ChatGPT vs Gemini vs Perplexity: Which Cites Business Sites More?

We cannot name a winner, and we are wary of anyone who does without showing a sample size. Our own numbers come from one tracker round on 2026-10-25 for the 7 sites of Netbase JSC (the company behind Leadee): CMSmart, Printcart (Netbase-owned), SoftWSP, NetbaseJSC, Web2Print, AdminKit and OutsourcingVN. Every answer came from an official API: OpenAI gpt-5 with the web search tool, Perplexity sonar and Google AI Mode through DataForSEO. No consumer chat window was scraped.

On the questions the three engines shared, the rates sit between 3% and 12%, and their 95% intervals overlap almost completely. A difference that small on samples that size is noise. What the data does show clearly is something else: the engines differ in whether they link to the brand they name, and in how much one answer costs to measure. Both matter more for a small business than a league table. For the full ChatGPT picture, start at the ChatGPT SEO overview.

Each brand had 8 US buyer questions and 4 questions for a second market (UK, Germany or Singapore). Perplexity sonar and Google AI Mode answered each question 3 times. OpenAI's account ran out of credit after 25 runs, so ChatGPT only answered the first questions of each brand, all of them unbranded "best ..." or "alternatives" questions. In total that is 492 runs: 25 ChatGPT, 252 Perplexity and 215 Google AI Mode.

Two events were counted per run. "Named" means the answer mentioned the brand in a form that cannot be confused with something else (full name, domain, or a citation). "Cited" means the brand's own domain appeared among the answer's source links. What ChatGPT citations are explains why these are different things.

Gemini itself was not measured: the Google AI Studio project had no billing enabled on that date. Google states that AI Mode runs on a custom version of Gemini 2.5, so AI Mode is the closest engine in our data, but it is a search feature with its own retrieval, not the Gemini app. Read every "Gemini" statement on this page with that in mind.

Across all runs, Perplexity named a Netbase brand in 46 of 252 runs and Google AI Mode in 36 of 215. ChatGPT named one in 3 of 25. These totals are not comparable as they stand: Perplexity and AI Mode were also asked questions with the brand name in them ("CMSmart vs Fancy Product Designer"), which name the brand most of the time, while ChatGPT never got one.

Restricted to questions without a brand name, Perplexity named a Netbase brand in 13 of 210 runs (6.2%) and AI Mode in 7 of 173 (4.0%). ChatGPT's 3 of 25 (12%) looks higher, but its interval runs from 4.2% to 30.0%, wide enough to include both. How to check if ChatGPT mentions your brand shows why a rate needs this many runs.

The fair comparison uses only the 13 questions ChatGPT also answered. On those, ChatGPT named a Netbase brand in 3 of 25 runs (12.0%, interval 4.2-30.0%), Perplexity in 4 of 39 (10.3%, 4.1-23.6%) and Google AI Mode in 1 of 38 (2.6%, 0.5-13.5%). All of the mentions came from CMSmart's two WooCommerce product-designer questions; the other 11 questions gave no Netbase mention on any engine.

So on the same questions, ChatGPT and Perplexity look alike and AI Mode a little lower, and none of the gaps is statistically meaningful. One brand in one niche drives the whole result. If your niche resembles CMSmart's (a plugin market where the brand's own site already ranks), you might see something similar; if it resembles NetbaseJSC's (software outsourcing, where lists come from directories), you might see zero everywhere, as we did. (Netbase-owned: part of Netbase JSC.)

This is the clearest difference in our data. Perplexity named a Netbase brand 46 times and linked to its domain 41 times. Google AI Mode named one 36 times and linked 29 times. In ChatGPT's 3 mentions, the brand's domain was in the citation list every time. ChatGPT's sample is tiny, so treat that as a hint, not a rule.

The gap between named and cited is mostly one brand. Web2Print Solutions (Netbase-owned) was named 7 times in the US (3 by Perplexity, 4 by AI Mode) and its domain was cited 0 times, even when the question asked about it by name. The engines took their facts about Web2Print from other sites. A mention without a link sends no visitor and gives you no control over what is said, which is why getting cited when ChatGPT prefers other websites treats the citation, not the mention, as the goal.

In OpenAI's API, a citation is a url_citation tied to a span of the answer text; Perplexity returns a citations list with each answer. Both engines also read pages they do not cite, so neither count measures how often your site was read.

Less than the title promises, and we prefer to say so. We have no Gemini API runs for these sites. Google AI Mode is powered by a custom Gemini 2.5 according to Google, and Google says it uses query fan-out, splitting a question into sub-topics and searching each. That makes AI Mode's behaviour a reasonable proxy for how Google's Gemini-based search features pick sources, not for the Gemini app with its own grounding settings.

AI Mode named Netbase brands at a rate close to Perplexity's across all runs and was the lowest of the three on the shared questions. It also had the largest share of names without a link (7 of 36). When we add Gemini runs, they will go into the same tables with their own sample size, and this page will be updated rather than replaced.

Because each API bills differently. In the same round, 25 ChatGPT runs cost 2.39 USD (about 0.096 USD per answer), 252 Perplexity runs 1.30 USD (about 0.005) and 215 Google AI Mode runs 0.26 USD (about 0.0012). A ChatGPT answer with web search cost roughly 20 times a Perplexity answer and 80 times an AI Mode answer.

That difference shapes what anyone can measure. With a fixed budget you get far fewer ChatGPT runs than Perplexity runs, so ChatGPT rates come with wider intervals. It is also why our OpenAI sample is the smallest: the account hit its credit limit first. The ChatGPT monitoring cost calculator shows what a prompt set costs per month.

For most sites, no. The groundwork is shared: crawlers allowed (OAI-SearchBot for ChatGPT search, PerplexityBot, Googlebot), the important facts in server-rendered HTML, and passages that state an answer with a number and a source. How to write pages ChatGPT can quote covers the passage work; ChatGPT SEO vs Google SEO covers what differs for OpenAI's crawler.

Where engines differ is in which third-party pages they lean on, and that differs by niche more than by engine in our data: software services lists came from directories such as clutch.co, as ChatGPT SEO for professional services firms shows. Measure your own unbranded questions on the engines your buyers use, with at least 3 runs per question, and compare engines only on the same questions. Manual checks vs automated tracking shows how much one run can mislead.