What Search Engine Does ChatGPT Use? And Is It Elasticsearch?

ChatGPT does not use one search engine: OpenAI says ChatGPT search "sometimes partners with other search providers," names Microsoft (which runs Bing) and Shopify among them, and runs its own crawler, OAI-SearchBot, to put websites into ChatGPT's results. OpenAI does not name Google and does not publish the database underneath, so nothing it documents supports the claim that Elasticsearch or OpenSearch decides whether ChatGPT sees your brand.

Asking "what search engine does ChatGPT use?" often hides a second question: which database holds your content. We were sent an AI-generated answer claiming Elasticsearch and OpenSearch are the retrieval "memory" of AI search and that getting indexed inside them "drastically increases" visibility. Here is what each AI engine actually documents about its own search, checked on October 16, 2026.

What search engine does ChatGPT use? What OpenAI documents

The most direct source is OpenAI's ChatGPT search help page, which says: "ChatGPT search sometimes partners with other search providers. When it does, ChatGPT search typically rewrites your query into one or more targeted queries that it sends those providers."

For how those providers handle your queries, it links exactly two privacy policies, Microsoft's and Shopify's.

Part of ChatGPT search	What OpenAI says	Source
Third-party providers	Sometimes used; queries are rewritten before they are sent. Microsoft and Shopify are the providers whose privacy policies OpenAI links.	ChatGPT search help page
Its own crawler	OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features."	OpenAI crawler docs
Its own index	For Codex's web search, "Cached mode uses an OpenAI-maintained index instead of fetching arbitrary pages live." OpenAI does not say whether ChatGPT search reads the same index.	OpenAI web search docs
Live page visits	When a user asks a question, ChatGPT "may visit a web page with a ChatGPT-User agent."	OpenAI crawler docs
When it searches	"ChatGPT may search the web automatically when your question would benefit from current information." Users can also start a search manually.	ChatGPT search help page
How results are ranked	"ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed."	ChatGPT search help page

Two consequences follow. When ChatGPT does not search, it answers from what the model learned in training, and no search engine is involved at all. When it does search, your customer's question has been rewritten first, so the words a provider receives are not the words your customer typed.

Does ChatGPT use Google?

OpenAI does not say so. In July 2025, Practical Ecommerce reported two informal tests suggesting Google results were reaching ChatGPT, including a made-up word on an unlinked page, submitted through Google Search Console, that ChatGPT then repeated while Bing returned nothing for it. That is two anecdotes, not documentation, and OpenAI's current page names no Google provider.

Does ChatGPT use Bing?

Partly, as far as OpenAI says. At launch in October 2024, SiliconANGLE quoted OpenAI's help page as sharing queries "with third-party search providers such as Bing." The current page still links Microsoft's privacy statement, but neither version says Bing is the only source.

What search engine do other AI assistants use?

Here is what each major AI assistant says about its own retrieval, from the operator or the closest documented source.

Platform	Where answers are retrieved from	Source
ChatGPT search	Partner providers (Microsoft and Shopify named) plus OpenAI's own crawler, OAI-SearchBot. No database named.	OpenAI help page and crawler docs, above
Perplexity	Its own index. PerplexityBot "is designed to surface and link websites in search results on Perplexity." Vespa says Perplexity built on its engine, "indexing billions of web pages," each stored "in both textual form and as collections of vector embeddings."	Perplexity crawler docs, Vespa, 2025-04-15
Claude	Brave Search, added to Anthropic's subprocessor list on 2025-03-19; Simon Willison matched Claude's citations to Brave's results. Anthropic also runs Claude-SearchBot to "improve search result quality."	Simon Willison, Anthropic crawler docs
Google AI Overviews and AI Mode	Google's own Search index. To appear, "a page must be indexed and eligible to be shown in Google Search with a snippet." Google adds: "There are no additional technical requirements."	Google Search Central
Gemini	Grounding with Google Search "connects the Gemini model to real-time web content"; the model "automatically generates one or multiple search queries and executes them."	Gemini API grounding docs
Microsoft Copilot	Bing's index. Bing Webmaster Tools' AI Performance report counts citations in "Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations."	Bing Webmaster blog, 2026-02-10

None of these engines documents Elasticsearch or OpenSearch as the layer that decides what gets retrieved. That does not prove none of them runs either one anywhere; large companies run Elasticsearch for logs, dashboards and internal tools. It means the claim that these two decide whether your brand is seen has no operator behind it.

Three caveats. Vespa's account of Perplexity is a vendor case study, though the architecture it describes, a web-scale index Perplexity fills itself, matches Perplexity's own crawler docs. We could not load Anthropic's subprocessor list to re-check the Brave entry today, and two 2026 write-ups say the same list now also names turbopuffer, a vector search company, for web search; that is unverified by us, and it is neither Elasticsearch nor OpenSearch.

For how each engine decides what to say once it has retrieved your pages, see how LLMs prioritise information about brands and how Perplexity cites sources.

The Elasticsearch and OpenSearch claim, checked

Claim	Verdict	What decides it
AI search engines run on Elasticsearch or OpenSearch	Not documented	No engine above names either. Perplexity, the one with a published stack, is on Vespa.
Faster vector search gets your brand cited more	Wrong scope	The source is Elastic benchmarking OpenSearch on a product catalog. It never mentions an AI engine.
Get your content indexed inside these engines	No publisher lever	You cannot write to another company's index. Its crawler fetches your page, or it does not.
Real-time vectorization closes the "readiness gap"	Operator's problem	Index freshness is engineered by the engine. Your side is being fetchable when the crawler arrives.
Both help engineers see inside their own AI apps	Accurate	A different question: observability, not how AI describes your brand.

Where the claim is true

Elasticsearch and OpenSearch are real retrieval backends for AI. They are what many companies use when they build their own assistant: a support bot over help docs, a search box over a catalog, an internal knowledge tool. If you run one of those, recall, filtering and index freshness decide whether your bot answers correctly.

That matters to your chatbot, not to ChatGPT. ChatGPT never queries your Elasticsearch cluster, and Perplexity never reads your OpenSearch index. External AI engines reach your brand one way: a crawler requests your URL over HTTP, and their own pipeline does the rest.

What the speed benchmark measured

The "faster recall" claim traces to a benchmark Elastic published on 2026-02-25: Elasticsearch 9.3.0 against OpenSearch 3.5.0, on a 20-million-document ecommerce catalog with 128-dimension vectors, across six cloud servers. Its headline is "Elasticsearch vector search is up to 8x faster than OpenSearch." It is a vendor measuring its competitor on a workload it chose, and it does not mention ChatGPT, Perplexity or any AI search engine.

Why the backend does not decide your AI visibility

Every AI answer about your brand passes through stages with different owners. The database question sits entirely inside the stages the operator owns:

• Vector recall, quantization and filtering. The operator's settings, on the operator's hardware.
• How your page gets chunked. Google's AI optimization guide says "There's no requirement to break your content into tiny pieces for AI to better understand it." Segmentation happens inside the operator's index.
• Where your page lands in the candidate list. C-SEO Bench (NeurIPS 2025) found that moving a document higher in the model's context was "significantly more effective" than the content rewrites it tested. No publisher can move it there.
• How fast the index refreshes. That is the operator's engineering problem. On your side, no AI vendor we checked says it weights page dates, and the IndexNow specification never mentions AI.

You do not put documents into Perplexity's index; its crawler takes them, or it does not. The same holds for Brave, Bing, Google and OpenAI, and whatever database sits behind them can only hold what its crawler managed to fetch and read.

What you control: five checks that work whatever the backend

Every engine in the tables above is fed the same way: a bot requests your page. If it gets a 403, a challenge page or an empty JavaScript shell, the fastest vector search in the world has nothing of yours to find.

1. Allow the search crawlers in robots.txt, one by one. OAI-SearchBot for ChatGPT search, PerplexityBot, Claude-SearchBot, Googlebot and Bingbot. Training crawlers such as GPTBot are a separate decision; OpenAI says "Each setting is independent of the others."
2. Request a page as each crawler and confirm a 200. OpenAI states eligibility as two conditions: allow OAI-SearchBot and "confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses." If you are on Cloudflare, read should I block AI crawlers on Cloudflare first.
3. Put the facts in the raw HTML. Vercel's AI crawler study found ChatGPT and Claude crawlers fetch JavaScript files but do not run them: "They can't read client-side rendered content." A price that appears only after a script runs does not exist to them.
4. Remove accidental suppression. noindex, nosnippet, max-snippet:0 and data-nosnippet take content out of AI answers while the page looks normal to people.
5. Make your site agree with itself. One price, one product name, one founding year across the pricing page, FAQ, footer and About page. When an engine retrieves two versions of a fact, you handed it the contradiction.

Since OpenAI names Microsoft as a ChatGPT search provider, Bing Webmaster Tools is also worth a login. Its AI Performance report shows "the key phrases the AI used when retrieving content," the only report we know of that shows the queries an engine actually ran. None of these checks makes an engine prefer you; they decide whether you are eligible at all, which is the only layer with a documented causal path and the only one you own.

How to check the next claim like this

An AI answer is a retrieval result: it tells you who published on a topic, not who is right about it. Before acting on one, ask three questions.

1. Does the source document the mechanism, or describe somebody else's system?
2. Is the claim additive or subtractive? "Do X and AI will favor you" has almost no surviving evidence. "Remove what blocks you" is where the documented levers are.
3. Could you tell if it failed? If the result is unobservable, the advice is unfalsifiable.

Then test the answer itself. Our free AI visibility checker shows what ChatGPT and Google AI Overviews say about your brand, and how ChatGPT decides which brands to recommend covers what shapes that answer.

Frequently asked questions

Does ChatGPT have its own search index?

OpenAI runs its own search crawler, OAI-SearchBot, which "is used to surface websites in search results in ChatGPT's search features." Its Codex documentation also refers to "an OpenAI-maintained index" for cached web search. OpenAI does not publish how that index relates to the partner providers it uses.

What database does ChatGPT use?

OpenAI does not publish its retrieval stack. It documents query rewriting, partner search providers and its own crawler. Anyone selling a ChatGPT tactic built around a specific database is guessing at a system OpenAI has not described.

What search engine does Perplexity use?

Its own. PerplexityBot crawls the web to build Perplexity's search results, and Vespa reports that Perplexity runs that index on Vespa's engine. To be in it, allow PerplexityBot in robots.txt and make sure your firewall lets it through.

What search engine does Claude use?

Brave Search, according to Anthropic's subprocessor list as of March 2025, with Claude's citations matching Brave's results in Simon Willison's test. Anthropic also runs Claude-SearchBot to improve search quality, and blocking it "may reduce your site's visibility and accuracy in user search results."

Should I move my site search to Elasticsearch or OpenSearch to improve AI visibility?

No. Your site search serves your visitors; ChatGPT, Perplexity, Claude and Google never query it. Pick it on its merits for your own users, and spend the AI-visibility effort on whether their crawlers can reach your pages.

To see which crawlers can reach you today, sign up free and run the 99Visibility crawler access check, which requests your pages as each AI crawler and flags every place your robots.txt and your firewall disagree.