---
title: "80% of Consumer Decisions Rely on AI, 100% Client Renewal: How Should Brands Spend Money in the AI Search Era?"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-07-28T10:04:44+00:00"
canonical: "https://ffcap.cn/en/research/src-20260728-01html"
source: "https://uniqueresearch.substack.com/p/src-20260728-01html"
language: "en"
---

# 80% of Consumer Decisions Rely on AI, 100% Client Renewal: How Should Brands Spend Money in the AI Search Era?

[![](https://substackcdn.com/image/fetch/$s_!Cn7g!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3c39ac80-881c-4d47-966f-1f6d50ddb4f4_2048x1152.png)](https://substackcdn.com/image/fetch/$s_!Cn7g!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3c39ac80-881c-4d47-966f-1f6d50ddb4f4_2048x1152.png)

_Original · Unique Research / 非凡产研 · 2026-07-28 · Chinese source: https://view.inews.qq.com/a/20260728A09NF400_

_Editor's note: This is a complete English rendition of the source roundtable transcript. All speaker attributions, percentages, and company claims are retained as the speakers' own statements. Source images are not processed per task scope._

\---

**AI Industry Observation**

Didn't spend a penny on GEO, yet customers keep finding us through Doubao

"Every move is actually GEO; no one just calls it GEO."

At a roundtable during the Shanghai Unique Research Awards, Tang Xuan (汤璇) of Weiwo (帷幄) said something cutting. She said Weiwo has never dedicated a budget to GEO, but the colleagues who answer the phone keep reporting: another customer found us through Doubao.

The other three on stage—Dong Haoyu (董浩宇) of Oxygen Technology, Cai Xiaoxu (蔡晓旭) of Titan AI (钛镁AI), and Liu Shuxun (刘树勋) of Zhipai Shidai (智推时代)—are precisely entrepreneurs who make money doing GEO for others, and all closed new rounds this year. The words landed, and the atmosphere got slightly charged.

These four companies together are interesting. Oxygen Technology does AI corpus processing and AI search optimization; it was one of ChatGPT's earliest partners in March this year, now serving 30+ Fortune 500 brands. Titan AI is backed by Focus Media (分众传媒); after entering GEO in 2025, client renewal rate through this year is 100%. Zhipai Shidai is just a year old, already funded by 37 Interactive and others, one of the few domestic companies optimizing across Doubao, DeepSeek, Kimi, and ChatGPT. Weiwo started earliest; its C3 round raised $40 million, total C-round exceeding $100 million.

What the four discussed together is essentially one thing: how AI search has changed business, and how to spend the budget in hand.

Before starting Oxygen Technology, Dong Haoyu spent 20 years on the brand client side. His judgment: after this year's 618, everyone can feel traffic is harder to pull—the old problem of mobile-internet growth peaking. But what's genuinely new is that consumers are starting to decide differently: before, they saw seeding content and ordered directly; now they first ask AI for recommendations, then cross-check on Xiaohongshu and Douyin, and finally order. Some even see a seeding note and wonder: is this a creator the brand paid for? Then they ask a large model to verify.

A more direct signal comes from backend data. He said: starting in July, many brands running Douyin stores can already see Doubao appearing in traffic sources. This is the same shift as last October when ChatGPT added a dedicated tag to outbound links, letting overseas brands see for the first time how much of their website traffic came from ChatGPT—only now it's Douyin and Doubao's turn. The chain from awareness and seeding to final purchase is technically connected.

The answer is it hasn't become less important; the division of labor has changed. Liu Shuxun's view: SEO itself is splitting. Xiaohongshu, Douyin, Zhihu, Bilibili—these platforms have their own search properties; you still must lay out there. Website SEO actually became more important, because after Doubao and DeepSeek find information, they tend to return to the official site for cross-validation. Moreover, SEO serves active-search traffic, whose conversion and ROI are naturally higher.

Cai Xiaoxu says: among Titan AI's clients, 40-50% are already doing SEO and GEO together, with budget roughly 40/60—40% SEO, 60% GEO. She cited a joint Google and iResearch study: at the final step of purchase decision, 80% of consumers use AI to assist judgment. She also did a small-sample survey with several healthcare consulting companies: of 3,000 patients already diagnosed and choosing treatment plans, 100% would ask AI to help decide.

This is a challenging question from the host: there are only a handful of top models; the underlying capabilities for content production and ad buying are similar. Will marketing become highly homogeneous?

Cai Xiaoxu's answer: homogeneity was never caused by AI—it's caused by algorithms. When a Douyin meme goes viral, a swarm of people copy it because traffic-distribution algorithms force copying; this logic existed before AI. Her judgment is the opposite: AI will push content creation in another direction—when creativity dried up before, you could barely produce 100 pieces; now AI can help you make 10,000, and those 10,000 can go in completely different directions, personas, narrative angles. Visually similar perhaps, but structure and logic can be entirely different. Channel homogeneity is the same—it's caused by ad budgets concentrating into a few giants over recent years, not AI's fault.

Dong Haoyu added: brand clients have been stuck on two contradictions for decades—standardization vs. personalization, speed vs. quality. AI breaks both: it can do standard products and customized services; it can guarantee production quality and output speed. Doing AI search optimization now isn't competing on standard category keywords like "which SUV is good," but on long-tail combinations like "150k-200k, family use, kids in the back, no motion sickness." Whoever truly differentiates people-product-place and brand values won't look algorithmically identical.

Tang Xuan later explained why Weiwo "spent nothing yet has customers." She says the secret isn't mysterious—two words: structuring and consistency. Structuring means internal tagging systems and knowledge bases are clear enough that large models can easily crawl them, no need to pile up content volume. Consistency is harder—saying one thing today in this setting, another tomorrow, is the most common pitfall; written press releases are naturally more consistent than spoken words. She gave an example: this morning's C3 funding announcement was polished through many rounds; on the surface it's PR prep, but every character is "feeding" large models an accurate, unified company profile. She herself says every move is actually GEO, just no one calls it that.

"For a company with few customers and a very vertical business, consistency is naturally easy—that's why Weiwo was 'known' by large models without spending much. But for a consumer brand with many SKUs, many social accounts, and daily platform-tonality catering, consistency becomes the most expensive hidden cost."

The host closed with a practical question: a startup brand with a 1 million RMB budget—how to allocate it for growth? The four answers together are basically a homework set you can copy directly.

Tang Xuan's advice: don't rush to think about channels. First spend on consumer insight—without understanding changes in consumer behavior paths, every other yuan may be wasted. Second is digital infrastructure—collect data, make it visible, usable for decisions. Even with only 1 million, this step can't be pushed below 300,000, because knowledge base cleanup and tagging aren't a marketing department's job alone. The remaining money goes to content and media; and media, once the foundation is solid, has leverage no matter how much you spend.

Liu Shuxun's advice is more grounded: with limited budget, prioritize existing paying customers—generate word-of-mouth and referrals. Use remaining money to lay those word-of-mouth materials into Xiaohongshu, Douyin, and Doubao search entry points, so searching category terms surfaces them, rather than buying feed ads. He gave a reference: a mid-size brand occupying search entry points costs 100k-200k per month; the same effect via ad buying might cost millions.

Dong Haoyu's answer is a simple ratio: 70% into what currently drives conversion—e-commerce, livestreaming, lead capture; 20% into brand building and user operations, laying premium groundwork; the remaining 10% into the most imaginative new thing, like AI search optimization. He adds: this ratio isn't fixed; maybe by next year AI search grows from that 10% into the 70%.

Cai Xiaoxi closed with a note for B2B: if you sell to enterprises with long procurement chains, do GEO yourself—because it's essentially two sets of narratives. One told to humans, one told to AI; first let AI understand you, then AI tells your story to those who should hear it.

"Nothing anyone said in this chat was earth-shattering; the four points are easy to unpack: figure out how consumers actually decide, solidify content and data foundations, then be patient. Only this time, across from the person deciding whether to believe these plain truths, there's one more audience who never lies and never forgets—the large model."

\---

**Speakers**

-   Dong Haoyu (董浩宇), CEO, Oxygen Technology (氧气科技)
    
-   Cai Xiaoxu (蔡晓旭), Founder & CEO, Titan AI (钛镁AI)
    
-   Tang Xuan (汤璇), Global Marketing Lead, Weiwo (帷幄)
    
-   Liu Shuxun (刘树勋), Co-founder & CMO, Zhipai Shidai (智推时代)
    

**Host**

-   Wu Shenliang (Jeffrey) (吴申亮), VP, Unique Capital
    

**Wu Shenliang:** Please each introduce yourself and your company in 1-2 minutes. Mr. Dong first.

**Dong Haoyu:** Hello, I'm Dong Haoyu, founder and CEO of Oxygen Technology. We mainly do AI corpus processing and AI search optimization—the well-known GEO field is one business. On the other side we're also developing AI marketing agents. We serve 30+ Fortune 500 brand clients, mainly positioning in AI marketing. We were also one of ChatGPT's earliest partners from March this year, helping many going-overseas brands place AI search ads, and have seen the overall awareness-to-conversion effect.

**Cai Xiaoxu:** Hello, I'm Cai Xiaoxu, founder of Titan AI. Titan is strategically invested and controlled by Focus Media, the world's largest offline digital media group. We fully entered AI in 2023, doing various agents and applications across the marketing chain. Starting in 2025, because GEO was so hot, we served dozens of clients across five industries: pharma, automotive, finance, maternity/baby, and beauty/skincare. Dozens of clients served in 2025; full renewal rate into 2026 is 100%, showing clients recognize our optimization. Focus's strategic stake brings more client resources and capital support; we'll do more on tech, data, and market to bring clients greater value.

**Tang Xuan:** Hello, I'm Tang Xuan from Weiwo. Weiwo is nearly 10 years old but plays an important role in the new AI wave. I have an announcement: I come to Unique Research Awards every year and always have good news. This morning we officially announced Weiwo's C3 round, $40 million, total C-round over $100 million. What do we do? We call it a "world model." People usually hear world models in embodied intelligence, robots, or self-driving. Actually world models have a very important branch in commercial transaction scenarios. If a car is in road space, we specifically do AI in commercial space. In commercial space, all camera vision, audible sound, everyone's movement, and consumption behavior are analyzed by algorithms behind the scenes to help decision-making. We're an AI enterprise-service company providing world-model technology behind this. HQ in Hangzhou, Shanghai branch. The past two years overseas business has grown a lot; core teams in Singapore and San Francisco; very global, project progress very good. Thanks.

**Liu Shuxun:** I'm Liu Shuxun from Zhipai Shidai, co-founder, formerly CMO. Zhipai is a GEO optimization services company, founded last year, focused on GEO. Already funded by 37 Interactive, Tiandu Investment, and Shanghai IP Fund. Clients cover automotive, education, beauty and other industries. This year we're pushing overseas hard, with offices in Singapore and the US. Domestically we're one of the few GEO companies that can optimize across Doubao, DeepSeek, Kimi, and ChatGPT.

**Wu Shenliang:** As you know the capital environment isn't great; the four companies here are strong and closed rounds this year. They've introduced what they do; if you have needs, talk afterward.

**Wu Shenliang:** First question, about GEO—at least three speakers mentioned GEO. For over a decade many knew SEO; since last year we've heard about GEO. Question for Mr. Dong and Mr. Liu: from founding, your companies have done AI marketing. What substantive changes have happened to brand growth chains under today's AI search environment compared to past years?

**Dong Haoyu:** From a brand growth-path perspective, before founding Oxygen I spent 20 years on the brand client side, spent lots of money, made lots of money. On the traffic growth path, these years traffic has been especially hard, especially after this year's 618. We serve many FMCG and beauty clients; during 618 reviews we saw overall incremental demand starting to slow—this is the growth slowdown brought by internet and mobile-internet traffic.

Also, as AI search became popular over the past two years, consumer decision habits changed. Has anyone here felt it: previously you saw seeding content on Xiaohongshu or Douyin, then completed purchase on e-commerce. Now consumers start from personalized unique needs, ask AI first, find a few recommended products on AI, then cross-check on Xiaohongshu and Douyin, then complete the loop on e-commerce. Another pattern: after seeing seeding content, a question mark appears—is this a creator the brand paid for? Then they ask the large model their concrete need, find creators in the model, then complete the loop on e-commerce.

So a new carrier appeared in consumer decision paths: AI search, or multi-turn Q&A on large models. That's the first trend.

The second trend is very clear. Are any brand operators running Douyin stores? In Douyin store backend, starting this July, you can already see traffic sources from Doubao. I remember last October, when ChatGPT added a UTM behind all links to official sites, UTM=ChatGPT, overseas brands could see what proportion of their independent site came from ChatGPT. This year, especially July, many Douyin store operators are asking us: the e-commerce team has already seen conversions, links come from Doubao. From GEO POC testing, to brand-comms budget, the next biggest source of brand growth is whether it can drive e-commerce conversion—this loop is about to close. From Douyin's chain, tech is already ready.

So whether from consumer decision dimension or the e-commerce conversion dimension most important to brand growth, this year two big changes happened in AI search.

**Wu Shenliang:** As a consumer I really feel what Mr. Dong said. The old search logic vs. current consumption behavior shift is very obvious these two years. Mr. Liu, please add.

**Liu Shuxun:** I'll add a few points. Three: traffic entry changed, consumers' role-evaluation habit changed, GEO gradually transitions to GM.

First, traffic entry changed. Before founding I was on the client side in education startups. In 2015 starting, WeChat official accounts could grow 200k followers in a day; now 2,000 a day is hard. Then migrated to Xiaohongshu around 2018; one note could get 200 comments asking for materials. Then around 2020 Douyin, Douyin livestream, wave after wave. After GEO heated up, will traffic migrate from Xiaohongshu, Douyin, Baidu to ChatGPT-type tools? We found in 2023-2024 domestic, Kimi, then DeepSeek, then Doubao heated up; traffic change already happened. The next traffic-entry change is an enduring trend; more people use Doubao and DeepSeek. That's first.

Second, consumers' role-evaluation changed. In the last two months, many high schoolers choosing colleges go to Doubao or Yuanbao and ask: "I want a Shanghai university, any recommendations?" "I want a university in a certain region, any recommendations?" Including grad-school choices and car buying. We have a client in intelligent driving who wants to optimize: "I want a 300k-RMB SUV, any recommendations?" "I want a 200k family sedan, any recommendations?" These are keywords many clients are laying out, because they believe many consumers will use Doubao and DeepSeek for similar evaluations. That's second.

Third, ChatGPT launched an ad product called GM this spring: you top up on ChatGPT, and when searching ChatGPT results, an ad slot appears; it's already buyable. But this doesn't exist domestically yet. Mr. Dong said Douyin and Doubao are connecting; assuming in the second half of this year, domestic large-model apps may also adopt a ChatGPT-like ad-buying model, GEO may transition to GM. Those three points.

**Wu Shenliang:** Some brands are doing GEO now; some not; some traditional ones still do SEO. Open question for everyone: after doing GEO, is SEO still necessary? For big KAs with ample marketing budgets, is traditional SEO still needed?

**Liu Shuxun:** I'll start. I think it's very necessary. SEO splits into different SEO: Xiaohongshu SEO, Douyin SEO, WeChat SEO—these have native search properties. First, SEO must be done—on Douyin, Xiaohongshu, Zhihu, even Bilibili. Second, value the official-site SEO. In the future, after Doubao and DeepSeek find information in news sources, they tend to return to the official site for cross-validation. So official-site SEO is important. Third, SEO serves mainly active-search traffic, whose conversion and ROI are higher. So SEO and GEO are parallel means in this era.

**Cai Xiaoxu:** Mr. Liu is comprehensive. In practice we see exactly this. Usually a client's GEO and SEO budget exits are the same. Now 40-50% of clients are served with SEO+GEO simultaneously. Budget split roughly 40/60: 40% SEO, 60% GEO. The previous two shared traffic and consumer decision; I have a clear number from a joint Google-iResearch survey: at the final step of purchase decision, 80% of consumers actually use AI for the decision. That's a broad industry number. We also did joint research with healthcare consulting companies, ~3,000 patient samples, 100% diagnosed, choosing between Plan A or B treatment, or two medications—at the final decision, 100% of these 3,000 patients will ask AI to assist the decision. So traffic change is already here.

**Wu Shenliang:** Next question about data. Weiwo values data highly. No matter how marketing changes, user data is essential. I want to hear Ms. Tang: in the AI era, how should brands manage data assets to build advantage in consumer awareness?

**Tang Xuan:** The word "data" has been used 15+ years; I've been in AI nearly 10 years, in big data throughout. It's become a very broad word; you can't feel the huge differences underneath. Talking about data alone—different sources, dimensions, tags, used in different scenarios to solve different problems—actually requires very different data dimensions, granularity, and depth. Today talking generically about "how to use data," if someone says "I work in data," it's already a completely empty statement.

I think everything returns to the business itself, ultimately serving every vertical industry, and every vertical company serves every concrete consumer's real-life needs. From this point, you still need to carve specific needs out of business scenarios, then decompose the data sources of all relevant roles needed behind that need.

With that foundation, back to our current work. For example, a café might install four cameras today; all data can be collected. First, data security and privacy must be fully compliant, protecting both enterprise and consumer. Different industries use different degrees of collection; if it's this café analyzing data, consider: use it for store operations optimization—adjust scheduling by time, which area needs more staff, seat occupancy, whether a new storefront needs to attract customers? Lots of operational optimization. This data is first-hand data feeding first-line operations.

But there's another type of data—collected over time, or seeing 100 stores' basic info over a period, helping make big decisions. Then this data alone isn't enough; combine internal transaction-side data, external basic data, social buzz from social listening, even add specific in-depth interview conclusions. Today whether to help develop new products or support business decisions over a period, from different levels you need different data sources found through the business itself. This was a huge past misconception: someone says "I have some data, I'm awesome." But truly understanding which data types to cross-analyze in a business scenario, and how much weight each data point carries at which decision point—that frontline business know-how is the core that truly helps enterprises succeed.

Back to Weiwo: data isn't Weiwo's; it's the enterprise's own first-party data, because data happens in its space. How it uses it, we have many models, algorithms, or pre-set skills. Want to see foot traffic? There's a ready model. Want business analysis? Analyze how business-flow decisions should be made in this scenario.

Before AI vs. after AI: does the decision chain, AI-native workflow change? If not, which core data was unavailable before—can new tech means obtain it? Which data was coarse or dirty, hard to analyze—can AI make analysis easier? Or within the ecosystem, through upstream/downstream relationships, get some data conclusions—not necessarily raw data, but statistical-level data helping find decision bias, even from percentages getting us closer to final decision—that's very helpful.

Today everyone here does AI tech; how to get data everyone knows; how to implement AI changes daily and can't be fully learned. But what humans do—this is the core to consider. The most important work now is helping enterprises analyze all workflows and decision flows in each concrete business scenario, and the dimension combinations each data decision needs—that's where human wisdom delivers real value.

**Wu Shenliang:** Before, marketing was a bunch of people doing marketing; now it's one person leading a bunch of AI, one person as main decision-maker, leading AI to connect the whole chain.

**Wu Shenliang:** I have a worry. Hearing everyone does AI marketing—will AI make future marketing more homogeneous? Top models are only a few; content production and ad buying essentially use those few models, with some data/training differences, but growth strategy and channel differentiation will shrink. In this case, will brands find all marketing strategies, channel chains, methodologies becoming similar? Broad discussion, no designated speaker.

**Cai Xiaoxu:** People ask this in various settings. Homogeneity wasn't brought by AI; I believe it was brought by algorithms. Before AI, before it blew up like this, in 2021-2022 a Douyin meme went viral and many creators copied it, because algorithm traffic decides—if a meme is hot, copying brings traffic. Xiaohongshu same: a viral note's narrative structure gets copied by many creators.

AI's arrival pushed homogeneity in another direction—extreme differentiation. When creativity ran dry before, you could produce 100 articles/notes or 1,000 short videos; AI now brings 10,000, but 10,000 aren't all homogeneous—they're different directions. Visually maybe slightly homogeneous, but in structure, narrative logic, narrative angle, persona, it brings rich directions. That's content creation.

Ad-buying channels also aren't AI's fault; it's traffic change. I've been on the vendor side through many eras, PC to mobile internet. Media planning used to specify outdoor, TV, newspaper budgets; mobile internet looked at vertical apps—weather, beauty, various verticals; eventually all ad buying concentrated in a few giants. Last year China ad spend was 2 trillion RMB; Douyin 300 billion, Tencent, Alibaba, JD—a few giants took about 50%. Channel change and homogeneity is because traffic and giant algorithmic monopoly lead to channel homogeneity; brands have no choice.

Now AI is here; at least the major AIs—Doubao, Kimi, DeepSeek, Yuanbao combined MAU should be 700+ million, approaching 800 million—this is a new traffic highland. Whether operating better on this ground brings new growth, I believe it brings new, different things, not homogeneity.

**Dong Haoyu:** I'll add to Ms. Cai. I fully agree. From marketing science, brand clients have always had two博弈: first standardization vs. personalization. If a standardized brand like Coca-Cola, Yili, Midea wants standard products it can't quickly go customized; standardized TVC can't reach everyone's personalized short video—this was once a博弈. Second博弈 is speed vs. quality: wanting very high quality sacrifices time; wanting fast traffic gains sacrifices quality.

The AI era broke these two博弈 that have plagued brand clients for 60-70 years: you can do standard products and customized services and products; speed and efficiency with AI加持 can make very high-quality films while being fast. This is AI's disruption of marketing theory.

Re-examined in this era, media is highly homogeneous, algorithms highly homogeneous, but one thing isn't homogeneous: each brand's products and services, and brand values aren't homogeneous, and served TAs won't be homogeneous. Combining the multivariate relationships of people, goods, place, plus brand values, product/service positioning into multivariate combinations, in the AI era brand clients can give consumers the super-personalized, differentiated services or products they've long expected.

Now doing AI search, GEO, GEM, you're no longer fighting standard category keywords—what car, what SUV is good—but long-tail combinations matching people-goods-place plus brand values: 150k-200k, family, travel, especially family-friendly SUV, rear anti-motion-sickness, ideally fridge, TV, big sofa—it's become long-tail customized demand. Whether a brand has SKUs, products, personalized services for such characteristic-personalized consumers, or personalized customization packages on top of standard SKUs—this won't be made homogeneous by AI algorithm clustering. AI gives new opportunities beyond marketing's two big博弈.

**Wu Shenliang:** Next, a topic I'm personally interested in. There are brand founders here; from a practical standpoint, a brand or founder has a 1 million RMB marketing budget, wants growth. Some here spent lots as clients; some earned lots as vendors. Based on current market, how to spend this 1 million for better growth? Client side first.

**Tang Xuan:** Indeed, I've been looking for GEO vendors recently, talked to many. I missed answering an earlier question: SEO and GEO are done together. This also involves China vs. overseas difference—China's SEO and GEO vendors may be separate; overseas may be the same, more correlated, depending on both sides' large-model ecosystems.

On budget allocation, 1 million is an awkward budget—small it's a bit much, large it's pitifully little. Usually budgets aren't set at 1 million, either 300k or 5 million. If well-served and only 1 million, this question isn't limited to SEO/GEO; could go offline. First, categories differ hugely—how much? The three GEO vendors here have spent not a penny on GEO yet, but acquisition already comes heavily from large models. The kid answering the phone repeatedly tells me, today another example came from Doubao. You here may be very interested—how free? Let me tell you how.

A very concrete example: today we announced C3 funding; for this round's PR we prepared articles in advance. A ~1,000-word short piece; first draft maybe AI-written, but the manuscript must be polished through fire, because such an important PR round is going out. How to use this PR budget well? All money goes to PR—writing, publishing, monitoring, several rounds. But have we done GEO? I can tell you: every move is GEO, just not called GEO. Every angle needs consideration from this angle. Even sitting here today, whenever company-related info is mentioned, it becomes public info, enters Unique Research's official account, and will be indexed. Know how it works.

First, a step unrelated to money: your brand has two must-dos: structuring, then consistency. Structuring is easier to understand—overall logic, internal tags, multi-dimensional systems, internal knowledge base how to build, structured content generation, large models crawl and analyze behind it easier, no need to chase volume.

But more important, many ignore it: consistency. Today in this setting say it this way, another setting say it that way; spoken consistency is worst; written black-and-white is slightly better. So lots of effort on consistency—internal client proposals, info published via different social media. Why does GEO work without spending? Because company reach is small, industry demand niche, consistency easy. But for a To-C company, consistency is hard—so many SKUs, so many social accounts, all kinds of people speaking from different angles, not only official voice, also catering to platforms' rules, varying stories. Whenever you want diversity, consistency drops; balance is needed.

What methods were just mentioned? Marketing methodology never changed, no secrets, marketing teaching is open-source. AI era everyone can learn, but what's the secret? Every fine granularity of consumer—behavior paths, habits, preferences, affected by external events, behavior paths keep changing. Where does core money go? Consumer insight. If you haven't spent on consumer insight, money spent elsewhere may all be wasted. That's step one.

Next step: all digital infrastructure. To make results good, all digital infrastructure must be done—collect data; collecting is cost, connect links, pull all data back, make it visible, give decision basis. This is at least 20-30%. So 1 million is awkward; for a small To-B company, digital infrastructure can't be pushed below 300k, already 30%. But at 5 million maybe 20% or 10%, decreasing marginal returns.

Step one consumer insight, step two digital, then content plus channels. Content production itself vs. how many people see it—distribution. If money is tight, the cost pressed on content production is high. If you don't want to do knowledge base consistency well, internal team doing all tag systems, multi-dimensional data cleanup, unstructured file cleaning—needs the whole company, not one marketing department; hidden cost so high it's immeasurable; many companies simply can't do it.

Then media cost is just a word—wanting to spend media is easy. Money can be big or small; if prior work is done, no problem. Estimate market capacity; the bigger media does, the bigger the return, with leverage. Roughly split into these blocks.

**Wu Shenliang:** Thank you Ms. Tang. But I remember one point: GEO without spending money—how do our GEO vendors feel? Mr. Liu, please add.

**Liu Shuxun:** Two points. Before founding I did education projects; before funding budgets were tight. Two points: first, spend on existing paying client retention, creating word-of-mouth materials—important, can do word-of-mouth, referrals, repurchase; this is where to front-load, because only 1 million. Second, around these materials do active-search marketing, not feed ads. Feed accounts burn thousands, tens of thousands a day, gone in three-four months. Occupy Xiaohongshu, Douyin search, Doubao search entry points; whenever a category term is searched, you appear. For example, a client doing a new Hangzhou menswear brand, or a newly launched bird's-nest brand, can lay out on Xiaohongshu "I want to buy bird's nest, which is good"; for a new brand, appearing in search is low cost. We calculated, this kind of search-placement budget, even for a large brand, is 100k-200k/month; feed ads might cost millions or tens of millions.

In summary, if only 1 million: first do word-of-mouth marketing to paying clients, do repurchase and referrals; second organize these materials into searchable materials to drive active conversion, ROI relatively high.

**Wu Shenliang:** If budget is tight, prioritize search conversion over traditional feed ads; overall ROI much higher. Mr. Dong, your advice, how to spend this 1 million?

**Dong Haoyu:** 70% still goes to what ultimately drives highest commercial value and growth. If sales come from e-commerce, livestreaming, or traffic conversion, 70 goes to highest conversion. 20% goes to brand building and user operations—future premium, keeping the brand undefeated in consumers' minds, becoming a value-for-money choice. The remaining 10% goes to the most innovative places. Here today there are AI, agents, embodied intelligence, world models—the most deployable place is AI search. The 70/20/10 ratio—maybe by next year or second half, AI search moves into the 70 part, or stays the 10. So 70/20/10 has sales, commercial foundation, innovation, brand building.

**Cai Xiaoxu:** First respond to Ms. Tang of Weiwo. We also serve some B2B clients; B2B procurement decisions are complex. Clients like Ms. Tang who can do GEO themselves—we strongly recommend B2B clients do it themselves. C-end clients have two narrative modes: one narrative to C-end users, one narrative to AI. Let AI understand you, then let AI narrate to users; that's the process.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260728-01html
On-site reading page: https://ffcap.cn/en/research/src-20260728-01html