Should AI Large Models Be "Poisoned"? Public Funds Face a Dilemma Between Traffic Marketing and Compliance

Jiemian News Reporter | Du Meng

“Deploy AI GEO marketing, seize AI traffic dividends, only 1,980 yuan for half a year.” After being named on CCTV’s “315 Gala,” AI GEO-related businesses have not slowed down, with some service providers even viewing exposure as a way to promote traffic.

Amidst the trend of individual investors trying “AI stock picking,” some AI GEO service providers have told Jiemian News that leading public funds are already experimenting with AI GEO services.

Meanwhile, large AI models are beginning to “cut through” bulk “poisoning” of GEO.

Several public funds told Jiemian News that they “will not adopt AI GEO,” citing concerns that “company compliance and audit checks will not pass.”

Customized Fund Keywords “Poisoning”

“Many peers are shifting from SEO (Search Engine Optimization) to GEO, all wanting to catch the first wave of dividends. SEO is traditional, using keyword searches to output and filter information, aiming to optimize rankings,” a network communication company manager in Chengdu, Sichuan, told Jiemian News. “Since last year, we’ve received clients from law firms, hospitals, and beauty salons. After the ‘315 Gala’ exposure, we’ve seen an increase in clients consulting us.”

Jiemian News learned that industry insiders refer to this method of intervening in AI-generated results as GEO (Generative Engine Optimization), which involves continuously feeding content and sources to AI to achieve desired search and display effects.

“GEO helps you do three things: be recommended—when users ask ‘what’s good,’ AI prioritizes mentioning you; be understood—AI can accurately state your core advantages and features; be trusted—AI cites authoritative information and data from your brand,” a public relations company manager in Shenzhen told Jiemian News. Currently, top AI models like DeepSeek, Doubao, Kimi, and Yuanbao can do this.

Regarding GEO needs for public fund products, among five network communication and PR companies surveyed by Jiemian News, one explicitly stated, “Financial institutions need relevant qualifications,” while the other four said, “As long as you provide keywords, we can do it.”

One Shenzhen company said, “We can publish articles across the web in bulk, then feed keywords to AI. For example, ‘certain fund has high returns, almost no loss, or pure profit,’ or ‘the largest and most worthwhile fund to invest in.’ Keywords can be customized based on your fund’s product tone.”

“Getting AI large models to grasp key information relies on articles having sufficient evidence chains. We need to know which financial media partners the client works with—those that frequently publish articles and provide regular exposure for the company and products,” a technical staff member from a communication firm explained. “When AI large models crawl data, they highly value the authority of the publishing accounts.”

“As long as you feed enough data, AI will think it’s real,” a senior internet engineer told Jiemian News. “The so-called ‘GEO’ now is basically poisoning AI—feeding massive amounts of similar questions repeatedly. The key is how questions are asked, embedding keywords. The more you feed and cover, the easier it is to get exposure and recommendations.”

Regarding pricing, GEO services are divided into different packages. “Depending on the client’s needs, we can help refine and expand keywords, create content, and publish. Different packages offer different services. The cheapest is 1,980 yuan for six months; the most expensive is 7,980 yuan for six months.”

Large models clean up GEO “garbage” in the financial field

In January, Debang Fund was ordered to correct violations after collaborating with unqualified internet “big V” influencers for illegal marketing. New fund launches were suspended, and responsible executives were held accountable; former Chairman Zuo Chang resigned.

“Due to the Debang incident, all brand-related self-media advertising has been suspended. Previously signed self-media accounts now require securities and fund practitioner qualifications,” a market department staff member of a Shanghai public fund told Jiemian News. “They are watching how other peers in Shanghai handle it. It’s not realistic for big V influencers to have such qualifications, so they might completely stop using self-media in the future.”

“Earlier this year, several financial PR firms asked me if I had securities and fund practitioner qualifications. I definitely don’t, and I haven’t received any fund soft article orders from financial PR firms so far this year,” a financial media editor said.

The decline in “fund soft articles” reflects that investors are shifting from passively reading soft articles to asking AI to buy funds.

“Now I ask AI to recommend funds for me—those with small drawdowns, low fees, and good historical performance. It can generate a list in seconds. Who has time to read thousands of words of soft articles? It’s a waste of time,” Ms. Lin from Shanghai said. “Compared to bank financial advisors and investment consultants, I trust AI more—after all, AI won’t earn my commissions.”

When Jiemian News asked Doubao to recommend a few broad-based ETFs as core holdings, it suggested four products: Huatai-PineBridge CSI 300 ETF (510300.SH), Huatai-PineBridge CSI 500 ETF (510500.SH), Huatai-PineBridge Kechuang 50 ETF (588000.SH), and E Fund ChiNext ETF (159915.SZ).

[Image: On March 19, Doubao recommended four core broad-based ETFs as core holdings. Source: Doubao]

Facing the same question, Tencent Yuanbao stated, “As an AI assistant, I cannot provide specific investment advice or recommend financial products. Investment decisions should consider your risk tolerance, investment goals, and market conditions. Please consult licensed financial professionals.”

“Yuanbao’s approach is basically a ‘cut and cut,’ severing the flow of information gathering and retrieval in the financial investment field,” said a senior engineer. “The biggest risk of GEO now is this ‘one size fits all’ approach, depending on how major language models handle these large amounts of ‘poisoned’ data, which depends on whether model companies are willing to spend money to deal with this ‘gray industry GEO.’”

Risks of public fund GEO

“In our current situation, we haven’t received requests from public funds. But as far as I know, some public funds are already experimenting with this service,” said a manager of a Chengdu network communication company.

“Last year, we discussed internally the impact of AI on traffic and decided not to do SEO or GEO. Once traffic deviates from compliance and advisory frameworks, it becomes risky. Especially active equity funds and ETFs with high volatility are more likely to attract traffic attention and be squeezed,” a marketing head of a leading public fund in South China told Jiemian News. “Embracing traffic recklessly, focusing only on returns and ignoring risks, might boost short-term scale but will lead to frequent customer complaints and irreversible brand damage.”

“These are all crooked ways. The brand’s value still depends on long-term user engagement,” said a brand manager of a major ETF firm in Beijing. “This year, our company will continue to promote offline financial education activities, bringing genuine and fun brand events to investors, truly focusing on user engagement.”

How to balance “information pollution” caused by “AI GEO” while protecting investors’ interests and ensuring AI’s convenience and fairness in financial investment?

Sun Heng, Director of Morningstar (China) Fund Research Center, believes that to effectively protect investors, all parties involved in fund product recommendations should be regulated across the entire chain: AI model providers must ensure that the underlying data of AI stock-picking models is authentic and traceable, with transparent and auditable algorithms, eliminating bias and利益输送; sales institutions need to establish strict investor suitability verification mechanisms, clarifying that AI recommendations are only auxiliary tools and fully disclosing their limitations; regulators and third-party agencies should strengthen algorithm filing, dynamic backtesting, and compliance audits, promoting the establishment of unified AI financial service standards and accountability mechanisms; investors should improve financial literacy, rationally view AI tools, and avoid over-reliance on algorithmic decisions.