI to be seen by. There are at least four, each with its own map of the same web.
This article walks through how each of them actually finds and picks pages. Not the mystical version. The mechanical one.
Why do AI engines give different answers to the same question?
Each engine runs on a different map of the web. ChatGPT calls on Bing plus its own crawler plus a few news partners. Google AI Overviews pulls from Google’s own index with query fan-out. Perplexity performs a fresh live search for every prompt. Only 11% of domains get cited by both ChatGPT and Perplexity for the same query.
That 11% number is striking. Two tools, same question, and almost nine out of ten cited domains are on one map but not the other.
It gets sharper. A study of AI citation behavior found that 71% of cited sources appear on only one platform. Only 7% are cited across all the major tools at once.
You are not imagining the inconsistency. It is built in.
Why? Because the engines were never trying to agree. Each one was designed by a different team, with a different product idea, using a different method to fetch the web. The answers diverge because the maps diverge.
How does ChatGPT find my page?
ChatGPT Search uses Bing’s index, its own crawler called OAI-SearchBot, and partner feeds from news and data providers. A page already indexed in Bing has an advantage. Only about 12% of pages cited by ChatGPT, Perplexity, or Copilot rank in Google’s top 10 for the same query. Being strong on Google is not the same as being visible in ChatGPT.
Read that again, slowly. You can be in Google’s top three for your main search phrase and still be invisible inside a ChatGPT answer about the same topic.
The reason is the Bing partnership. OpenAI has stated that ChatGPT Search uses third-party search engines, and for its enterprise customers it specifically names Bing. So the first question for a site owner is not “am I ranking on Google?” but “am I in Bing?”
Most site owners have never checked.
Bing’s webmaster tools are free. Submitting a sitemap there takes a few minutes. That alone can open a door that was quietly closed.
How does Google AI Overviews find my page?
AI Overviews does not have its own crawler. It draws from Google’s existing index — the same one regular search uses. Since January 2026 it runs on Gemini 3, which breaks each query into 8 to 12 smaller questions. It searches each one separately and assembles the answer from multiple sources.
This is the one piece of good news for site owners who have invested in traditional SEO. If Google already knows about you, AI Overviews already knows about you too.
The complication is what AI Overviews does with that knowledge.
Gemini 3 does not look up your main query and pick one winner. It fans the query out. Take “how do I fix a slow WordPress site” — that one question might become eight different searches, covering plugins, hosting, images, caching, and more. Each sub-search picks its own source. The final answer is stitched together from all of them.
So your page can be in Google’s index and still not get cited. The sub-query that would have found you may never have been run.
How is AI Mode different from AI Overviews?
AI Mode is a separate product on a separate tab, not the summary box inside regular results. Both now run on Gemini 3, but they cite different sources for the same query. SE Ranking found about 10.7% URL overlap between them. Ahrefs found 13.7% citation overlap. Same company, same AI model, different answers.
This surprises almost everyone who hears it.
Most site owners do not know AI Mode exists. They see the summary box on google.com — AI Overviews — and assume that is where Google’s AI work stops. It is not.
AI Mode is its own tab at the top of Google Search. The user has to click into it. Once they do, the whole page is a conversation with the AI rather than a list of results.
The overlap between AI Overviews and AI Mode citations is under 14%. A business can be cited in one and ignored by the other for the exact same query. Same AI model. Same month. Different tab.
If you check only AI Overviews when you audit yourself, you are missing most of Google’s AI surface.
How does Perplexity find my page?
Perplexity runs a fresh web search for every single prompt. Its own crawler, PerplexityBot, builds an index. Third-party search APIs from Bing and Google fill the gaps. Unlike ChatGPT, Perplexity cannot cite a page that has not already been crawled. The freshest-updated pages win more citations than older ones.
Perplexity is the most literal of the big engines. Every prompt triggers a new search. Nothing is pulled from a cached conversation or a memorized answer.
That has two consequences for a site owner.
First, discoverability matters more than it does anywhere else. If PerplexityBot has never reached your site, you are not in the running — no matter how good the content is.
Second, freshness matters more. Perplexity biases toward pages with recent updates. A thoughtful article from 2023 often loses to a competitor’s rough draft from last month.
Reddit is Perplexity’s favorite source — about 47% of its top citations for some query categories come from Reddit threads. That says something about how the engine thinks. It trusts conversation. It trusts recency. It trusts pages that look like the community talked through the question, not pages that look like marketing.
Do the engines agree on who to cite?
Not usually. Only 11% of domains are cited by both ChatGPT and Perplexity for the same query. 71% of cited sources appear on just one platform. ChatGPT leans on Wikipedia. Perplexity leans on Reddit. AI Overviews leans on YouTube. Different maps, different favorites.
And the maps are unstable.
One analyst found that between 40% and 60% of cited sources change month-to-month across Google AI Mode and ChatGPT. The list of who gets cited on Monday is not the list from a month ago.
This is why audit tools that check only one engine on one day give you a distorted picture. You can look terrible in one, fine in another, and invisible in a third — all at the same time.
There is a disconfirming framing worth holding alongside all of this. Google still processes roughly 417 billion searches per month. ChatGPT sees about 72 billion messages. The gap is huge. For most small businesses, Google’s index remains the primary path. AI search is adding noise to the picture, not replacing it.
But the noise is uneven. On some queries, in some categories, for some audiences, the AI answer is the only thing the customer sees.
Other questions worth answering
What happens when my site renders content through JavaScript instead of plain HTML?
AI crawlers from OpenAI, Anthropic, and Google fetch raw HTML only — none of them execute JavaScript. A March 2026 Writesonic study of 60+ page elements across six major AI platforms confirmed this for ChatGPT, Claude, and Gemini. If your site builder loads paragraphs or product details through client-side JavaScript, those sections are invisible to every major engine. Server-side rendering is the only safe path.
Why do free users get fewer brand-website citations than paid users for an identical prompt?
Because of how the paid tier handles prompts. A March 2026 Writesonic study of about 1,161 ChatGPT citations measured the gap. The paid tier cites brand sites about 56% of the time versus 8% for free — roughly a 7x gap. The paid tier breaks one prompt into about 8.5 sub-queries and searches brand domains directly, so testing on free ChatGPT understates real reach.
What technical mistakes silently block crawlers from reaching your site?
Three things show up most often:
- Your firewall may block crawlers by default. Cloudflare identified roughly 226 distinct AI-related crawlers in early 2026, and many bot-management rules drop them automatically.
- Your robots.txt could block GPTBot when you only meant to opt out of training. Blocking OAI-SearchBot by mistake also closes ChatGPT Search.
- Bing has not indexed you, which closes the ChatGPT door entirely.
How long does it take a brand-new article to enter LLM citation lists?
Perplexity is the fastest. Its crawler reliably picks up and cites new content within 1 to 2 weeks of publication, per 2026 crawler analyses. ChatGPT and AI Overviews lag on AI citation because they rely on Bing’s and Google’s indexes.
The available studies don’t pin a precise window for either. Don’t audit a brand-new page the day after publish, since most pages take several weeks to appear.
How does Claude crawl the web compared to OpenAI’s setup?
ClaudeBot is one crawler doing both training and live retrieval, where OpenAI splits the role across three separate bots. ClaudeBot roughly doubled its crawl rate between Q3 2025 and Q1 2026.
Like every other big engine, Claude only fetches raw HTML, so any content rendered through JavaScript on your site stays invisible to it. The mechanic differs. The blind spot does not.
How should you test where each AI engine sees you?
There is no single AI-search strategy. Check which tools your customers actually use. Test the same question in each one. Decide whether to prioritize the index Google already has of you, or the different map each other engine keeps. The strongest pages are visible on two or three at once — not all four.
Here is a small plan that fits a small business.
Pick one question your customers ask most often. Open four tabs — ChatGPT, google.com with AI Overviews on, google.com/aimode, perplexity.ai. Ask the same thing in each. Write down what each one shows.
That exercise takes fifteen minutes. It tells you more about where your business stands with AI than most paid audits would. The four-surface test for AI visibility writeup walks through this same exercise with a logging sheet you can reuse each month.
If you are cited in one, you now know which map you are already on. Most of the work is figuring out why you are on that one and not the others.
If you are cited in none, your first job is not AI optimization. Your first job is to check whether each engine’s crawler can even reach your site. The AI crawler block diagnostic article walks through that check.
Every engine keeps its own address book. You cannot force your way into all four at once. But you can start by knowing which book you are in — and which one you are missing from.
If you want a calm second opinion on which engines your customers actually use and which ones are citing you today, you can contact me. No pitch. Just a clearer view of where you sit on four different maps of the same web.