“SEO’s New Overachiever” — Pollo AI Founder 阿彪 on SEO, Influencer Marketing, and Paid Ads
Summary
- Pollo AI reached roughly 4M monthly visits within months, while its company grew to around 90 people and eight-figure ARR; the growth engine was not an SEO trick, but the compound effect of entering early and applying a three-pronged playbook of SEO, influencers, and Google/Facebook ads. 阿彪 says outsiders only see the “tenth bun” （“第十个包子”）that finally fills you up; the first 9 came from years of going global and building multiple products. Internally, there was no single breakout event—traffic simply climbed steadily. He estimates marketing “should still account for half” of the result, while stressing that timing mattered and the product experience was decent too.
- Pollo’s SEO methodology starts by building pages at scale that can capture search intent, then using product experience, content, and authority to push them up the rankings. The pages fall into 4 categories: core features, utilities, models, and articles. “100 pages, 10 visits per page” creates 1,000 visits a day; pages for models such as 可灵, Runway, and Sora can also capture branded searches and multilingual queries the model providers have not covered. The key detail is that landing pages work without login: before login, they retain descriptions, examples, and tutorials for Google to understand; after login, that material disappears and the user enters an immersive workspace.
- SEO is neither free traffic nor an instant cure-all; the normal feedback cycle is 3 to 6 months. Articles, feature pages, PR, and backlinks all cost money, and any weak link—from content and site structure to URLs and keyword difficulty—can drag down the result. Backlinks should mimic the natural growth of a website: do not start with only expensive links or suddenly spend $100,000, and keep the cadence smooth. 阿彪’s one-line definition of deep SEO is “satisfy user intent,” but he is clear that SEO alone cannot support $10M-plus revenue, let alone more.
- The real constraint in influencer marketing is not how many KOLs you can find, but whether the product has a “hook” worth liking, debating, and sharing. As a wrapper around aggregated models, Pollo lacked the proprietary-model halo and had to push hard in the early days. It later promoted fun effects, but users often “liked it, then left—they didn’t pay.” 阿彪 still admits that ROI is “very, very difficult to measure”; for now, the team uses CPM, views, and budget limits to manage campaigns, while trying to strengthen the “one-stop everything bundle” pitch.
- Timing, differentiation, and the variable that determines success must be assessed at the product-selection stage; otherwise, harder marketing may simply mean a more expensive education. 阿彪 spent more than half a year on an e-signature product and lost several million yuan, only realizing afterward that the category depended on brand, trust, and compliance, while startups faced mature rivals such as DocuSign and Adobe Sign. Online PDF editing likewise lacked both timing and a differentiated entry point. AI music failed because the MiniMax and 天工 APIs were expensive, users bought only the cheapest tier, and Suno was “better and cheaper than us”; what remained was “fishing for dead fish.”
- 阿彪 does not deny that Pollo currently lacks a hard moat; the practical response is to move faster on the product, make it painful for followers to catch up, and turn the marketing dial “up a little more.” He also does not think a multi-product strategy is inherently wrong: he prefers going from 0 to 1, even if that means he has not taken a particular product from 10 to 100. Whether to focus depends on the product’s attributes and ceiling—B2B products may merit years of depth, while a B2C tool may have little runway left after a year. HIX.AI only started in May 2023, which he sees as missing the window; the company that truly caught the rhythm was Monica.
- In areas where the founder lacks expertise, 阿彪 advocates hiring senior talent directly and bringing in trusted outside experts to help with interviews. “Never hire mid-level people,” he says, because the company ultimately pays for the trial-and-error cost of cheaper hires, while senior talent has often had its tuition paid by previous employers. Over the past year, the company’s best organizational decision was filling out its influencer and advertising skill trees. At the same time, he no longer believes in extracting a week or two of schedule through forced overtime: “We picked the right product direction.” On the right slope, arriving a little earlier or later makes little difference; founders should “let themselves and others off the hook” （“放过自己，也放过他人”）.
Deep dive
1. Pollo Treats the “Everything Bundle” as Market Expansion, Not a Technical Moat
阿彪 was born in 1991 and graduated from Sun Yat-sen University. He previously worked at Wondershare, left in 2018, spent 2 years freelancing, and formed the company in 2020. The team now has around 90 people, has built at least 20 products, and has reached eight-figure ARR.
Pollo’s goal is a “one-stop AI creative production platform”: models are simply tools users can choose from, while the platform also covers asset management, community, generation, and editing. Professional creators get a Studio-style workflow; general users are served through image and video effects. The “everything bundle” is the way to expand from vertical creative work into mass-market entertainment.
The host asks whether this dual positioning is bound to make the product bloated. 阿彪 concedes that “this could indeed be a problem,” citing 剪映 as a reference: its app covers both effects and editing, while its web product focuses on professional creation. At the time, Pollo still planned to support both groups on the web, using interaction layers to separate them. As of recording, the new version was expected in roughly a month and a half—around June.
2. Video Ran Ahead of Strategy; the Product Filled Out the Creative Chain
Pollo was not the result of a grand plan. The team was experimenting with AI music and AI video at the same time, initially favoring music and building that first. Video ended up performing better. In August or September 2024, through internal connections, the team became one of the earlier products to integrate 可灵’s API; the original idea was simply to “put a wrapper around it and give it a try.”
The first shift in understanding was to build “Poe for video,” gradually integrating video models with public APIs. At first, the team did not even prominently show which underlying model—可灵—was being used. Only later did multi-model aggregation become an explicit product form.
The second shift came from real workflows: users were more likely to turn images into videos than text into videos. Pollo therefore added image models, followed by light-editing functions such as upscaling, erasing, and background removal. After video generation, it added background music, sound effects, slow motion, and upscaling, connecting image generation, editing, and video generation into one chain.
3. The US Is the Biggest and Hardest Market; Start with Multilingual Low-Hanging Fruit
7 years of experience going global meant the team defaulted to English plus multiple languages at launch, rather than betting first on a single country. 阿彪 naturally wanted to attack the US, the “fattest cake,” but competition was stronger, KOLs demanded more from products, and the team lacked offline relationships. A wrapper product cannot win over top influencers simply by spending money; even $100,000 or $200,000 does not guarantee success.
The practical strategy was therefore to cover the entire market, working through the influencers available in each country and “picking all the lowest-hanging fruit first.” Each country contributed some revenue; together, that could still reach meaningful scale. Once the product and the company had enough strength to win over high-profile top KOLs, they could concentrate on the core markets.
4. Roughly 4M Monthly Visits Came from a Three-Pronged Stack, Not an SEO Breakout
Asked about the claim that Pollo reached roughly 4M monthly visits within months, 阿彪 first corrected the framing: the company was running SEO, influencer marketing, and advertising in parallel, with advertising split between Google and Facebook. The “three-pronged playbook” was working simultaneously. He described the outcome as the “tenth bun”: viewers see the last one, but not the first 9 that came before it.
The internal growth curve also had no sudden milestone-driven spike. It moved steadily upward, which 阿彪 attributed to the ongoing accumulation of 3 channels rather than a single launch, keyword, or influencer.
The broad SEO framework is to build 4 types of pages—core features, utilities, models, and articles—then use PR and backlinks to build authority and trust. The arithmetic is simple: “100 pages, 10 visits per page” produces 1,000 visits a day. Without landing pages that capture intent, there is nowhere for the traffic to land.
5. Directly Usable Landing Pages Serve Both Users and Crawlers
The host points out that this framework sounds like an introductory textbook and contains no secret trick. 阿彪’s answer is that the gap lies mainly in the details. Pollo’s key design is that users can use a feature as soon as they enter its page, without first viewing a showcase page, clicking “Try it now,” logging into a dashboard, and then spending time looking for the original feature.
The page “looks like this” both before and after login, but the information density changes. Before login, the feature area occupies the top half, while descriptions, examples, and usage guidance remain below as SEO content. After login, the lower section is hidden and the page becomes an immersive workspace. 阿彪 believes this serves both Google’s understanding and the user experience—and that “the core of SEO is also the user experience.”
Model pages cover branded terms users search directly, including 可灵, Runway, Sora, Lumina, and 海螺. High-volume US keywords may not rank well, but the same English term may rank better in Australia, New Zealand, or Singapore. If the model provider has not built Spanish or other multilingual pages, Pollo can capture the local-language traffic it leaves uncovered.
6. AI Writing Is Only a Copilot; SEO Has Never Been Free
When writing blog posts, the team uses “half AI, half human” because users have become more demanding. 阿彪 prefers an opening that does not mechanically list “the 10 best AI video tools,” but instead speaks as a tester: what he tried, what problems he found, and what happened, told in the conversational manner of a friend “walking you through it.”
His view is that AI can assist, but “simply becoming a hands-off manager is unrealistic.” When machine-written content becomes uniformly stiff and impersonal, human experience and voice become differentiators in their own right.
Another common misconception is that SEO costs nothing. Article production, page and feature development, PR, and backlinks all require budget. SEO is not “simply milking Google for free traffic”; it may just offer better ROI than other acquisition channels.
7. Backlinks Should Grow Naturally; Wait at Least 3 to 6 Months for SEO Feedback
Google does not, in principle, allow purchased backlinks and may simply nullify suspicious links. 阿彪 suggests working backward from the website’s lifecycle: a new site might first accumulate ordinary links from Medium, Quora, and similar sources, then gradually attract higher-quality links as attention grows. That looks more natural than starting with nothing but expensive authority endorsements.
A new site will generally not be penalized for buying 3 to 5 links early on at more than $1,000 each, because search engines need some tolerance; otherwise, competitors could easily attack one another. But suddenly spending $100,000 on premium backlinks can backfire. The cadence also needs to be smooth: a dozen-odd to 20-30 links per month is far better than 100 in the first month, zero for 2 consecutive months, then another 100 in month 4.
SEO usually takes 3 to 6 months to show results, and changes during that period offer little immediate feedback. Content, site structure, URLs, backlinks, keyword direction, and competitive intensity all shape the outcome. For someone without enough experience, there is a “fairly high possibility” of doing SEO for 6 months without results; the resulting sense of helplessness is not unusual.
8. The SEO Divide Is Architectural Thinking, but the Skill Tree Has a Ceiling
When interviewing SEO candidates, 阿彪 first asks them to explain the complete process of taking a website from 0 to 1. Someone who only knows how to find keywords, build pages, and place backlinks is still operating at a shallow level. The next-level answer is “satisfy user intent”—whether the user wants information or wants to complete a specific task.
Top talent must also understand product positioning and future direction, combining marketing and SEO capabilities. They need to place user experience, page-level conversion, and the broader growth logic inside one architecture rather than optimizing a single module.
SEO employees at 阿彪’s company often had no relevant experience before joining; he trained them through internal documentation. He recommends English-language material from Ahrefs, Semrush, Authority Hacker, and Backlinko: “Reading 100 articles” is enough to fill in the theory, after which candidates can dissect 50 to 100 strong websites—their architecture, copy, URLs, and layouts—to build case-based intuition.
But SEO is no cure-all and does not fit every industry. It has a clear traffic ceiling and cannot single-handedly support $10M-plus revenue, let alone more. The real global judgment is to understand the ceiling and use case of every marketing channel first.
9. Influencer Distribution Needs a Hook First; Views Do Not Equal Payment
Pollo’s influencer-marketing challenge was that it had neither the technical boost of a proprietary model nor much controversy, spectacle, or novel interaction. 阿彪 believes people scroll short videos looking for “something fun.” GPT-4o’s earlier interaction update was good in itself, but would not have generated the same scale of distribution on its own; the Ghibli-style trend pushed it to another level. Digital-human products going viral through “郭德纲 speaking English and foreigners speaking Chinese” followed the same logic.
The early team pushed mechanically and paid its tuition. It later shifted to promoting fun effects, but a new contradiction appeared: “He likes it, OK, and then he’s gone. He doesn’t pay.” More recently, the team has tried emphasizing that Pollo offers one-stop access to the latest video models, though 阿彪 does not claim the results have been proven.
His honest answer on influencer ROI is, “Even now, I don’t know,” because attribution is so difficult. The team mainly uses CPM, views, and budget to decide whether to continue a partnership, rather than relying on a detailed revenue-recovery model.
10. If You Don’t Understand Paid Acquisition, Hire Senior Talent First; Channel Fit Is Decided at Product Selection
Asked about advertising tactics, 阿彪 has one answer: hire a senior media buyer, then pay an experienced friend to help with the interview. For a product, marketing function, technology, or development stack that the founder knows nothing about, “never hire mid-level people—always hire senior people.” Senior talent has usually had its tuition paid by a previous employer; founders pay for the trial-and-error of cheaper hires.
Over the past year, he believes the company’s best decision was filling out its influencer and advertising skill trees. Coming from SEO himself, he understands the limits of a single channel and no longer wants the company to depend on only one acquisition capability.
Channel fit is often determined at the moment of product selection. E-signature is boring and lacks a viral hook for influencer marketing, while paid acquisition means competing against listed companies that can comfortably spend several dollars per click. Cross-border e-commerce is also a poor fit for SEO; 阿彪 cites Shein’s heavy reliance on advertising. The three-pronged playbook does not mean every product should receive equal effort—it means finding the channel that fits best.
11. Timing and Product Attributes Determine Whether to Diversify or Go Deep
阿彪 rejects the outside image that he always gets the timing right. HIX.AI only started in May 2023, after ChatGPT had already gone through several waves of attention; the company that truly captured the initial dividend was Monica. Manus has recently shown him that, after raising funding and bringing in better talent, a team really can build a more innovative product.
His approach to finding opportunities is not mysterious: keep tracking new products, technology developments, and open-source projects, then think about how to embed them into existing products or create new use cases. Because he is not from a technical background, he often relies on Chinese AI media accounts to process news about GitHub, Twitter, and major companies, marking items for later and researching them more deeply at work.
The cost of a multi-product strategy is that no single product has truly gone from 10 to 100, but 阿彪 does not regret it. Going from 0 to 1 excites him and increases the odds of turning new ideas into products. The host contrasts this with another founder who later regretted not focusing enough. 阿彪’s response is that the answer depends on the premise: a B2B product can be developed deeply for years, while a B2C tool may have “reached the end of the road” after a year. The ceiling also has to be worth concentrating on.
On his entrepreneurial goals, he offers no fundraising-style slogan: “Entrepreneurship is simply my life.” A sense of achievement, pride, and recognition from peers all matter. As for the ambition to build a company worth tens of billions of dollars, he would rather “keep it hidden in my heart and quietly protect it.” He also admires Lei Jun for remaining on the front line making cars after achieving success and recognition.
12. Several Million Yuan in Tuition Bought 3 Product-Selection Questions
阿彪 now evaluates a new product by asking 3 questions: Is the timing right? Is there an entry point distinct from the existing product? And what determines success in the category—product strength, brand, trust, compliance, marketing, or advertising? “Choosing matters more than working hard,” because if the decisive variable lies beyond a startup’s capabilities, diligent execution may not reverse the outcome.
E-signature was the most painful lesson. The team worked on it for more than half a year, lost several million yuan, and spent heavily on marketing, only to realize afterward that users cared most about brand trust and compliance. Mature players such as DocuSign, Adobe Sign, and HelloSign already dominated the market. Timing, differentiation, and the required core capability all failed, making the product nearly doomed at the point of selection.
The online PDF editor developed afterward also had poor timing, no differentiated entry point, and intense competition. 阿彪 adds that before AI, genuinely new use cases were scarce: the software market was either crowded with giants or too small to matter. The continuing evolution of AI technology has reopened a window in which startups can line up roughly alongside large companies.
AI music failed through a different mechanism. The team used the MiniMax and 天工 APIs, which were expensive, while users mostly bought the cheapest tier. Compared with Suno, the product was neither better nor cheaper, and it had no brand advantage. The remaining users simply did not know Suno—the “small shrimp” in 阿彪’s analogy. He calls this kind of business “fishing for dead fish,” essentially “a blind cat stumbling onto a dead mouse.”
13. Without a Hard Moat, 阿彪 Is Betting on Speed, Iteration, and Organizational Capability
The host puts the sharpest question directly: Is Pollo something anyone could build, with no moat? 阿彪 answers, “Not at the moment.” He admits that others can copy it and that the company has not yet created a clear gap. His response is to keep improving the product, move relatively faster so that followers find it “a little difficult to catch up,” turn marketing “up to full,” and hope that future interaction iterations create differentiation.
Among comparable products, he most admires 剪映’s systemic capabilities: its features, operations, and growth are all “systematically unbeatable.” A team like that could win with any product. But Pollo does not want to—and “doesn’t dare to”—compete head-on; it is betting on an AI-oriented user mindshare. On Krea AI, he questions whether its $83M in cumulative funding, $8M-plus ARR, and $500M valuation contain some inflation and froth, while acknowledging that its influence is undeniable and that it may have product directions or ideas the team does not yet understand.
Chinese teams have the advantage of relatively cheaper engineering talent overseas, plus a currency advantage when revenue is in dollars. Their weaknesses are limited insight into strong B2B use cases and weaker overseas local sales and business development. ChatGPT was the dividing line in 阿彪’s view: before it, only a limited number of players were going global; after AI, “everything began to bloom,” and new use cases gave Chinese founders a chance to start roughly alongside major companies.
He rejects treating capability boundaries as fixed “genes”: “Genes are what happens when you limit yourself.” Whether a company can build cloud infrastructure, models, or overseas business depends on whether the founder keeps evolving. His management style has also shifted from demanding more overtime to pursuing a sustainable pace: the company gives employees a 1.5-hour lunch break and a collective 20-minute afternoon break. If the direction is right, forcing people to work through the night to arrive 1 or 2 weeks earlier has little value; founders should “let themselves and others off the hook.”