Quick answer: ChatGPT doesn’t run on a single search engine. Microsoft confirms its Web IQ platform, built on Bing’s index, powers ChatGPT. OpenAI’s own help center points users to Microsoft’s privacy statement for its search partners. Independent captures from mid-2026, before GPT-6, show the rest of the picture: free ChatGPT leaned mainly on an index OpenAI runs itself, paid Thinking mode pulled most of its results from scraped Google, and Deep Research was logged running on Bing. As of October 2026, even shopping is split, with US results drawn from OpenAI’s own product feeds and European results from Google Shopping. OpenAI has never confirmed any of the split. Here’s the paper trail, what’s confirmed versus assumed, and the two gates your site has to pass to show up.
Last time, we followed the paper trail behind Claude and found Brave Search named plainly on Anthropic's vendor list (What Search Engine Does Claude Use?). OpenAI is less forthcoming. You won't find ChatGPT's search engine on a sub-processor page. You'll find it scattered across a privacy link, a Microsoft product page, a court filing, and the data stream ChatGPT sends to your own browser. Put those pieces together and the answer isn't one engine. It's several, routed by mode, by market, and by the question itself.
Last verified: 19 October 2026. Most observations below predate GPT-6, which OpenAI rolled out to paid ChatGPT plans on 7 October 2026 and to Free and Go users from 8 October (OpenAI). OpenAI says GPT-6 starts answering web-search questions faster; it hasn't said whether it searches differently. We found no published independent study of GPT-6's retrieval as of 19 October 2026.
ChatGPT's digital footprint
Exhibit A: Microsoft said it first (2023). At Build 2023, Microsoft announced it was bringing Bing to ChatGPT as the default search experience. That was the plugin era, before ChatGPT's search was rebuilt. It's a historical marker: it tells you who the partner was, not who it is today.
Exhibit B: OpenAI's launch post (October 2024). When ChatGPT search launched in its current form, OpenAI said it draws on third-party search providers plus content supplied directly by partners (OpenAI). It named no providers and published no split. That silence has held ever since.
Exhibit C: The privacy link. OpenAI's current help center says ChatGPT search sometimes partners with other search providers, rewriting your question into targeted queries before sending them out. To learn how those partners handle your queries, it points you to exactly two documents: Microsoft's privacy statement and Shopify's privacy policy. OpenAI never writes "Microsoft is our search provider." It simply sends you to Microsoft to find out what happens to your searches. There is one exception. On its help page for Enterprise and Edu workspaces, OpenAI says ChatGPT may share "disassociated search queries with the Bing search engine" to return web results (source). The same page says ChatGPT for Healthcare workspaces don't use Bing at all. So OpenAI does name Bing, but only for business plans.
Exhibit D: Microsoft names ChatGPT (2026). Microsoft is less coy. Its product page for Web IQ, the AI grounding platform it announced in June 2026 as built on the Bing global index, says it "powers leading frontier AI assistants, including Microsoft Copilot and ChatGPT." OpenAI names Bing only for its business workspaces. Microsoft names ChatGPT as a customer outright, and we found no other supplier that has done the same.
Exhibit E: Under oath (2025). In the US government's antitrust case against Google, the plaintiffs' proposed findings of fact, filed 29 May 2025, cite testimony from OpenAI's Nick Turley about the plumbing. OpenAI began grounding ChatGPT on third-party search results in 2023. In 2024 it asked Google for access to its search API, and Google declined. Turley described significant quality issues with OpenAI's non-Google search partners. The plaintiffs' responsive findings, filed the same day, add that Turley testified OpenAI was not on track to serve even 80% of queries with its own "index technology", his term for the whole ranking and retrieval system, by the end of 2025.
Exhibit F: The Deep Research logs (June 2026). Peec AI recorded ChatGPT Deep Research sessions across more than ten free accounts. Every search payload carried the same label, web_with_bing, with no Google and no fallback. The scope matters: this is Deep Research, the long-running research agent, not the everyday answer most people see. Bing may not be confined to Deep Research. RESONEO reports that in June, Oncrawl saw a separate bing pipeline supply 93.7% of sources on a ChatGPT Team account, while RESONEO's own accounts never showed it. RESONEO suspects the Team plan, but couldn't verify it.
Exhibit G: OpenAI's own index (July–August 2026).
- OpenAI's own index: on the free tier, most search results came from "labrador," an index OpenAI runs itself. OpenAI's help center describes "indexed and cached web content" for offline search in eligible workspaces (source); it has never described the consumer index or used the name. RESONEO showed the index is not Bing: a quarter of its titles run longer than Bing ever displays.
- Google: two other pipelines turned out to be scraped Google results.
- The mix depends on the mode: in August, free users who clicked Think drew 74.7% of their results from OpenAI's index and 3.1% from scraped Google. Paid Thinking drew 75.3% from scraped Google (RESONEO).
- Corroboration: Peec AI's September investigation found the same index and counted at least eight outside providers, including Yelp, TripAdvisor and Microsoft's Web IQ.
Exhibit H: Shopping splits by continent (October 2026). The newest finding concerns products. Peec AI's debug data shows ChatGPT Shopping in the US now runs on product feeds OpenAI indexes itself, while Europe still pulls from Google Shopping. Same assistant, same month, two supply chains.
Exhibit I: Regulators call it a search engine (August 2026). On 31 August 2026, the European Commission designated ChatGPT a Very Large Online Search Engine under the EU's Digital Services Act, the first AI chatbot to receive that status. The Commission classed it as a hybrid service that qualifies as a search engine because it can search the web in response to prompts. OpenAI reported about 159.1 million average monthly EU users of ChatGPT's search function in the six months to 31 March 2026. The designation brings ranking transparency, independent audits and data access for regulators and vetted researchers. It does not reveal which engines ChatGPT uses, but it settles whether ChatGPT is a search engine in law.
Put together, Bing is no longer the only engine in the record. Microsoft is a confirmed supplier, confirmed by Microsoft. OpenAI documents an index of its own for workspace offline search, and independent captures show an OpenAI index carrying most free-tier search. Scraped Google carries most paid reasoning, despite Google refusing OpenAI's request for official access. Bing surfaces in Deep Research. "ChatGPT uses Bing" is true but incomplete. "ChatGPT uses one search engine" is false. And OpenAI has confirmed none of the split.
The two months ChatGPT showed its plumbing
For a brief window, OpenAI's machinery was visible to anyone who knew where to look. Between 21 May and 21 July 2026, ChatGPT's data stream carried a field called result_source that named the pipeline behind every search result, according to Peec AI. It took only four values: labrador, OpenAI's own index; bright and oxylabs, two contractors that scrape Google; and serp, a residual channel.
Researchers moved quickly. Chris Green ran 1,000 prompts up to ten times each, capturing 9,946 completed searches, and found labrador behind 88.1% of primary search sources (Search Engine Land). The routing was not stable, either. 11.6% of prompts switched their main source between identical runs, and when they did, the overlap in cited URLs fell from 0.273 to 0.149. Ask the same question twice and you may be reading from two different corpora.
RESONEO found that the question itself decides the route (RESONEO):
- Stable facts, free instant mode: 99.9% of results came from OpenAI's index.
- News and anything changing within the day: split almost evenly, 52.6% OpenAI's index and 46.1% Google.
- Paid Thinking: scraped Google dominated every category of question.
Where ChatGPT's search results came from, by mode
Observed by independent researchers, May to October 2026, before GPT-6. Not confirmed by OpenAI.
web_with_bing (June)Routing was not fixed: 11.6% of prompts switched their main source between identical runs. GPT-6's routing has not been independently measured. Sources: RESONEO, Peec AI, Search Engine Land.
Then, on 21 July, the field disappeared overnight. Most of the source-split figures in this post come from that window, or from methods built to reconstruct it.
Why OpenAI built its own index
The motive is on the court record. According to the plaintiffs' proposed findings of fact in the US antitrust case against Google, OpenAI's Nick Turley testified that ChatGPT began grounding answers on third-party search results in 2023. In 2024 OpenAI asked Google for access to its search API and was refused. Turley also described "significant quality issues" with OpenAI's non-Google search partners (plaintiffs' proposed findings of fact). The plaintiffs' responsive findings add that he testified OpenAI was not on track to serve even 80% of queries with its own index technology by the end of 2025 (plaintiffs' responsive findings of fact).
The response has been to build. Peec AI found OpenAI job postings describing indexing systems at multi-region, exabyte scale. It also found live experiments testing OpenAI's own index against scraped results, most visibly in shopping, where one experiment, "prefer-index-over-serp-v3," was running on 8% of chats in mid-August 2026 (Peec AI). The index is not a finished replacement. It is being tested, slice by slice, against the providers it is meant to replace.
From retrieved to cited: the funnel
Being found and being cited are separated by a steep funnel. Across RESONEO's July 2026 corpus, ChatGPT retrieved about 58,000 URLs. Roughly 7,600 were promoted to the visible sources, about 5,000 were cited in the answer text, and only around 760 pages were actually opened and read (RESONEO).
That last step is where citations are won. In RESONEO's July sample, a page ChatGPT opened was cited 74% of the time, against 7% for a page that was merely retrieved. In July, opening a page was almost entirely confined to paid Thinking mode. Since August, the free tier's Think button opens pages too, though rarely.
What gets read is also narrower than most publishers assume. In RESONEO and Oncrawl's July 2026 observations:
- Copies are shared and reused: ChatGPT cached the pages it fetched and shared the copy across users. A copy was treated as fresh for about 30 minutes and could be served for months afterwards.
- Structured data doesn't reach the model when ChatGPT reads a page: JSON-LD was stripped first. What the index does with it is unknown.
- A hard size limit: pages over 4 MB were rejected outright with an HTTP 400, with no partial read (Search Engine Land).
Scripts are a separate risk. Peec AI found Deep Research reads pages through a text browser that runs no JavaScript (Peec AI), and OpenAI lists script-dependent content as a common reason pages are missing from its index (OpenAI).
What's left is the text in your HTML, and its position matters. Deep Research reads pages top to bottom in source order, in a capped first window of roughly 5,000 to 6,000 characters, so a heavy navigation block can crowd out your answer (Peec AI). OpenAI's own index works the same way: its snippet is taken from the start of your page body, anchored on your H1 (RESONEO).
How rarely ChatGPT searches at all
All of this applies only when ChatGPT decides to search, and on the free tier that is the exception. Ahrefs analysed 66.6 million free, logged-out responses from GPT-5.6 Luna in August and September 2026. Only 14.4% triggered a web search, averaging about 1.6 queries each (Ahrefs). For most prompts, ChatGPT answers from what it already knows.
When it does search, it increasingly goes straight to the source:
- Brand sites first: 75.9% of the most common site-restricted searches targeted brand websites.
- "Official" rising: the word appeared in 13.1% of English fan-out queries in September, up from 10.1% in August.
- Summary sites falling: Reddit citations in US answers dropped 88% between March and September, while regulators, research bodies and brand support pages climbed.
The implication is uncomfortable for anyone whose content summarises other people's facts. ChatGPT is learning to cite whoever publishes the fact first.
Shopping: two continents, two rulebooks
Peec AI's analysis of ChatGPT's shopping debug data shows the split runs deeper than a choice of supplier (Peec AI).
In the US, products come from feeds OpenAI indexes itself, and the rules are strict:
- Feeds: 669 at the time, mostly built by scraping retailer sites, alongside direct brand partnerships and Shopify.
- Verified sellers only: a shop without the flag drops out of results entirely.
- Short descriptions: only the first 50 tokens of each product description are scored, roughly 40 words.
In Europe, ChatGPT still pulls from Google Shopping through a scraping service. There's no seller check and no relevance scoring of its own; results are taken in turn from each search query ChatGPT runs. For now, a European brand's ChatGPT shopping visibility is effectively its Google Shopping visibility. Peec expects that to change when OpenAI's own index reaches Europe.
What we don't know
An honest account of this subject has to mark where public knowledge ends.
- The current mix. The best measurements date from July and August 2026. GPT-6 reached all ChatGPT users in October, and we found no published study of how it routes searches as of 19 October 2026.
- How pages enter OpenAI's index. RESONEO's test pages saw OAI-SearchBot almost silent while the index kept serving results, and no new page entered during a 16-day watch. Which crawler adds new pages to the index is unconfirmed.
- Bing's real share. Oncrawl's Jérôme Salomon reported a
bingpipeline supplying most sources on a ChatGPT Team account in June; RESONEO never saw it on its own accounts. The field that would settle it is gone. - The split itself. OpenAI documents an index for workspace offline search and names Bing for business plans. It has never published how consumer traffic divides between its sources.
The two gates most guides miss
Most advice stops at "get indexed in Bing." That was incomplete when it was first written, and it is more so now. There are still two gates. What has changed is who guards the first one.
Gate one: being found. Which index finds you depends on the route the question takes. Pre-GPT-6 studies observed OpenAI's own index on the free tier, scraped Google in paid Thinking, and Bing in Deep Research; GPT-6's routing hasn't been independently measured yet. So the safe move is the same either way: be properly indexed in both Google and Bing, and let OpenAI's crawler in.
How OpenAI's index stays current: RESONEO and Oncrawl found that its snippets line up with visits from OAI-SearchBot, OpenAI's search crawler. How a page first gets in is still unknown.
What ChatGPT sees of you: in OpenAI's index, that snippet is usually your full title plus roughly 200 characters from the start of your page body, anchored on your H1.
Meta descriptions: OpenAI's index ignores them, though Google-fed results still use them about one time in three.
Make the first lines under your H1 say what you most want ChatGPT to know.
Gate two: being read. Discovery isn't permission. OpenAI's crawler documentation says sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers," though they can still appear as navigational links. You're findable but not quotable, which is arguably worse: a competitor's words get read aloud while your name sits underneath as a footnote.
Here's the twist. OAI-SearchBot now matters at both gates. Block it and OpenAI says your content won't be shown in search answers. Its visits also appear to keep OpenAI's own index current, which matters for the free tier's main discovery route. Passing one gate isn't enough. You need both.
OpenAI's crawlers, and which one actually matters
| User agent | What it does | Affects ChatGPT search? |
|---|---|---|
| OAI-SearchBot | Surfaces sites in ChatGPT's search features; independent research links its visits to updates in OpenAI's own index | Yes. This is the one. Block it and OpenAI says your content won't be shown in search answers. |
| GPTBot | Collects content that may be used to train OpenAI's models | No. Opting out of training doesn't hide you from search. |
| ChatGPT-User | Visits a page when a user's request triggers it | No. OpenAI says it isn't used to decide search eligibility, and robots.txt rules may not apply to it. |
| OAI-AdsBot | Checks landing pages submitted as ChatGPT ads | Only if you advertise. |
The settings are independent. You can block GPTBot from training on your content and still allow OAI-SearchBot to show you in search. OpenAI uses exactly that configuration as its own example. If your robots.txt has a blanket "block all AI" rule, you've probably blocked the one bot your visibility depends on. Check your CDN or firewall too: OpenAI asks sites to allow traffic from its published OAI-SearchBot IP ranges, and a security rule can turn the bot away even when robots.txt lets it in.
Copy-paste robots.txt: show up in ChatGPT search, but keep your content out of model training.
User-agent: OAI-SearchBot Allow: / User-agent: GPTBot Disallow: /
Want your content used for training too? Delete the last two lines. If your platform doesn't let you edit robots.txt directly, check its AI crawler settings instead, and make sure OAI-SearchBot isn't caught by a blanket "block AI" option.
Confirmed vs. assumed
Last verified: 19 October 2026. Most observations predate GPT-6.
| Claim | Status | Source |
|---|---|---|
| ChatGPT search uses third-party search providers | Confirmed by OpenAI | OpenAI launch post; OpenAI Help Center |
| Microsoft supplies ChatGPT search | Confirmed by Microsoft, 2026 | Microsoft Web IQ |
| Bing powers search in Enterprise and Edu workspaces | Confirmed by OpenAI (Healthcare workspaces excluded) | OpenAI Help Center |
| Bing was named ChatGPT's default search experience | Confirmed for 2023 only. Historical, not current. | Microsoft, Build 2023 |
| OpenAI runs its own search index | Partly confirmed. OpenAI documents "indexed and cached web content" for offline search in eligible workspaces. The consumer index, its name "labrador," and its share of traffic come from independent observation, July to September 2026. | OpenAI Help Center; RESONEO; Peec AI |
| OpenAI sought Google's search API and was refused | Sworn testimony, 2025, as cited in the plaintiffs' proposed findings of fact | US v. Google, plaintiffs' proposed findings of fact |
| OpenAI was not on track to serve 80% of queries from its own index technology by the end of 2025 | Sworn testimony, 2025, as cited in the plaintiffs' responsive findings of fact. Turley used "index technology" to mean the whole ranking and retrieval system. | US v. Google, plaintiffs' responsive findings of fact |
| Free Think leans on OpenAI's index; paid Thinking leans on scraped Google | Observed, August 2026: 74.7% OpenAI index (free Think); 75.3% scraped Google (paid Thinking) | RESONEO |
| Deep Research retrieves through Bing | Observed in logs, June 2026 | Peec AI |
| US shopping uses OpenAI's own feeds; Europe uses Google Shopping | Observed, October 2026 | Peec AI |
| Regulators classify ChatGPT as a search engine | Confirmed by the European Commission, August 2026: designated a Very Large Online Search Engine under the Digital Services Act | European Commission |
| ChatGPT uses one search engine for everything | False | All of the above |
| Ranking well on Google or Bing carries over to ChatGPT | Not guaranteed. In one 2026 sample, only 1.5% of OpenAI-index URLs appeared in Bing's top 20 for the same queries. | RESONEO via Search Engine Land |
| How GPT-6 retrieves | Unknown | No public study found as of 19 October 2026 |
The ten-minute check
- Open yourdomain.com/robots.txt. Make sure OAI-SearchBot isn't disallowed, either by name or by a wildcard rule. If it's blocked, OpenAI says your content won't be shown in ChatGPT search answers.
- Check your firewall or CDN. OpenAI asks sites to allow traffic from its published search-bot IP ranges (openai.com/searchbot.json). A security rule can block the bot even when robots.txt allows it.
- Check you're indexed in both Google and Bing. Search
site:yourdomain.comon google.com and bing.com. Pre-GPT-6 studies found paid Thinking pulling most of its results from scraped Google and Deep Research running on Bing. If key pages are missing, submit your sitemap in Google Search Console and Bing Webmaster Tools. - Wait a day. OpenAI says robots.txt changes take about 24 hours to register for search.
- Measure it, with two caveats. ChatGPT adds
utm_source=chatgpt.comto the links it sends, so referral traffic shows up in Google Analytics under that source. First, after GPT-6's release, links in ChatGPT's answer text began arriving without the tag or a referrer, so those visits landed in Direct (Search Engine Roundtable). Links in the Sources panel and shopping links kept the tag. Reports place the start around 8 October. Adobe's Cedric Huesler reported that OpenAI told him it was a bug and began rolling out a fix on 15 October; some measurements showed recovery, but OpenAI hasn't confirmed the fix publicly and the exact end date is unconfirmed (Search Engine Journal). Annotate your reports from 8 October until your own ChatGPT numbers recover. Second, independent research found that pages ChatGPT opens and reads itself in Thinking mode carry no tag, so the citations worth the most are the ones your analytics miss (RESONEO). - Front-load your answers. In OpenAI's own index, ChatGPT usually sees your full title plus roughly 200 characters from the start of your page body, anchored on your H1. That index ignores your meta description, though Google-fed results still use it about one time in three. Deep Research reads more, but only a capped first window of each page's HTML, so heavy navigation can crowd out your content before the agent reaches it. Put your core answer directly under your H1.
- Watch your AI citations in Bing Webmaster Tools. Its AI Performance report, in public preview since February 2026, shows how often your pages are cited in AI answers across Microsoft Copilot, Bing's AI summaries and select partner integrations, plus the grounding queries the AI used to find them (Microsoft). Since June 2026 it also shows your citation share for each query and groups grounding queries by intent and topic, so you can see why your pages get cited, not just how often (Microsoft). In October 2026 Microsoft added Topics Portfolio, a beta tab that groups your grounding queries into themes and lists queries where other sites are cited but yours isn't. Its topics are AI-classified, its figures are estimates, and it only appears once Bing has enough data for your site (Search Engine Roundtable). Microsoft doesn't say which partners are included or name ChatGPT, so treat all of this as a signal for Microsoft's side of the picture, not a ChatGPT report.
FAQ
Does ChatGPT use Google to search the web?
In some modes, yes. Independent captures from August 2026 found paid Thinking mode drew 75.3% of its results from scraped Google, and as of October 2026 ChatGPT Shopping in Europe pulls from Google Shopping. It isn't by agreement: according to testimony cited in the US v. Google case, OpenAI asked Google for search API access in 2024 and was turned down. OpenAI has never confirmed using Google, and its help center links only to Microsoft's and Shopify's privacy policies for its search partners.
Does ChatGPT have its own search index?
Yes, by every independent measure. OpenAI's help center describes "indexed and cached web content" for offline search in eligible workspaces, and ChatGPT's data stream labelled a consumer index "labrador" until July 2026. RESONEO showed it is neither Bing nor Google, and it supplied most search results on the free tier in July and August 2026. OAI-SearchBot's visits appear to keep it current; how a page first enters it is still unknown.
If I block GPTBot, will I disappear from ChatGPT search?
No. GPTBot governs training. Search visibility is controlled by OAI-SearchBot, and OpenAI treats the two settings independently.
Can ChatGPT show my site even if I've blocked it?
Yes, partially. OpenAI's crawler documentation says sites that opt out of OAI-SearchBot can still appear as navigational links, without their content in the answer. For its Atlas browser, OpenAI recommends the noindex meta tag to keep a page out entirely, and its crawler has to be allowed in to read that tag (OpenAI). One caution: independent research found noindex pages still stored in ChatGPT's page cache, so treat noindex as a request, not a guarantee.
Does a good Google or Bing ranking guarantee I'll be cited?
No. OpenAI says placement isn't guaranteed, and the overlap is thin: in one 2026 sample, only 1.5% of URLs from OpenAI's own index appeared in Bing's top 20 for the same queries. Being indexed in Google and Bing gets you considered in some modes. Allowing OAI-SearchBot keeps you eligible to appear and lets your content be read. Being cited depends on whether your page is the clearest, most credible answer, and pages ChatGPT actually opens get cited far more often than pages it only retrieves.
Not sure whether Google, Bing and OpenAI's own index even know your site exists, or whether a security rule is quietly turning OAI-SearchBot away? That's part of what we check. Get your free consultation, and we'll show you exactly where you stand.
Get your free consultationSources (verified 19 October 2026)
OpenAI
- Introducing ChatGPT search (Oct 31, 2024)
- GPT-6 and Intelligent UI for everyone (Oct 7, 2026)
- Help Center: Searching the web with ChatGPT
- Help Center: ChatGPT search for Enterprise and Edu
- Help Center: Offline web search for ChatGPT workspaces
- Overview of OpenAI Crawlers
- Publishers and Developers FAQ
Microsoft
- Microsoft Web IQ (names ChatGPT among the assistants it powers)
- Microsoft Bing Blog: Announcing Microsoft Web IQ (June 2, 2026)
- Microsoft Bing Blog: Introducing AI Performance in Bing Webmaster Tools (Feb 10, 2026)
- Microsoft Bing Blog: New AI Visibility Insights in Bing Webmaster Tools (June 16, 2026)
- Microsoft Bing Blog: Bing at Microsoft Build 2023 (May 23, 2023, historical)
- CNBC: Microsoft says Bing can be default search engine for ChatGPT users (May 23, 2023, historical)
Court records
- US v. Google: Plaintiffs' Remedies Proposed Findings of Fact (May 29, 2025)
- US v. Google: Plaintiffs' Remedies Responsive Proposed Findings of Fact (May 29, 2025)
Regulatory records
Independent research
- RESONEO: What ChatGPT pulls, what it shows, what it cites (July 2026, updated August 2026)
- Search Engine Land: Inside ChatGPT's retrieval stack (Aug 17, 2026)
- Search Engine Land: ChatGPT citations change when hidden search pipelines switch (July 8, 2026)
- Peec AI: How ChatGPT Deep Research reads your site (June 22, 2026)
- Peec AI: ChatGPT built its own search index (Sept 4, 2026)
- Peec AI: ChatGPT Shopping optimization in 2026, US vs. Europe (Oct 16, 2026)
- Ahrefs: How ChatGPT does query fan-out, 15 million queries studied (Oct 12, 2026)
- Search Engine Roundtable: Topics Portfolio In Bing Webmaster Tools AI Reports (Oct 13, 2026)
- Search Engine Roundtable: OpenAI Drops UTM Parameters From ChatGPT Results (Oct 14, 2026)
- Search Engine Journal: ChatGPT Traffic Drop Appears To Be A Tracking Problem (Oct 15, 2026)
- Danu Agency: What Search Engine Does Claude Use?
