Five AI engines, five source preferences: how to approach GEO in Taiwan
We ran 14,598 AI queries in Taiwan. Each of five engines cites different sources, and B2B flips B2C: vendor-site citations on ChatGPT range from 19.0% to 96.1%.
Five AI engines, five source preferences: how to approach GEO in Taiwan
Key takeaways
- Each of the five engines has its own source preferences, and the gaps are wide. ChatGPT cites government sources 10.7% of the time, three to five times more than the other engines. Review and editorial sites make up 40.5% of Perplexity's citations, so nearly half of what it cites lives outside your website. Gemini cites brand websites the most (33.1%) and social media not at all. Which engine you want to show up in determines where your content needs to live.
- B2B and B2C point in opposite directions. On B2B purchasing prompts, 81.2% to 96.1% of citations across the five engines come from vendor websites, while third-party content accounts for only 1.0% to 11.8%. Put the same engines on consumer product prompts and vendor websites drop to 19.0% to 34.2%, while third-party content rises to 33.9% to 60.4%.
- Engines don't share sources, so winning one doesn't lift another. Across 166 matched sets where Google AI Overviews, AI Mode and Gemini all returned citations, 80.7% had not a single source in common. All three are Google products.
We ran 14,598 AI engine queries. Of those, 8,120 followed a matched design: same brand, same prompt, same day, asked across all five engines at once, with source overlap compared only within each matched set. That design cancels out differences in prompts, timing and industry, so whatever differences remain come from the engines themselves.
We set out to answer one question: before each engine writes its answer, where does it go to read?
We tested five major engines, and each prefers different sources
The engine you want to reach determines whether your resources belong on your own website, on review sites or on social media. The table below shows citation sources for a Traditional Chinese consumer prompt set for Taiwan, covering five categories (home appliances, health supplements, software tools, travel and personal finance) and built from real consumer questions.
| Source type | AI Overviews | AI Mode | ChatGPT | Perplexity | Gemini |
|---|---|---|---|---|---|
| Review and editorial sites | 25.7% | 17.0% | 36.6% | 40.5% | 28.1% |
| Brand websites | 28.6% | 32.1% | 18.9% | 19.9% | 33.1% |
| Magazines | 8.3% | 12.1% | 10.7% | 10.7% | 11.0% |
| E-commerce | 3.7% | 10.3% | 3.4% | 1.7% | 6.2% |
| Government | 3.3% | 2.2% | 10.7% | 2.2% | 2.4% |
| Forums | 2.9% | 7.6% | 3.2% | 6.6% | 2.1% |
| News media | 4.6% | 3.6% | 6.8% | 1.0% | 2.4% |
| Blogging platforms | 3.7% | 0.9% | 1.6% | 4.4% | 4.5% |
| Price comparison sites | 5.4% | 0.4% | 3.6% | 2.7% | 2.9% |
| Social media | 5.4% | 9.4% | 1.0% | 1.7% | 0% |
| Video | 1.7% | 0% | 1.2% | 1.2% | 1.1% |
ChatGPT checks official documents and ignores online chatter. Government sources make up 10.7% of its citations, while the other four engines all sit between 2% and 3%. It also has the highest share of news media at 6.8%, yet social media accounts for just 1.0%. If you have regulatory references, official statistics or public-sector partnerships, this is the only engine that takes them seriously.
AI Mode picks up social media and e-commerce. It leads both categories, with social media at 9.4% and e-commerce at 10.3%. On social media, only AI Overviews comes close at 5.4%; ChatGPT, Perplexity and Gemini are all under 2%. The content you build up on social platforms gets picked up here and nowhere else.
Perplexity bets on third-party reviews and forums. It has the highest share of review and editorial sites at 40.5%, and its forum share of 6.6% is second only to AI Mode. Together these two categories account for nearly half of its citations, all of them outside your own website. To get into Perplexity, you need other people writing about you.
Gemini cares most about what you write yourself. It has the highest share of brand website citations at 33.1% and cites social media 0% of the time. Self-published content such as case studies, tutorials and spec sheets carries the most weight with this engine.
AI Overviews leads on price comparison sites at 5.4%, while AI Mode cites them just 0.4% of the time. Two interfaces from the same company differ thirteenfold on this one source type.
On Perplexity, review and editorial sites (40.5%) outweigh blogging platforms (4.4%) roughly ninefold. Ten posts on Medium will do far less for you on this engine than a single feature on a review site.
Beyond the engine, consider your industry: B2B is the mirror image of B2C
In B2B, your website is almost the only place that matters. Put the same engines on enterprise purchasing prompts and the source mix flips entirely:
| Engine | Vendor websites (B2B) | Vendor websites (B2C) | Third-party content (B2B) | Third-party content (B2C) | Social media and forums (B2B) | Social media and forums (B2C) |
|---|---|---|---|---|---|---|
| ChatGPT | 96.1% | 19.0% | 1.0% | 59.6% | 0.2% | 4.2% |
| Gemini | 89.4% | 34.2% | 6.0% | 50.5% | 0.5% | 2.1% |
| AI Mode | 85.2% | 32.1% | 8.4% | 33.9% | 2.6% | 17.0% |
| AI Overviews | 82.0% | 29.1% | 9.9% | 48.5% | 4.2% | 8.4% |
| Perplexity | 81.2% | 20.3% | 11.8% | 60.4% | 0.1% | 8.4% |
On B2B purchasing prompts, 81.2% to 96.1% of citations across the five engines point to vendor and brand websites. Move the same engines back to consumer product prompts and that share falls to 19.0% to 34.2%.
Third-party content moves the other way. Review sites, media, blogs and price comparison platforms make up only 1.0% to 11.8% of citations in B2B, but more than a third in B2C, reaching 60.4% on Perplexity. The gap for social media and forums is even wider: in B2B, ChatGPT cites them 0.2% of the time and Perplexity 0.1%.
Why such a big difference? Niche B2B categories have no unboxing reviews, no forums debating specs and little media coverage. When AI has to answer "which supplier can do this," the only material available is what suppliers have written about themselves. Whether third-party content about your category exists online determines whether AI has anywhere else to look.
So the standard GEO playbook (get featured on review sites, build word of mouth on forums, put out press releases) accounts for less than a tenth of citations in B2B, all three combined. The advice isn't wrong; its data simply comes from consumer scenarios.
For B2B brands, that means flipping the order of priorities. Spec sheets, use cases, technical decision guides and FAQs are the pages AI actually cites in your category. ChatGPT is the most extreme case: 96.1% of its citations come from vendor websites, so how complete your website is almost single-handedly decides whether you get mentioned there.
Google's own three interfaces don't share sources either, so winning one won't lift the others
If you make it into an AI Overviews answer, there's close to a 60% chance you're nowhere to be found in AI Mode.
For the same prompt on the same day, when AI Overviews, AI Mode and Gemini all returned citations, the three shared at least one source only 19.3% of the time. In the remaining 80.7%, each went its own way, and together they cited an average of 4.9 distinct sources.
| Engine pair | Matched samples | Source overlap rate | No shared sources at all |
|---|---|---|---|
| AI Overviews × AI Mode | 339 | 23.0% | 57.8% |
| AI Mode × Gemini | 370 | 15.5% | 65.4% |
| AI Overviews × Gemini | 254 | 9.6% | 78.3% |
| ChatGPT × Gemini (cross-vendor comparison) | 181 | 9.2% | 74.0% |
| ChatGPT × Perplexity (cross-vendor comparison) | 246 | 3.7% | 91.1% |
Bring in the two cross-vendor pairs for comparison: AI Overviews and Gemini have zero overlap 78.3% of the time, higher than the 74.0% for ChatGPT and Gemini. Two products from the same company share fewer sources with each other than with a competitor.
Intuition says one company's products should share an index. Judging by the results, the more plausible explanation is that their retrieval pipelines run independently: AI Overviews sits on the search results page and draws on the existing search index, while Gemini runs its own grounding process. The two differ in when they fetch, how much they fetch and how they rank it.
AI Overviews and AI Mode are the most similar pair of all, and they still have zero overlap on 57.8% of prompts.
Being cited on ChatGPT won't carry over to Perplexity automatically. Coverage across all engines needs its own budget line; it isn't a bonus you get for free after winning one.
How to allocate content resources: check whether your category has third-party content, then set the balance between your website and off-site
The first question in deciding where to put resources is: "Does third-party content about my category exist online?" Answer that before deciding which engine to target, because it reverses everything that follows.
Your category has no third-party content (most niche B2B categories). Nobody writes reviews, nobody discusses it on forums, the media doesn't cover it. Put nearly all your resources into your website: spec sheets, use cases, technical decision guides, FAQs. In this scenario, 96.1% of ChatGPT's citations come from vendor websites, so the completeness of your site almost single-handedly decides the outcome. Investing in review sites and press releases before your website is complete has limited payoff.
Your category has mature third-party content (most B2C categories). Review sites, price comparison sites and forums are all writing about your category. Your website can only capture 19.0% to 34.2% of citations; the rest are off-site. Here, one feature on a review site does more than ten blog posts of your own.
A sequence that works in both scenarios. For reach, start with AI Overviews and AI Mode: they sit on the search results page, so users encounter them without opening any separate tool. For high-intent traffic along the decision chain, go after ChatGPT, and finish your website first, since 96.1% of its B2B citations come from vendor websites. For third-party word of mouth, go after Perplexity, where 40.5% of citations come from review and editorial sites.
Where AI goes for answers depends half on the engine and half on what your category has left behind online. That's why the same ChatGPT draws only 19.0% of its citations from brand websites on consumer product prompts, but 96.1% on B2B purchasing prompts.
If nobody reviews your category or discusses it on forums, that apparent disadvantage is actually good news: AI has no choice but to come back and read your own website, the one place in the whole chain you fully control.