Every major AI engine uses a search index. This is a database of web pages an engine checks before generating an answer. Google's AI uses Google, Copilot uses Bing, and Perplexity built its own index.
ChatGPT uses OpenAI's index and Claude reportedly used Brave but appears to have switched to its own. And neither Grok nor Meta AI disclose what index they use.
Ranking highly in Google is one of the most impactful things you can do to improve your AI search visibility, especially in Google AI Overviews, AI Mode, and Gemini. Still, rankings are only one part of your AEO strategy. We all know off-site brand authority matters now, whether that's getting mentions on sites like Reddit, LinkedIn, or YouTube, or even third-party media sites.
But because each AI engine uses its own index, you can rank #1 in Google and still be invisible on an AI platform that never looks at Google-ranked results.
What's a brand to do?
I went engine by engine to find out which index each AI platform uses and when it searches that index. Then, I looked at how you can measure your performance in each. Here's what I found, including tips for monitoring your visibility in every AI engine.
First, Let's Clarify: A Search Engine And A Search Index Are Not The Same Thing
People use the two terms interchangeably, but they are not the same. Google describes Search in three stages:
- Crawlers discover pages
- Indexing stores them in a database
- Serving pulls the most relevant results from that database when someone searches
The index is the database. The search engine is the system built on top of it. That includes the crawler, the index, and the results page.
Think of a search engine like a library. The index is the catalog of books and shelves they're stored on. Every book is logged so it can be easily found. The search engine is the librarian that takes your question, checks the catalog, and hands you the books that best match what you're looking for.
An AI engine is different from a search engine.
A search engine hands you a list of results, but an AI engine reads sources and writes the answer for you, like a researcher writing a summary. Sometimes it asks the librarian (searches an index) before it provides an answer, and sometimes it answers directly from its existing knowledge (e.g., what it already learned in training).
Checking an index before answering is called grounding, or retrieval-augmented generation (RAG). This step affects how current AI answers are, and the citation sources the AI uses to generate its answer.
Training memory has a hard limit, though. Every model stops learning at a cutoff date, so it knows nothing about what happened after it, like a price change or a product you launched last month. A researcher who answers from memory can only tell you what they knew when they last studied. But if they check the library first, they can give you today's answer and show you where they found it.
That's why grounded answers are more current and include citations that you can check.
Not every researcher uses the same library (index), either. Google, Microsoft, and Perplexity each use their own. OpenAI and likely Anthropic have built new ones. And the rest haven't disclosed what search index they use.
Here's Which Index Each AI Engine Uses
Here's the short version, as of October 2026.
| AI engine | Index it uses | When it searches |
|---|---|---|
| Google AI Overviews | When Google decides an AI answer adds value | |
| Google AI Mode | Google, plus Knowledge Graph and Shopping Graph | Every search, split into subtopics that it searches simultaneously |
| Gemini | When the model decides search will help | |
| ChatGPT | Its own index, plus live Google results from third-party scraping services | When the model decides, or when the user turns on Search from the tools menu |
| Microsoft Copilot | Bing | When the model decides, Copilot runs short keyword searches (query fan-outs) instead of your full prompt |
| Perplexity | Perplexity's own index | Nearly every question |
| Claude | Not disclosed (suspected Brave, likely its own since October) | When the model decides |
| Grok | X posts, plus an undisclosed web provider | When the model decides |
| Meta AI (Muse Spark) | Not disclosed, possibly an in-house index | When the model decides |
The Google family
AI Overviews, AI Mode, and Gemini use Google's index. To show up in AI Overviews or AI Mode, a page needs to be indexed and eligible to show a snippet (the short description under a blue link). Pages qualify by default unless a noindex or nosnippet rule blocks them. There's no special markup or page-one ranking required to be eligible.
Being eligible isn't the same as being cited, though. Google says these features use its core ranking systems to pick which pages to retrieve. So the higher a page ranks for the searches Google runs, the better its odds of making it into the answer. That's why SEO is still the foundation for showing up in Google's AI features.
Google says both features may use "query fan-out," running several related searches across subtopics before generating an answer.
AI Mode takes this a step further. On top of the core index, Google's AI Mode adds:
- Knowledge Graph (Google's database of billions of facts about people, places, and things)
- Real-time data
- Shopping data
Gemini uses the same index, but its API documentation says the model decides whether a search would help before it runs one.
The Bing family
Copilot runs on Bing, Microsoft's own search engine. Microsoft's Copilot documentation says it sends a short search query to Bing, but it doesn't send your entire prompt. Its example, "New York City weather," is short enough to send as-is, while it cuts longer prompts down to a few keywords.
Take, for instance, a prompt like:
"What are the best CRMs for a 20-person sales team? Please provide current pricing tiers for each option you recommend."
Copilot will break this down into smaller web searches to check for current pricing:
In June 2026, Microsoft also launched Web IQ, a set of grounding APIs for AI systems built on the Bing index. Developers use it to connect their AI agents to current web pages, news, images, and videos. That means Bing's index can now feed AI apps and agents beyond just Copilot.
The AI engines building their own index
ChatGPT
In 2023, Bing was ChatGPT's default search engine. Today, Bing is only one of several sources ChatGPT uses.
OpenAI's help center says it "sometimes partners with other search providers," but it doesn't list them. However, it does link to Microsoft's and Shopify's privacy policies when it explains how providers handle your queries, so they're likely involved.
Fortunately, we can also look at independent testing for more insight. RESONEO, a French search marketing consultancy, captured ChatGPT's retrieval data across 1,200 answers and found OpenAI's own index at work.
Only 1.5% of its URLs appeared in Bing's top 20 for the same queries. In paid Thinking mode, though, 75.3% of results were live Google results, pulled through third-party scraping services instead of OpenAI's own index. OpenAI hasn't confirmed this.
Perplexity
Perplexity has its own index, too. Its Search API covers hundreds of billions of pages. Its index receives tens of thousands of updates every second as it adds new pages and refreshes old ones.
Perplexity does this by breaking each indexed page into smaller passages. In Perplexity's words, its system "divides documents up into fine-grained units" that are "individually surfaced and scored" against the search. Put simply, this means Perplexity can cite just one relevant passage instead of an entire page.
Its crawler, PerplexityBot, collects the pages that go into the index. Perplexity doesn't explain how the bot discovers pages, but it says the bot follows robots.txt, so blocking it will keep your pages out of Perplexity search.
The ones that don't say ...
Claude, Grok, and Meta AI don't say which search index they use.
Claude
Anthropic's web search documentation says Claude decides when to search. In 2025, Brave Search appeared on Anthropic's subprocessor list, suggesting Claude used to use Brave's index. But that seems to have changed.
Writesonic found that Claude's default search moved off Brave around October 5th, 2026. It most likely runs on an index Anthropic built itself.
Grok
Grok has two separate search tools:
- web_search for the open web
- x_search for posts on X
That gives it direct search across X, including keyword, semantic, and user search. Other engines generally see X posts only when public posts show up in their own web index.
Meta AI
Meta's Muse Spark announcement says the tool leans on Instagram, Facebook, and Threads content. Meta also runs a crawler built to improve Meta AI search. Unconfirmed reports suggest Meta may also be building its own index.
Why AI Engines Are Building Their Own Indexes
Why would these AI platforms build their own indexes when Google and Bing already exist? It comes down to costs, competition, and control.
Renting search results has gotten harder
Microsoft retired Bing Search APIs in August 2025. Its suggested replacement, Grounding with Bing Search, costs 40% to 483% more, depending on the pricing tier.
Google is moving customers off its Custom Search JSON API by January 1, 2027, and doesn't offer open access to its full index. It also sued SerpApi, a service that scrapes and resells Google results, for getting around its anti-bot protections.
In July 2026, a federal judge dismissed most of the case, ruling that plain search results are public facts, and not copyrighted work. Google was allowed to amend only its claims about knowledge panels and other third-party content.
Google and Microsoft are competitors
At Google's antitrust trial, ChatGPT head Nick Turley said OpenAI had problems with its search provider and asked Google for access.
An OpenAI email shown in court said:
"We believe having multiple partners, and in particular Google's API, would enable us to provide a better product to users."
Google said no, citing "too many competitors." Meta reportedly started building its own index to rely less on Google and Bing.
Freshness and control matter
Perplexity says "information staleness is one of the biggest failure modes for AI agents."
Perplexity doesn't say how often it recrawls a given page, but running its own index means it sets that schedule itself. It also decides how to break up pages for citations.
The #1 brand in ChatGPT changes 52% of the time.
Get the report and find out why.
"When" Matters As Much As "Which"
All of these AI engines decide what to search for, but they also decide whether to search at all. Instead of learning from past searches, the engine decides whether to search the web every time a user asks a query, so the same prompt doesn't always trigger a search.
For example, how often ChatGPT searches the web depends on the model and your plan.
Ahrefs tracked 66.6 million free, logged-out ChatGPT responses and found only 14.4% triggered a web search.
Writesonic's tests of paid ChatGPT found the opposite. GPT-5.5 searched in 94% of conversations, and GPT-5.6 Sol searched in all of them. When ChatGPT skips search, it answers only from training data, with no citations.
AI Mode and Perplexity are the exceptions, since they conduct searches for nearly every question.
This means that in many ChatGPT answers, rankings might not matter. When an engine skips the search, no search index is involved at all. The model answers from what it learned about your brand in training.
A user's chat mode will also change which index an AI platform uses.
In the RESONEO report, free Think mode pulled mostly from OpenAI's index, while paid Thinking mode pulled mostly from Google. A model update could change that again.
Our Claude data shows the same pattern. Claude's standard and extended search modes shared just 10% of cited domains across 50 prompts.
Writesonic also found that GPT-6 Astra runs about 40% fewer searches per answer than the model before it. Most of its searches go straight to brand-owned pages (e.g., pricing page, feature pages, etc.).
You don't need a top-10 ranking to get cited in AI Overviews.
Ahrefs found only 37.9% of AI Overview citations rank in Google's top 10 for the query, and 31% are outside the top 100. So the keyword you track isn't always the one that gets you cited. The follow-up searches matter too, and Copilot will show you exactly what those are.
Keyword Rank Trackers Provide Limited Visibility
It's tempting to rely on keyword rank trackers to see how your brand is doing in AI search, but these tools aren't perfect. In fact, classic rank trackers primarily track only Google and traditional search engine rankings:
- Ahrefs' Rank Tracker only monitors Google keyword rankings.
- Semrush's Position Tracking has Google, Bing, and Baidu rank tracking.
- SE Ranking tracks Google, Bing, Yahoo, and YouTube.
None of these tools track where you rank in the indexes OpenAI and Perplexity built, or in whatever Claude, Grok, and Meta AI use. I couldn't find a tool that does, which makes sense because you can't search or access most of these indexes directly.
Perplexity is the exception. Perplexity's Search API returns ranked results from its own index, so a technical team could check where its pages land for a given search.
Anthropic's and Meta's developer APIs show you the raw search results behind an answer, but that isn't the same as searching their index. The model decides what to search, the results don't come with a position, and neither company says whether its API uses the same index as the app/user interface. Tools sold as "Perplexity rank trackers" track your position and visibility in generated AI answers, not your position in the index.
That leaves Google and Bing as the only indexes where you can see your rank.
And yes, those rankings still matter. Google's index feeds AI Overviews, AI Mode, and Gemini, and Bing's feeds Copilot. But a high ranking doesn't guarantee a citation, even in those four engines.
Search Atlas data reported by Search Engine Journal found ChatGPT citations overlap with Google's results on fewer than 10% of URLs. Assuming that you're fine because you're ranking in Google isn't enough. The only reliable measure is whether you show up in the answers themselves.
Writesonic's turnover report found that AI engines don't agree with each other, either. Across 161,286 prompts, no two engines shared more than 24% of their cited sources. Only 3.8% of sources were cited by ChatGPT, Gemini, Perplexity, and AI Overviews alike.
How To Monitor Your Visibility In Every AI Engine
As you can see, monitoring your visibility across AI engines is easier said than done. You can't see your rank in most of these indexes, so monitoring AI visibility comes down to five things.
The first four use tools you likely already have. The last one is the only way to see all nine engines.
1. Make sure every crawler can access your website
You can't be retrieved from an index if you're not in it. Check your robots.txt and your CDN or firewall settings for the bots that feed AI search. Make sure these bots have access to your site:
- Googlebot
- Bingbot
- OAI-SearchBot
- PerplexityBot
- Claude-SearchBot
- Meta's meta-webindexer
Some engines also fetch pages live, at the moment someone asks a question. Those visits come from separate user-triggered bots, like ChatGPT-User and Claude-User.
Anthropic says blocking Claude-User stops Claude from retrieving your content in response to a user's question. OpenAI says robots.txt rules may not apply to ChatGPT-User, so check that your CDN and firewall aren't blocking it either.
Adding an llms.txt file won't hurt, but don't expect it to help. Google's AI optimization guide says Search ignores llms.txt files, and that includes AI Overviews and AI Mode. A November 2025 Search Engine Journal report analyzed about 300,000 domains and compared each site's AI citation frequency with whether it had the file. It found no link.
I also couldn't find any other AI engine that says it uses llms.txt for search.
2. Use the free reports from Google and Bing
Search Console now has a generative AI performance report for AI Overviews and AI Mode. It shows impressions only (no clicks or queries) and combines both AI features.
Bing Webmaster Tools added an AI Performance report that counts how often Copilot cites your pages.
However, neither covers ChatGPT, Perplexity, Claude, Grok, or Meta AI, so you won't get a full picture of your search engine visibility with these free reports.
3. Read your server logs
Your server logs will tell you:
- Which AI bots visit
- Which pages they fetch
- Which ones fail
A crawler hitting a 403 on your pricing page is a problem that no rank tracker will pick up, but server logs can. Still, reading server logs can get a little technical. We cover a setup in our guide to tracking AI search traffic.
Writesonic's AI Agent Analytics does this without adding any code to your website. It reads your server logs through integrations like Cloudflare, Vercel, and WordPress, and shows which AI bots visit which pages.
It also flags pages that serve errors to bots and tracks clicks from AI answers back to your site, shown by platform.
4. Separate AI referrals in analytics
GA4 added a default AI Assistant channel in May 2026 to measure visits from AI engines like ChatGPT, Gemini, and Claude. Keep in mind, though, that traffic without a referrer still displays as Direct traffic, so this isn't a perfect solution.
5. Track AI answers, engine by engine
Checking answers once tells you almost nothing, because AI answers change constantly. You need a scalable way to reliably monitor daily performance.
This fifth method is the only way to see how often you're mentioned in each AI engine. With it, you can monitor the prompts your buyers ask in every engine you care about, plus run daily checks. Writesonic's turnover report found that 52% of ChatGPT's #1 brand picks rotate to a different brand when the same prompt is rerun. Daily checks build a history and give you an accurate measure of your visibility over time.
In reality, that means using an AI visibility platform, not just randomly checking a prompt in ChatGPT. An AI visibility platform like Writesonic runs your prompts in each engine every day. It checks through the user interface and not the AI's API to record:
- Mentions
- Citations
- Which pages are cited if you aren't
You can see this in Writesonic with the Platform filter:
Writesonic supports nine AI platforms:
- Google AI Overviews
- Google AI Mode
- Gemini
- ChatGPT
- Perplexity
- Claude
- Copilot
- Grok
- Meta Muse Spark
Find Out Where You're Missing
Nine engines read at least five indexes. On your own, you can only see your rank in Google and Bing indexes.
Those rankings still matter because they feed AI Overviews, AI Mode, Gemini, and Copilot. But even then, ranking doesn't guarantee a mention or a citation. What you can measure in every AI engine is how often you show up in the answers to the prompts your buyers ask.
You don't have to overhaul your AI search strategy overnight. This week, start by looking at crawler access. Make sure you aren't accidentally blocking bots so every engine can reach your website. Then use an AI visibility platform to track your search visibility, share of voice, and citation share in each engine (checked every day).
If you want to see where you stand across all of them, try Writesonic for free.
The #1 brand in ChatGPT changes 52% of the time.
Get the report and find out why.
Director of Growth Marketing
Cody Slingerland is the Director of Growth Marketing at Writesonic. He leads SEO and AI search (AEO) strategy. He has over 12 years of search experience, helping companies like Zapier, Instapage, Nutshell, CloudZero, and others improve their visibility and position in organic search.
