How does ChatGPT search work and where do its answers come from?

The short answer to "how does ChatGPT search work" is this: ChatGPT search is the feature that lets ChatGPT look things up on the live web and answer with links to the pages it used, instead of relying only on what its model learned in training. When your question needs current information, ChatGPT rewrites it into one or more web searches, sends those to outside search providers, reads what comes back and writes an answer in its own words, with citations you can click. This guide explains where those answers come from, which OpenAI crawlers decide whether your website can be part of them, and what that means for a business in Australia.

What ChatGPT search is, in plain words

OpenAI introduced ChatGPT search on 31 October 2024, first for paying Plus and Team users and people on the SearchGPT waitlist. It opened to all logged-in users on 16 December 2024 and to everyone in regions where ChatGPT is available, with no sign-up, on 5 February 2025.

As of October 2026, OpenAI lists web search as available on the Free, Go, Plus, Pro, Business, Enterprise and Edu plans, and to people who are not signed in, on chatgpt.com and in the desktop and mobile apps. Searches count against your plan's usage limits.

Two places ChatGPT's answers come from

Source one: what the model learned in training. OpenAI says its foundation models are built from three kinds of information: publicly available internet content, data accessed through agreements with third parties, and information provided by users, human trainers and researchers. The model keeps no copies of those pages; training adjusts its internal parameters so it gets better at predicting the next most likely word. Because several words can plausibly come next, there is built-in randomness, which is why the same question can produce a different answer on a different day. Whatever the model knows about your business from training is paraphrased, undated and sometimes wrong.

Source two: live web results. When a question would benefit from current information, ChatGPT may search automatically, or you can choose Search from the tools menu. OpenAI says the search feature uses third-party search providers and content provided directly by partners, including news and data providers for weather, stocks, sports, news and maps. For a business these are two separate doors, controlled by two different crawlers.

What happens when you ask ChatGPT a question

This is the sequence OpenAI describes.

1. ChatGPT decides whether to search. OpenAI says it searches when your question would benefit from current information, so a question about a local business or a recent event is a likely trigger and a request to tidy up an email is not.
2. It rewrites your question into search queries. ChatGPT typically turns your prompt into one or more targeted queries for its search providers. OpenAI's example: asked about the latest drugs that target CCR8 for cancer, ChatGPT first searches for "CCR8 immunotherapy drug development 2025", then, after reading those results, may send a narrower query such as "CHS-114 conference 2025".
3. It adds context. ChatGPT works out your approximate location from your IP address and may pass that general location to search providers, so "good restaurants near me" can become "top restaurants San Francisco". If memory is on, saved details count too: a vegan in San Francisco might get "good vegan restaurants San Francisco". OpenAI says it does not pass your IP address or account details to search providers.
4. It reads the results and writes an answer, with citations. Selecting a citation opens the source, and the Sources button lists the cited pages and other relevant links. Follow-ups keep the full context of the chat, so the next search is shaped by everything said before it.

You may hear step two called query fan-out: one question becomes several searches behind the scenes, each worded by the model, which is how a page can be cited for a question that never contained its keywords.

Does ChatGPT use Google or Bing?

OpenAI does not name a single engine, so anyone who gives you a confident one-word answer is guessing. Its help page says ChatGPT search "sometimes partners with other search providers" and points readers to two privacy policies for how those providers handle queries: Microsoft's and Shopify's. So Microsoft is one documented provider, OpenAI says nothing more specific, and Google is not named as a search provider on that page. Shopify's product catalogue is integrated into ChatGPT's shopping results.

OpenAI also runs its own search crawler, OAI-SearchBot, which it says is used to surface websites in ChatGPT's search results. So ChatGPT does more than re-show another engine's top ten: it can draw on providers' results, partner content and pages its own crawler has visited, and the model decides what to use. Keep your site open to that crawler and keep doing ordinary SEO, rather than optimising for one engine and assuming you are covered.

How ChatGPT chooses which sources to cite

OpenAI's published position is short: ChatGPT ranks search results using multiple factors intended to help people find relevant, reliable information, and placement is not guaranteed. Because the output is a written summary with inline citations and a Sources panel, a page can inform an answer without being named in the main text.

Shopping is the one area where OpenAI says more. When a question shows buying intent, ChatGPT can show product options with images, details and links, chosen on your query and context (including memory and custom instructions), structured product data such as price and description, and OpenAI's safety and product policies. When several merchants sell the same item, OpenAI says it weighs things such as stock availability, price, quality and whether the seller made the product or is its main seller. OpenAI says product results are not ads and are not influenced by its partnerships.

For ordinary pages, our advice is good practice rather than OpenAI rules, because that is all the evidence supports. The system reads pages and summarises them, so a page that answers the real question near the top, in plain language, with current facts and clear headings, is easier to quote accurately than one that buries the answer under a brand story.

The three OpenAI crawlers and what each one controls

OpenAI publishes a page describing its crawlers. Three matter to a business, and they do different jobs.

• OAI-SearchBot is the search crawler, used to surface websites in ChatGPT's search results. OpenAI says sites that block it will not be shown in ChatGPT search answers, though they can still appear as navigational links, and recommends allowing it in robots.txt along with requests from its published IP ranges.
• GPTBot is the training crawler. It collects content that may be used to train OpenAI's foundation models, and disallowing it tells OpenAI your content should not be used for training.
• ChatGPT-User fetches a page when a person's request in ChatGPT or a Custom GPT needs it. OpenAI says it does not crawl automatically, robots.txt rules may not apply because the fetch is user-initiated, and it plays no part in search inclusion.

The settings are independent. OpenAI's own example is a site that allows OAI-SearchBot to appear in search results while disallowing GPTBot to opt out of training; if you allow both, it may use one crawl for both purposes. After a robots.txt change, OpenAI says its search systems can take about 24 hours to adjust.

robots.txt in one paragraph

robots.txt is a plain text file at the root of your site that gives rules to crawlers, grouped by crawler name. The Robots Exclusion Protocol behind it was published as an IETF standards-track specification, RFC 9309, in September 2022. Its rules are requests that well-behaved crawlers are asked to honour; the standard itself says they are not a form of access authorisation. A minimal file that keeps you in ChatGPT search while opting out of training looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Whether to block GPTBot is a business decision; the second group is there only if you want to make it. The common snag sits outside robots.txt: a CDN or firewall that blocks OpenAI's IP ranges even though robots.txt says yes.

Shopping, local and deeper research in ChatGPT

Shopping

Merchants can apply to share a product feed with OpenAI; Shopify and Etsy catalogues are already integrated. OpenAI says a feed is optional if ChatGPT already crawls your site, but gives you more control over how products appear. OpenAI's merchant page says, as of October 2026, that shopping is live in the United States first, with more regions to follow, and that purchases are completed on the merchant's own site or app. So for an Australian retailer a direct feed is mainly preparation for now; in the meantime, crawlable product pages are what ChatGPT's ordinary web search can read and cite.

Local

ChatGPT uses an approximate location from your IP address for local results. Precise device location is optional and off by default; if a user turns it on, OpenAI says ChatGPT can answer questions about nearby businesses, local news and weather more specifically. OpenAI also says it may share location with trusted third-party providers of local information, such as business listings. OpenAI's own tip for better local results is to put the city, neighbourhood or postcode in the question. For an Australian business, keep your location details accurate and consistent on your site and in your listings.

Deep research

Deep research, launched on 2 February 2025, is a separate mode that OpenAI says finds, analyses and synthesises hundreds of online sources into a report, so the answer a customer reads may have been built from dozens of pages.

Why this matters for your business

In September 2025 OpenAI published a study of 1.5 million consumer conversations, alongside a figure of 700 million weekly active users. About half of messages, 49 per cent, were what the researchers called Asking: seeking information or advice. OpenAI's own reading is that people value ChatGPT most as an adviser. When one of those people asks which accountant to use in Parramatta or which heat pump suits a Melbourne winter, the businesses named in the reply are the ones that get considered.

Ads exist in ChatGPT too. OpenAI began testing them in the United States on 9 February 2026 for Free and Go users, and says they run on separate systems, appear below a response, are clearly labelled, and that advertisers cannot shape, rank or alter the answer itself. Being cited organically is therefore a separate job from advertising, and it is the job our SEO for ChatGPT in Australia service exists to do.

How to get started: making your site findable by ChatGPT

For most small businesses the first pass is an afternoon's work.

1. Check robots.txt allows OAI-SearchBot. Open yoursite.com.au/robots.txt. If a blanket rule blocks every crawler, add a group for OAI-SearchBot that allows it. Decide about GPTBot separately.
2. Check your host or CDN is not blocking OpenAI's IP ranges. Many security products block bots by default. Ask your developer or host to confirm that OpenAI's published searchbot addresses can reach the site.
3. Make your key pages answer real questions. List the ten questions customers ask before they buy and give each a page or section that answers it plainly near the top, with a clear heading and current prices or dates. This is good practice rather than an OpenAI rule, but plain pages summarise well.
4. Keep business details consistent. ChatGPT localises searches and pulls local data from listing providers, so your trading name, address, suburb, phone number and hours should match everywhere they appear.
5. Retailers: consider a product feed. Shopify and Etsy catalogues are already integrated and other merchants can apply for direct feed access. OpenAI says shopping is live in the United States first as of October 2026, so treat this as preparation.
6. Keep doing core SEO. ChatGPT search draws candidates from search providers and its own crawler, so crawlable, well-linked, well-written pages feed both. If the terms are new to you, our beginner's guide to AI SEO explains how it differs from classic SEO.

If you would rather hand the work over, we offer it in Australia and as SEO for ChatGPT in Singapore.

Common mistakes and myths

• "Blocking GPTBot keeps me out of ChatGPT search." It does not, and allowing GPTBot does not get you in. The two crawlers have separate jobs and separate settings; OAI-SearchBot controls search.
• Blocking every AI bot in one blanket rule. Plenty of sites added a catch-all block, or a firewall setting that does the same thing, and then wonder why they never appear.
• "ChatGPT is just Google, or just Bing, with a chat box." OpenAI describes multiple third-party providers, partner content and its own crawler, and the model typically rewrites your question before it searches.
• "You can pay to be recommended." OpenAI says product results are not ads and that advertisers cannot shape, rank or alter answers.
• Treating ChatGPT's answer as always right. OpenAI itself warns that search results and citations can be incomplete, outdated or incorrect, and tells users to open the cited source. Apply the same scepticism to what it says about your business.

How to tell if ChatGPT is sending you visitors

Start with your analytics. In Google Analytics the Source dimension records where a visit originates, and visits from ChatGPT usually show chatgpt.com as the source; when a link carries a utm_source tag, that value also fills the manual source dimensions. Tags can go missing, though. In October 2026, for example, ChatGPT's inline links lost their tracking tag (and arrived without a referrer) for about a week, and those visits landed in Direct instead; we wrote up that ChatGPT referral drop in GA4 at the time. If your ChatGPT traffic falls off a cliff, check Direct first.

Clicks are only part of the picture, because many people read the summary and never click. The second measure is a prompt check: write down the 10 to 20 questions your customers ask, put them to ChatGPT with search on each month, and record whether you are named, linked or absent, and who is named instead. Because answers vary and are personalised, treat one screenshot as an anecdote and a quarter of results as a trend.

Frequently asked questions

Does ChatGPT search work in Australia?

Yes. OpenAI makes web search available in every region where ChatGPT is offered, on the free plan and to people who are not signed in. The one caveat is shopping: OpenAI's merchant page says shopping is live in the United States first, as of October 2026, with other regions to follow.

Can I get ChatGPT to correct something wrong about my business?

For search-based answers the most reliable lever is the source: fix the page the answer came from, since ChatGPT re-reads live pages, and flag the poor answer with ChatGPT's feedback controls. OpenAI's privacy portal handles requests about personal information; for business details we are not aware of a published process beyond that, and no outcome is guaranteed.

Is ChatGPT search the same as Google's AI Overviews?

No. They come from different companies, use different crawlers and run on different systems, so being visible in one does not mean you are visible in the other. The groundwork overlaps, though: our guides to generative engine optimisation and answer engine optimisation explain how the different AI answer engines choose what to quote.

What should I do first?

Check robots.txt and your firewall for OAI-SearchBot, then ask ChatGPT your customers' ten questions in a fresh chat with search on and write down whether you are named, linked or missing. That tells you whether your problem is access or content. The full method, from how generative engines choose sources through crawler access and pages that get quoted to measuring a channel with no rankings, is in the GEO playbook.