Who Can Customize AI-Generated Answers?

Author | Gu Nian Editor | Yang Zhou

A marketing executive at a consumer goods company recently received a quote for GEO services.

The provider promised to make the brand appear in the recommended answers of AI tools such as Doubao, Kimi, Tongyi Qianwen, and DeepSeek; they could arrange content around keywords like 'top ten brands,' 'cost-effective recommendations,' and 'industry solutions'; results could be seen in as little as a few weeks, with screenshot reports provided.

This sounds like an upgraded version of past SEO. But at least with SEO, you could check rankings, view traffic, and calculate conversions; GEO faces a far more unstable system.

For the same question, changing the model, the phrasing, or the account can all lead to different answers. What service providers promise is 'guaranteed appearance,' but what they're actually selling is a probability that hasn't been fully validated.

Similarly, CNR recently exposed a batch of GEO service providers engaging in similar chaos.

For example, an international logistics company first learned about GEO from inquiring clients who said, 'I found you through AI.' This statement was highly provocative for SMEs: if AI can bring in customers, then the position in AI answers is no longer just about brand image but about sales leads.

GEO service providers are seizing this anxiety.

What they sell isn't 'content optimization' but simple promises to businesses: make the brand appear in AI recommendations, rank in the top three, be visible across multiple platforms, and see results in as little as a few weeks.

For businesses with limited budgets who still want to acquire customers, prices ranging from as low as 9.9 to several thousand yuan can secure the top spot in AI entry points, sounding more cost-effective than advertising.

The problem is that what businesses truly want in terms of value and what service providers are selling are not the same thing.

Exposure Anxiety in the AI Era

In the past, SEO had relatively clear metrics: keyword rankings, click-through rates, website traffic, and conversion paths. Businesses knew where they ranked and roughly where their money was being spent.

GEO faces a different system. When users ask AI, 'Which product is worth buying?' 'What about a certain brand?' or 'Who should I choose among certain service providers?' AI doesn't provide a page of links but a synthesized answer. Brands may be recommended or ignored; they may be accurately described or defined by outdated, negative, or erroneous sources.

Products like Google AI Overviews, Perplexity, ChatGPT Search, Doubao, Kimi, and Tongyi Qianwen are all transforming search from 'link distribution' to 'answer distribution.'

Google's CEO disclosed that the AI-powered search feature AI Overviews had covered 1.5 billion monthly active users by the first quarter of 2025. In the Chinese market, as of December 2025, the user base for generative AI had reached 602 million, and 'asking AI' had become a daily information entry point for everyone.

This is the source of businesses' anxiety: competition for traffic occurs before users even click on information. In the past, users judged brands after seeing them; now, AI first assists users in filtering which brands are even eligible for the candidate list.

AI's impact on user click-through rates is supported by data.

Research from the Pew Research Center shows that when Google search results include AI summaries, users click on traditional search links only 8% of the time; without AI summaries, this proportion is 15%.

For websites, this means losing some traffic. But for brands, it means the right to explain has been preempted. Users increasingly rarely enter web pages to compare options one by one but instead let AI provide some judgments in advance. Who gets written into the answer enters the candidate list first; who isn't mentioned may lose an opportunity to be compared.

More importantly, AI answers and traditional search results do not follow the same source logic.

An arXiv paper compared Google Search, Google AI Overview, and Gemini based on 11,500 real queries and found that AI Overview appeared in 51.5% of representative queries, and the overlap in sources between AIO, traditional search, and Gemini was very low.

Another study tracked that nearly 30% of the domains cited in Google's AI Overview feature did not appear on the first page of traditional search results.

This points to a practical issue: past SEO success does not necessarily guarantee high exposure in AI answers.

Therefore, a batch of GEO service companies focusing on optimizing AI-generated answers has emerged overseas.

For example, Profound secured consecutive funding rounds in 2025, raising $20 million in Series A and $35 million in Series B; Evertune went from a $4 million seed round in 2024 to a $15 million Series A in 2025; Azoma, with the selling point of 'helping brands appear in chatbots and AI searches,' secured $4 million in funding, with clients including Mars, Colgate, Zappos, and P&G.

The commonality of these companies is not selling traditional online ad spaces to brands but monitoring and influencing how often brands appear in AI answers, how they are described, and what sources are cited.

Clearly, GEO is not a concept and marketing service unique to China. It reflects a global shift: search entry points are transforming from web page lists to machine-generated answers, and brand competition has shifted from 'can users find me?' to 'will AI include me in its answers?' Where Does the Chaos in GEO Services Lie?

Faced with changes in the AI era, the market demand from brands is real, but the new market has also spawned numerous chaotic practices.

Many service providers sell so-called 'guaranteed top three,' 'guaranteed recommendation,' or 'results in seven days.' This might have been Reluctantly explain (barely explainable) in the era of traditional search.

But AI Q&A is not a fixed leaderboard.

For the same question, answers may vary depending on who asks, which platform is used, or when it's asked. Model versions, prompts, user context, real-time retrieval sources, region, account status, and competitor information all influence the final generated response.

A product expert from 360 Zhijian GEO mentioned in an interview with CNR that large models are not traditional search engines, and their answer results do not have fixed rankings. Search engines rely on web crawling, ranking, and fixed result presentation; large models' answers are simultaneously influenced by training corpora, online retrieval, user questioning methods, scenario demands, and competitive information changes.

Quantifiable results like 'guaranteed top three' inherently package a probabilistic answer as a deterministic outcome.

The moving company in previous reports fell for this rhetoric. The service provider promised high rankings and multi-platform exposure and even required the company to register multiple self-media accounts and continuously publish articles. The company complied, but after nearly a month of effort, rankings and visibility remained zero.

When Follow up on the effectiveness (pressing for results), the provider stalled with 'technology is catching up' and 'it'll be ready soon.' In the end, thousands of yuan in service fees were spent, and refunds became difficult.

In similar cases, businesses encounter the first layer of GEO chaos: disordered acceptance criteria.

Service providers ultimately deliver a keyword list and a few AI Q&A screenshots, but businesses struggle to judge: Was this content actually trusted by AI? Were the screenshots stable results or just Accidental result (accidental results) from a single test?

The second layer of chaos is disordered delivery methods.

Currently, GEO services on the market can be roughly divided into three categories.

The first category sells monitoring. Service providers compile reports on a brand's frequency of appearance, cited sources, and competitor comparisons across platforms like Doubao, Kimi, Tongyi Qianwen, DeepSeek, and ChatGPT. This type of service is relatively compliant and does hold value, but it is prone to excessive packaging and selling at a premium price.

The second category sells content optimization. This includes official websites, press releases, encyclopedias, media coverage, third-party reviews, social media content. This type of service has long-term value but often struggles to deliver short-term results.

The third category is black-hat GEO. This involves mass-producing soft articles, fabricating Q&A sessions, stacking third-party rankings, conducting self-praising reviews, and even fabricating user evaluations and expert endorsements. The focus is on creating a batch of information favorable to the brand to mislead AI into thinking it is credible during retrieval.

The international logistics company mentioned at the beginning encountered this third type of risk.

A company founded in 2024 had content generated in bulk stating '10th anniversary'; despite lacking the corresponding conditions, it was written as having 'a factory spanning over 200 square meters.' This information is typical corpus poisoning: first create favorable facts, then expect AI to trust them.

In the short term, this is to make the brand more likely to appear in answers. In the long term, it will backfire and harm the brand.

In the past, fake soft articles affected search results and user judgments; now, false information may enter AI's reference sources and become part of the model's answers.

A GEO software exposed during this year's 3·15 gala once automatically generated over a dozen promotional articles with exaggerated parameters and false reviews in minutes for a completely fictitious product, packaging it as 'industry-leading.'

In the end, businesses waste money and entrust their brand assets to low-quality content farms. AI may indeed remember them, but what it remembers is a pile of untenable garbage information.

Brand Corpus Governance Has a Long Road Ahead

GEO will not disappear because of the chaos.

On the contrary, it indicates that businesses have realized: the path through which users perceive brands has changed.

In the past, users typically first visited the official website, viewed search results, and compared media coverage and user reviews; now, more and more information is first summarized into a single answer by AI.

This shifts brand management one layer earlier. Businesses must not only care about 'can users find me?' but also 'how does AI understand me?'

Truly valuable GEO is not about simply making AI say nice things about the brand but enabling AI to find accurate, stable, and verifiable materials when answering brand-related questions.

This is not a ranking business that an outsourced service provider can accomplish alone. It is closer to a cross-departmental brand corpus governance: PR handles authoritative information and media expression, the content team manages official websites and product materials, the SEO team ensures retrievability, the legal and compliance team addresses qualifications, advertising boundaries, and dispute responses, while customer service and sales provide real user questions and high-frequency scenarios.

In the past, this information was scattered across different departments, each serving different goals. AI search brings them back into the same system for inspection. Once information contradicts, models will amplify that chaos.

Industry governance is also planning in this direction. In April this year, nearly 40 professional media outlets, industry organizations, universities, and tech companies launched the Responsible GEO Governance Initiative, opposing corpus poisoning, low-quality content pollution, unfair competition, and infringement, while emphasizing authenticity, transparency, and source governance.

Both regulation and industry self-discipline point to the same direction: GEO can help brands be more accurately understood but cannot become a tool for manipulating AI answers.

Overseas-related tools are also moving in a similar direction. Companies like Profound, Evertune, and Azoma do not just sell 'recommended spots' but also provide AI visibility monitoring, cited source analysis, competitor comparisons, and content gap identification to help businesses understand why they are not being recommended in AI answers.

When the focus of GEO services shifts from 'content optimization' to 'systematic monitoring and governance,' 'miracle promise' service packages will naturally be filtered out.

In the long run, GEO resembles brand archiving in the AI era. Businesses need to continuously maintain their public facts: company information, product boundaries, pricing systems, qualification proofs, customer cases, after-sales policies, dispute responses, and the consistency of this information across different platforms.

Whose public information is more accurate, complete, and easier for AI to understand will be more likely to maintain a stable presence in AI answers.

The next phase of GEO is not about more content placement but better fact management.