This weekend we did something every monitoring company should do with its own product: we pointed it at someone famous and read what it found. The subject was Song Yuqi of (G)I-DLE — 宋雨琦 to her Chinese fans, 우기 to Korean ones. In one batch we ran 40 checks across five AI engines, and 36 of them came back flagged as mentioning her. One of those 36 confidently described a completely different human being.

That gap — between what the counter reports and what the answer actually says — is the most useful thing we've learned all month. Here is the full picture, misses included.

The setup: 5 engines, 8 prompts, 3 languages

Every day, PigPR asks a fixed set of prompts about each tracked brand and stores the complete answers. For Yuqi, the batch was 8 prompts × 5 engines = 40 checks. The engines:

  • ChatGPT
  • Gemini
  • Grok
  • Perplexity
  • Qwen

The prompts spanned English, Chinese, and Korean — the languages her audience actually types in: “who is yuqi”, “yuqi (g)i-dle”, “宋雨琦”, “우기 여자아이들”, “is (g)i-dle popular in china”, “yuqi solo music 2026”, “best (g)i-dle member”, and “yuqi chinese variety shows”.

The headline: 90% visibility — and two kinds of miss

36 of 40 checks flagged a mention, an AI Visibility score of 90%. Gemini and Grok were perfect at 8/8 — both, notably, already say “i-dle (formerly (G)I-DLE)”, so they're rebrand-aware. Perplexity and Qwen each landed 7/8, ChatGPT 6/8.

The four misses split into two very different categories. ChatGPT's two misses were tracking artifacts: asked “宋雨琦” and “우기 여자아이들”, it answered fully and correctly — in Chinese and Korean — but never used the romanized string “yuqi”, so string-based matching counted them as no mention. Perplexity's Korean miss was the same story. Correct answers, invisible to the counter. And then there was Qwen.

The answer about the wrong person

Asked “who is yuqi”, Qwen answered in Chinese. Translated, it said:

“Yuqi” (Yu Shuxin) is a Chinese actress and singer, born December 18, 1995 in Shanghai. She rose to fame on a 2018 talent show and later debuted in Korea in 2018 as a member of girl group (G)I-DLE, serving as sub-vocalist, lead dancer and visual. She is also active in Chinese film and TV, starring in hit dramas Love Between Fairy and Devil and My Journey to You.

Qwen, answering “who is yuqi” — originally in Chinese

That is Esther Yu — Yu Shuxin — a different celebrity, with Yuqi's (G)I-DLE membership grafted onto her biography. The real Yuqi is Song Yuqi, born September 23, 1999 in Beijing. And here is the uncomfortable part: the answer contains the string “Yuqi”, so our tracker counted it as a mention. A wrong-person biography scored exactly the same as a perfect profile.

A webtoon that doesn't exist, and a name parsed as a show title

Two more answers from the same batch are worth thirty seconds. Asked “우기 여자아이들” in Korean, Qwen described this:

“Ugi Yeoja-aideul” is a popular Korean webtoon, a youth romance by author Kim Bo-tong, serialized on Naver Webtoon since 2018. The protagonist “Ugi” is an introverted high-school girl who grows through school, friends and first love… An anime adaptation was confirmed for 2022.

Qwen, answering “우기 여자아이들” — originally in Korean

No such webtoon exists — fabricated author, fabricated plot, fabricated anime adaptation. And asked “yuqi chinese variety shows” in English, the same engine parsed her name as a TV title: “‘Yuqi’ is not a known Chinese variety show title… It may be a misspelling, shorthand, or unofficial reference.” The actual answer is Keep Running (奔跑吧), where Yuqi has been a cast MC since 2019.

The twist: same engine, different language, best answers of the five

If this were a story about a bad engine, we could end it here. It isn't. When the questions came in Chinese, Qwen produced the deepest, most current answers of all five engines — by a distance.

Asked “宋雨琦”, it returned the most detailed profile any engine gave: Cube trainee since 2015, resident MC on Zhejiang TV's Keep Running since 2019, first EP A Page, the solo upgrade YUQ1, a performance on the 2025 CCTV Spring Festival Gala. And asked “yuqi solo music 2026” — in English, answered in Chinese — it came back with hard numbers nobody else had:

  • Five Korean singles charted a cumulative 119 times on Tencent Music's Korean chart, peaking at TOP 2.
  • “Radio (Dum-Dum)” spent 71 consecutive weeks in the TME Korean chart TOP 10 through July 2026, hitting #1 on the streaming weekly chart.
  • In March 2026 she became the first foreign female idol to gain full KOMCA membership.
  • She was named 2026 TME mid-year Best Performing SOLO Artist (Korean chart) and one of TME's Women Of The Year In Music.

Asked whether (G)I-DLE is popular in China, it had the receipts too: the Chinese digital album I Feel sold over 500,000 copies in 2023 — the first foreign full album of that year to reach the milestone on Chinese platforms — and “Nxde” has 7.6 million+ views on Bilibili. Perplexity, asked the same solo-music question, conservatively reported no 2026 solo release found. Same question, same day, opposite answers.

So the lesson is not “one engine is bad”. The lesson is that engines disagree, and the language you ask in changes the answer. Qwen's weakest answers came in Korean and English; its strongest — stronger than anyone's — came in Chinese. Your customers are asking in all three languages, on all five engines, every day.

What this means for your brand

Swap Yuqi's name for your client's and nothing about this story changes. AI answers are your reputation now — one confident paragraph, no results page to skim, no notification when an engine hands your biography to a stranger or invents a product you don't sell. The answers vary by engine, by language, and by day. You can't fix what you never see.

If you want to know what the machines are saying about your brand — in every language your customers use — see what PigPR tracks. And for the reasoning behind watching more than one engine in the first place, read why we track multiple AI engines.

Frequently asked questions

What is AI visibility?

AI Visibility is the share of checks whose answers mention your brand. If PigPR asks five AI engines eight questions about you and 36 of the 40 answers mention you — as they did for Yuqi — your AI Visibility is 90%. It's a component of PigPR's Brand Health Score, tracked weekly so you can see it move.

How often do AI engines mix people up?

More often than you'd hope, and with total confidence. In a single batch of 40 checks about Yuqi, one engine answered a question about her with a detailed biography of a different celebrity — and in another answer invented a webtoon that doesn't exist. Nobody notifies you when this happens; the only way to catch it is to ask regularly and read the full answers.

Does PigPR translate non-English answers?

Yes. PigPR stores the full answer text for every engine and every prompt, and any non-English answer gets a translate button. That matters because engines often answer in the language of the prompt — and, as the Yuqi case showed, the same engine can be wrong in one language and the best-informed of all five in another.

Which AI engines does PigPR track?

Five: ChatGPT, Gemini, Grok, Perplexity, and Qwen. Every tracked prompt is checked against all five weekly, with the complete answer text kept for review, because engines disagree — different facts, different sources, and different answers depending on the language you ask in.