Dan Taylor, Partner at SALT.agency and founder of International Search Insights (ISI), is speaking at the Asia SEO Virtual Summit 2026 on 23 October 2026. His talk, “How Asia’s Search Engines Built Their Own AI Answers: Being chosen, and accurate, beyond Google”, looks past Google and Bing to how Baidu, Naver, DeepSeek, Tencent, Cốc Cốc and Yandex now generate answers, and what that means for brands.
The talk puts three questions to every engine. Is the brand chosen, meaning named and recommended rather than used as raw material? Is the answer accurate on product, price and claims? And is the demand it creates measurable in citations, clicks, brand search and enquiries?
Baidu: a million citations, 39 clicks
Baidu rebuilt its default experience around ERNIE. By July 2025, more than 90% of Baidu App monthly users were interacting with AI Search, according to an ISI report, and the app has 700M+ monthly active users. Robin Li has named search agents as Baidu’s main agent priority for 2026.
Dan cites a Politecnico di Milano audit of 500 intents. It found Baidu AI overviews on 82.0% of Chinese queries but only 4.0% of English ones. Citations are heavily concentrated: the top host takes 27.1% on Baidu against 6.6% on Google, with Baidu’s own Baike, Baijiahao and Zhidao among the most cited. Source overlap with Google is tiny, a host-set Jaccard index of 0.032.
Then there is what Dan calls the most honest number in AI search this year. In Baidu’s new AI Traffic Analysis reporting, one household-name brand site logged roughly 1,004,742 AI citations and about 39 AI clicks over 30 days. Baidu calculates AI CTR as clicks ÷ citations, so it can’t be compared directly with Google Search Console.
Naver: cited is not chosen
AI Briefing covers roughly 20% of Naver’s integrated search queries. A 650-keyword analysis found that why/how queries trigger it reliably, while “recommend” and “compare” commercial queries often get no Briefing at all.
A study of 635 non-branded queries and 85,392 citations found Naver Blog made up 46.5% of AI Briefing citations, 7.7% of Google AI Overviews citations, and 0% in ChatGPT and Gemini. Naver’s robots.txt rules block external AI crawlers. Even when a brand’s own channel was cited, AI Briefing named the brand in only 31.9% of answers, against 83.7–92.3% on Google, Gemini and ChatGPT.
The talk also covers this month’s changes: Related Questions becoming a standalone block from 22 October, the Blog and Cafe reshuffle reported by practitioners, and Advertiser Centre becoming N Ads on 4 November.
DeepSeek: from chatbot to search layer
On 17 October, DeepSeek opened Search and Fetch APIs at $0.75 per 1,000 successful requests. Every result carries a published authority tier (0–4), a rerank score (0–3) and a 600–800 character summary “intended to be read by a model”. DeepSeek has also open-sourced DeepInfer, its inference framework. Its deterministic mode matters for anyone trying to measure AI visibility in a repeatable way.
On reach, QuestMobile puts DeepSeek third among China’s AI apps at roughly 130m MAU, and Mediascope says 13m Russians used it in September 2026. In the US, a DHS order bans it from federal civilian agencies.
Tencent, Sogou, Cốc Cốc and Yandex
Tencent has opened Yuanbao Search as an API at ¥14–¥26 per 1,000 calls. Coverage says the technology derives from Sogou. Pages are compressed to about 100 tokens, and the index draws on 157 authoritative sources. In Vietnam, ISI’s 100-query test found Cốc Cốc AI answers on 40 of 50 informational queries but just 3 of 50 commercial ones. Yandex says Alice AI quick answers reached 49.5M monthly users in Q2 2026, covering 42% of queries. Ecommerce quick answers rose from 1 in 1,000 queries in April to about 15% in July.
Takeaways
- Cited is not chosen. One million citations became 39 clicks on Baidu, and Naver named brands in 31.9% of answers. Track whether you are named, not just used.
- Every engine feeds its own garden. That means Baidu Baike, Naver Blog and Cafe, Yandex Maps and Dzen, and Tencent’s authoritative sources. Plan presence engine by engine, not one “AI search” plan.
- Accuracy and measurement are yours to own. Answers are built from short extracts, so put the claim, price and date first. Report citations, clicks and brand demand separately for each engine.