ChatGPT vs Gemini vs Perplexity: how each picks the brands it names

In any ChatGPT vs Gemini vs Perplexity comparison, the part that matters to a brand is where each one looks: ChatGPT searches with its own OAI-SearchBot crawler and, at times, outside search providers; Gemini grounds some of its answers in Google Search; Perplexity searches its own index for every question. In October 2026, we know of no public study that shows which of the three names the most brands, and the studies that do exist mostly show that they cite different websites. Being named by one tells you little about the other two, or about Google's AI Mode and AI Overviews.

Hypervisible, the software we make, asks all three the same questions on its five-engine plans, along with Google's two AI features, so the gaps between them show up side by side in one report.

Three assistants, three ways to answer

ChatGPT and the Gemini app decide when to search the web, while Perplexity searches every time. Google's AI Overviews and AI Mode, which run on Gemini models inside Google Search, follow rules of their own, so a brand tracking AI answers really has five patterns to deal with.

Based on help pages and documentation from OpenAI, Google and Perplexity, read on October 22, 2026.
Engine	When it searches	Where it looks	How it shows sources	Crawler to allow
ChatGPT	When current information would help, or when you ask	Pages from OAI-SearchBot, partner search providers	Inline citations and a Sources button	OAI-SearchBot
Gemini app	For some answers	Google Search results	A Sources button, when there are sources	Googlebot, with Google-Extended not blocked
Perplexity	Every question	Its own index	Numbered citations	PerplexityBot
AI Overviews	Only when Google judges it adds something	Google's index	Links to supporting pages	Googlebot
AI Mode	Every question asked in AI Mode	Google's index	Inline links, or a list of links when confidence is low	Googlebot

The split between always-search and sometimes-search matters more than it looks. When ChatGPT answers without searching, it works from what its model learned in training, which OpenAI says draws on publicly available internet content, data from partners, and data from users and trainers. Citations come with the answers that go through search.

For a brand, that means two separate jobs. One is what the model already knows about you. The other is what it finds when it goes looking, and which pages it trusts once it gets there.

ChatGPT: web search and citations

ChatGPT searches the web on its own when it judges that current information would help, or when you tell it to. Those answers carry citations, and a Sources button lists the cited pages along with other links ChatGPT found relevant.

Search is open on every plan, Free through Enterprise and Edu, and to people who aren't signed in. To force it, choose Search from “View all tools,” or type / in the message box. On desktop, hovering over a citation shows a preview.

The most revealing part of OpenAI's help page sits in its privacy section. ChatGPT sometimes works with outside search providers, and when it does, it typically rewrites your question into one or more targeted queries and sends those instead. The page links to the privacy policies of Microsoft and Shopify. ChatGPT also passes along an approximate location based on your IP address, and it may use saved memories: OpenAI's own example turns a request for nearby restaurants, from a user who has said they're vegan and live in San Francisco, into “good vegan restaurants San Francisco.” Two buyers typing the same question can set off different searches.

The rule for websites is written down. To be eligible, a site has to let OAI-SearchBot crawl it and accept traffic from OpenAI's published IP addresses. According to OpenAI's crawler documentation, a site that opts out of OAI-SearchBot disappears from ChatGPT search answers, apart from navigational links, and robots.txt changes take about 24 hours to register. GPTBot is a different crawler, used for model training. You can block it and keep OAI-SearchBot.

OpenAI doesn't publish its ranking criteria. Its help page says only that ChatGPT ranks results on multiple factors meant to help people find relevant, reliable information, that placement isn't promised, and that results and citations can be incomplete, outdated or wrong. Ads don't buy a place in that ranking: OpenAI says they run on systems separate from the chat model, as covered in our guide to ChatGPT ads. For the full mechanics, read how ChatGPT chooses its sources.

Gemini: built on Google results

The Gemini app grounds some of its answers in Google Search results, according to Google's Gemini Apps Privacy Hub. When a response has sources, a Sources button appears below it or inline and opens a side panel of links.

Not every response gets one. Google's help page is plain about it: no Sources button means Gemini provided no links for that answer. When Gemini quotes a long passage from a web page, that page shows up in the list, and images taken from the web come with their source.

Google also warns that Gemini can misrepresent how it works, including how it cites sources. Asking Gemini why it never mentions your company won't produce a reliable diagnosis, and the same goes for every other assistant.

The setting site owners need to know is Google-Extended. This robots.txt token lets a publisher refuse the use of its pages for training Gemini models and for grounding answers in the Gemini app, which Google defines as feeding the model content from the Search index at prompt time. Google states that Google-Extended has no effect on a site's inclusion in Google Search. A site that blocks it can still be cited in AI Overviews and AI Mode while dropping out of the pages the Gemini app can draw on.

AI Overviews and AI Mode work differently from the app. Both run several related searches across subtopics in Google's index, a technique Google calls query fan-out, and link to pages that are indexed and eligible to show with a snippet. Google shows an AI Overview only when its systems judge that one adds something to classic results, and says that often isn't so. AI Mode answers whatever it's asked, and falls back to a list of web links when it isn't confident in a written answer. Google also notes that the two may use different models and techniques, so their links vary. Controls for both go through Googlebot rules in robots.txt, the nosnippet, data-nosnippet, max-snippet and noindex tags, and a Search Console setting open to every site since August 31, 2026. Our guide to AI Overviews covers where their links come from.

Perplexity: numbered sources

Perplexity searches the web in real time for every question, according to its help center, and writes an answer studded with numbered citations that link back to the original pages.

Pro subscribers can pick the model that writes the answer. The Research mode goes further, running dozens of searches and returning a full report.

Perplexity says it runs its own index. When it launched its Search API on September 25, 2025, the company described that index as covering hundreds of billions of web pages and as the same infrastructure behind its public answer engine. Those are Perplexity's own figures.

On the crawler side, PerplexityBot is the one that matters. Perplexity's documentation says it is designed to show and link websites in Perplexity's search results and isn't used to collect content for training AI foundation models, and the company recommends allowing it along with its published IP ranges. A second agent, Perplexity-User, fetches pages when a person asks a question, and because a user requested the fetch, it generally ignores robots.txt.

The numbering has a practical upside for brands. You can see which page supports which sentence. If Perplexity quotes a price you changed last spring, the citation points to the review that still lists the old one, and that's the page to get updated.

Which one recommends brands most often

We know of no public study that answers that across industries, and we won't invent a ranking. Three dated studies compare how these assistants cite, each on a narrow task, using 2025 versions that have changed since.

Published studies, read on October 22, 2026. All tested 2025 versions.
Study	What was tested	Measure	Finding on sources
EBU and BBC, October 2025	2,709 answers, 22 public service media in 18 countries, May 24 to June 10, 2025, free versions	News answers rated by journalists	Significant sourcing issues in 72% of Gemini answers, 24% for ChatGPT, 15% for Perplexity
Tow Center, Columbia Journalism Review, March 2025	1,600 queries across 8 tools, February 2025	Find the article, publisher and URL behind a news excerpt	Perplexity wrong 37% of the time, ChatGPT on 134 of 200; over half of Gemini answers cited made-up or broken URLs
University of Toronto, September 2025	Consumer ranking questions, sent through each engine's API	Type of sites cited	For well-known brands, 93.5% third-party sources for ChatGPT, 23.8% social for Perplexity, 25.1% brand sites for Gemini

The EBU and BBC study is the largest, and NPR was among the participants. Across all answers, sourcing was the biggest single cause of significant problems, affecting 31% of responses. Gemini stood out: 42% of its answers gave no direct link to a specific piece of content. Each prompt asked the assistant to use the participant's sources where possible. On one NPR question, Gemini framed its answer as drawing from NPR sources, and none of the 11 sources it listed came from NPR.

The Tow Center tested something narrower, matching a news excerpt to its source. Together, the eight tools got it wrong on more than 60% of queries. For brands, the lesson concerns risk: an assistant can cite the wrong page, or a link that goes nowhere.

The University of Toronto study comes closest to buying questions, with prompts like “best smartphones” and “best electric car for commuting.” AI engines leaned heavily on third-party sources such as publishers and review sites, where Google mixed in more brand sites and social platforms. For electric cars, ChatGPT cited Wikipedia, AP News and Reuters, Perplexity relied on YouTube and Car and Driver, and Google's top results included Reddit and Quora. The overlap was thin: in consumer electronics, of all the domains ChatGPT or Perplexity cited, fewer than 9% were cited by both. The study queried APIs, which don't always behave like the consumer apps.

None of the three crowns a winner for brands. They do agree that each assistant has its own sourcing habits, and that alone is reason to check them one by one.

What to do if one cites you and another doesn't

Start with access, then compare the sources each assistant uses, then measure over several weeks. A brand can be named by Perplexity and missing from ChatGPT on the same question. Before reading that as a verdict on your brand, check the plumbing.

Check crawler access. Look at robots.txt and at your CDN or firewall rules for four names: OAI-SearchBot (ChatGPT), PerplexityBot, Googlebot (AI Overviews and AI Mode) and Google-Extended (Gemini app grounding). A bot-blocking rule at the hosting level can remove you from one assistant and leave the others untouched.
Compare the cited pages. Ask each assistant the questions your buyers ask, such as “best HVAC company in Phoenix” or “best CRM for a 10-person real estate team.” If Perplexity names you through a YouTube review while ChatGPT cites three publications that never mention you, the gap lives in those publications.
Work on what the model already knows. For answers given without search, a clear and consistent description of your company, on your site and across the web, matters more than last week's blog post.
Re-ask on a schedule. OpenAI notes that the same question can produce different answers from one query to the next. One missing mention proves nothing; a gap that holds for weeks is worth acting on.

Because answers vary from one session to another, a fixed place in them isn't something anyone can honestly sell you. Getting cited still takes work, and the results of that work can be measured. For Copilot and other engines, see AI search engines compared.

Comparing the three side by side with Hypervisible

Tracking this by hand across three assistants and Google's AI answers adds up fast, which is why NEXUS GEO, our agency in France, which does nothing but GEO, built Hypervisible on data from Google and the AI engines. We make it, so here are our reasons for putting it forward, for you to weigh.

The same questions on every engine. ChatGPT and Gemini are read together with Google AI Overviews on every plan, and the five-engine plans bring in Perplexity and Google AI Mode. The studies above found that the assistants cite different sites, and one report on identical questions is where that gap becomes visible.
Sources as well as names. For each brand you see how often the answers mention it (its citation rate), where it usually sits in them, its standing against competitors in that same set of answers, the domains each engine cites, and what is said about it and in what tone.
A weekly series. On a paid plan, each Monday brings a fresh run of the same questions and the gap in points since the last one, which is step 4 of the list above done for you (pricing).

Its limits in one sentence: the reading is weekly, on Mondays, and neither Claude nor Copilot is among the engines it measures. For a baseline first, try the free AI visibility checker. It works with 10 questions, five about your market where your brand isn't mentioned and five where it is, each answered by three engines: ChatGPT, Gemini and Google AI Overviews. Reading those 30 replies takes 5 to 10 minutes, at no point do you hand over credit card details, and the results are stored in your account. Perplexity and Google AI Mode are read on the five-engine plans.

Your brand

The free analysis asks 10 questions to three AI engines (ChatGPT, Gemini and Google AI Overviews), reads the 30 answers in 5 to 10 minutes and keeps your dashboard in your account.

Frequently asked questions

Is Gemini better than ChatGPT?

It depends on what you use them for, and no public study ranks them on how often they name brands. Their sources differ: the Gemini app grounds some answers in Google Search, while ChatGPT relies on its OAI-SearchBot crawler and partner search providers. In the EBU and BBC study of free versions in mid-2025, 72% of Gemini's news answers had significant sourcing issues, against 24% for ChatGPT. Both have changed since, so a brand should track both.

Is Perplexity better than ChatGPT?

Perplexity is easier to check, since every answer carries numbered citations to its sources. It still makes mistakes. In the EBU and BBC study (May and June 2025), 30% of Perplexity's news answers had at least one significant issue, compared with 36% for ChatGPT. In the Tow Center's February 2025 test, Perplexity gave a wrong answer on 37% of queries asking for the source of a news excerpt. Open the cited pages before relying on any answer.

Which AI cites its sources best?

By design, Perplexity: its help center says every answer includes numbered citations linking to the original sources. ChatGPT shows citations and a Sources button on answers that use web search. The Gemini app shows links on some answers only. In the EBU and BBC study, run in May and June 2025 on free versions, 42% of Gemini's news answers had no direct link to a specific source.

Why would one AI mention my brand and not another?

Because they don't search the same index or trust the same sites. ChatGPT uses OAI-SearchBot and partner search providers, Gemini and Google's AI answers use Google's index, and Perplexity uses its own. A 2025 University of Toronto study found little overlap between the domains ChatGPT and Perplexity cited. A blocked crawler, or an answer given from memory without any search, can widen the gap further.

Does Gemini use Google search results?

Partly. Google's Gemini Apps Privacy Hub says some Gemini responses are grounded in Search results, and those answers can show a Sources button with links. Site owners can opt out of that grounding with the Google-Extended robots.txt token, which, according to Google, has no effect on inclusion in Google Search. Google AI Mode is a different product: it runs searches for every question inside Google Search.

Does blocking GPTBot remove my site from ChatGPT?

No. GPTBot is the crawler OpenAI uses to gather content that may train its models. ChatGPT's search answers depend on OAI-SearchBot, and OpenAI's documentation says the two settings are independent: a site can block GPTBot and still allow OAI-SearchBot. Blocking OAI-SearchBot is what keeps a site out of ChatGPT search answers, apart from navigational links. Changes take about 24 hours to apply.

How can I track which AI assistant mentions my brand?

Hypervisible, the software we make at NEXUS GEO, puts the same questions to ChatGPT, Gemini and Google's AI Overviews every Monday, and to Perplexity and Google's AI Mode on its five-engine plans, then shows each engine's citation rate, your position and the sites it cites. Its free checker reads 30 answers from three of those engines (ChatGPT, Gemini and Google AI Overviews) in 5 to 10 minutes without a credit card. By hand, ask each assistant your buyers' questions on a fixed schedule and log who gets named and which pages are linked.