Investment Research on AI Search and the Browser Rebuild

Core Conclusions

• The real position of AI search in the value chain is not a single "smarter search box." It repackages web indexing, answer generation, ad distribution, browser permissions, task execution and personalized context into a new information entry layer, where Search, Browser and Agent are gradually converging rather than staying independent. Google has placed AI Overviews, AI Mode, Shopping, Deep Search and Agentic capabilities on the same Search roadmap; Microsoft has wired Copilot Search, Copilot Answers, Edge/Copilot and Azure AI Search together; OpenAI and Perplexity combine ChatGPT Search and the Comet/browserization direction respectively.

• The first thing to scale commercially over the next two years is not the "fully autonomous browser Agent" but the cited answer engine plus search-ad extension plus shopping results plus enterprise knowledge search. Google has already extended AI Overviews ads to desktop and started testing ads in AI Mode; Microsoft has launched interactive ad formats designed for Copilot and let AI Max cover Copilot Search / Copilot Answers; Perplexity has experimented with ads, run a Publisher Program, and shipped the Sonar API and Enterprise Pro; OpenAI's public materials look more focused on subscriptions, search enhancement and product discovery than on immediately replicating traditional search advertising.

• The profit pool will not leave Google at scale in the near term. Google still holds roughly 90.0% of the global search engine market and Chrome still holds about 68.0% of the global browser market. This means that even if AI search changes user behavior, the first beneficiary is still usually the strongest default entry, the largest query log, the biggest advertiser network and the most complete holder of product/maps/local data. Microsoft's priority is to grab the second profit pool, especially through the interplay of enterprise, Edge/Bing, Copilot and Azure AI Search.

• But the marginal increment of the profit pool is shifting toward two kinds of companies. One is professional database companies that own high-quality proprietary content and workflows, such as RELX, Thomson Reuters and Wolters Kluwer; the other is companies that provide "pick-and-shovel" visibility, data, search and enterprise knowledge-layer products for AI search, such as Glean, Semrush, Similarweb, Yext, Azure AI Search and Elastic. They do not depend on general traffic but on high-value retrieval, professional answers and enterprise budgets.

• AI browsers will become a new entry point, but they are more likely to take hold first in high-frequency knowledge work and high-complexity tasks rather than immediately replacing the mass distribution of Chrome/Safari/Edge. Chrome, Safari and Edge still enjoy enormous distribution advantages in operating systems, default settings, account systems, bookmarks and extension ecosystems. The real challenge for standalone AI browsers like Comet, Dia and Opera Neon is "whether they can turn a side-panel assistant into the primary browsing workflow" and "whether they can achieve long-term retention and payment under security constraints."

• Looking at market structure, the AI chat entry is growing fast but is still far smaller than traditional search. StatCounter shows that in the global AI chatbot share for April 2026, ChatGPT was about 76.9%, Gemini 9.0%, Perplexity 7.7% and Copilot 3.8%; meanwhile global search is still dominated by Google. This means "AI replacing search" has already happened, but more as an incremental substitution for complex information queries than a near-term wholesale replacement of traditional commercial search.

• The most direct change the AI answer engine makes to traditional blue links is to compress low-value clicks and amplify high-value clicks. Pew research found that on Google result pages showing an AI summary, only 8% of users clicked through to traditional links, versus 15% when there was no AI summary; only 1% of visits clicked further on a source link inside the AI summary. Similarweb also noted that since Google launched AI Overviews, organic traffic to news sites has fallen 26%, while referral visits from ChatGPT to publishers grew 25x, showing that AI referrals are indeed growing but still struggle to offset the overall traffic decline.

• The content hit hardest is therefore not "all content" but content that can be summarized, substituted, and lacks first-hand data and brand preference: low-quality FAQs, generic tutorials, general reviews, some affiliate-marketing pages, content farms and weak news aggregation. By contrast, content sources with proprietary databases, strong communities, high update frequency, strong brands, professional accountability and paid-workflow lock-in actually gain bargaining power. The value of assets like Reddit, Dow Jones, Reuters/Thomson Reuters, LexisNexis and UpToDate/ClinicalKey is rising.

• SEO will not be fully replaced by GEO/AEO; it will be rebuilt into a broader budget around "crawlable, understandable, citable, recommendable and measurable." Semrush has positioned itself as a brand-visibility platform covering AI search, SEO, PPC and social; Similarweb launched AI Share of Voice, Citation Analysis and Prompt Analysis; Ahrefs launched Brand Radar; Yext directly studied the citation logic of different models. This means the budget will not disappear but will shift from "pure ranking" toward "cross-model visibility and structured content management."

• AI answer ads will form a new ad format, but it is hard to fully replicate the high margins of Google's traditional search advertising. The reason is not insufficient ad demand but higher inference cost for answer generation, fewer ad slots, more complex brand safety and attribution, and possibly added content-licensing costs. Both Google and Microsoft are making answer-engine ads the "natural next action," such as sponsored product cards, contextual shopping recommendations and interactive conversational ads, but this looks more like new inventory than a 1:1 port of old inventory.

• The clearest and most sustainable revenue paths for AI search today, in order, are: upgrading existing search advertising, high-ARPU subscriptions, enterprise knowledge search / workflow licensing, AI-assistant premiums in professional databases, shopping guidance and commissions, and API/developer platforms. Among these, professional databases and enterprise search generally have better margins than open-ended answer engines, because they more easily control content cost, reduce liability for wrong answers, and charge on a seat-based or high-stickiness workflow basis.

• The companies where the current "strong AI search/browser narrative, weak evidence" gap is most concentrated fall into two groups: one is standalone AI browsers / sidebar tools, where public evidence on distribution, retention, DAU/MAU, enterprise procurement and ad revenue is still weak; the other is GEO/AEO niche tools and agency services, where the story exceeds budget validation. For Dia, what is publicly confirmable today is that The Browser Company lists it as a core product, but it discloses no users, payment or ad validation; Opera Neon has gone from launch to public early access, but its financial contribution is undisclosed; Comet is more advanced, but security and distribution remain the biggest obstacles.

• The biggest catalysts over the next 12–24 months are not model benchmarks but changes in default entry points, ad-load validation, enterprise payment expansion, publisher agreements, regulatory rulings and security incidents. Watch in particular: the pace of Google AI Mode ad loading, whether Apple grants AI search more Safari/system entry points, whether OpenAI starts truly scaling advertising, whether Perplexity secures sustained API/enterprise/browser revenue, and the constraints the DOJ and the EU place on Google's distribution and AI Overviews.

• The biggest risk is not "no one uses AI search" but that usage will grow while profit does not necessarily grow in step. If clicks fall, ad loading stays slow, inference and licensing costs run high, and browser Agents are restricted in permissions for security reasons, then user time will grow while the profit pool may first be diluted. OWASP has listed Prompt Injection as the top risk for LLM applications; NIST points to agent hijacking / indirect prompt injection as a key issue; public research from Google, Brave, Trail of Bits and Opera also shows that web injection and unauthorized execution are real-world problems, not theoretical ones.

Value-Chain Panorama and Product Stages

AI search does not replace the traditional search chain; it layers three new dimensions on top of it: natural-language query understanding, answer synthesis and citation, and task execution. What really matters is not who first builds a product that "can answer questions," but who can synthesize answers, advertising, transactions, permissions, accounts and data into a continuously monetizable entry point. The strongest hubs today are still Google Search/Chrome/Android, Microsoft Bing/Edge/Copilot/Windows, OpenAI ChatGPT, Perplexity, and professional database and enterprise search platforms.

Table note: The "commercialization stage, margin profile, benefit intensity and investment elasticity" columns above are composite judgments; representative products and public commercialization evidence are drawn mainly from Google Search/Ads, Microsoft Advertising/Azure AI Search, OpenAI ChatGPT Search, Perplexity Hub, Brave, Anthropic, RELX, Thomson Reuters, Wolters Kluwer, Semrush, Similarweb, Yext and other public sources.

The table below places the core products users care about most into a framework of "launch — trial — retention/payment validation — revenue landing — scaled adoption":

Business Models, Profit Pools and Scenario Forecasts

Where the Profit Pool Will Sit

The core judgment is that AI search will reshuffle the profit pool but will not distribute it evenly.

In the near term, the largest profit pool most likely stays with Google, for a simple reason: it simultaneously controls the default search entry, query logs, advertiser budgets, search/maps/shopping/local data and the world's largest browser distribution. Google has rolled out AI Overviews to more than 200 countries and regions worldwide and continues turning AI Mode into the front-of-house interface for Search; at the same time, it has not abandoned traditional search advertising but treats AI results as new ad inventory and a new expander of commercial search queries.

The second profit pool is more likely to concentrate in Microsoft and professional databases / enterprise search. Microsoft's most important advantage is not Bing's share per se but stringing together Consumer Search, Copilot, Edge, Windows, Microsoft 365, Azure AI Search and its ad platform; search and news advertising ex-TAC grew 20% in FY25, showing that "AI defense" has at least partly converted into revenue growth. At the same time, enterprises will not buy "the flashiest general answers" first; they will first buy "permission-aware, secure-and-compliant search and Agents that can land in workflows." That is precisely the structural opportunity for Glean, Microsoft 365 Copilot Search, Atlassian Rovo and ServiceNow AI Search.

OpenAI and Perplexity look more like the main challengers for "incremental user time and new entry points" than the biggest near-term winners of the profit pool. OpenAI has already proven strong subscription scale and overall revenue capability, with an annualized revenue run-rate of 10 billion dollars in June 2025 and overall 2025 annualized revenue exceeding 20 billion dollars; but this is OpenAI's total revenue, not search-ad revenue. Perplexity has simultaneously bet on subscriptions, API, enterprise and browser, but its distribution cost, security challenges, copyright litigation and high valuation all mean it is still proving a "sustainable profit pool."

The truly most underestimated profit pool may instead come from "professional content + AI assistant." What RELX, Thomson Reuters and Wolters Kluwer provide is data and software for high-willingness-to-pay, high-accountability scenarios with high-stickiness workflows, not general Q&A traffic. RELX reported 2025 revenue of 9.59 billion pounds and adjusted operating profit of 3.34 billion pounds; Thomson Reuters explicitly cited AI products such as Westlaw and CoCounsel among its 2025 growth drivers; Wolters Kluwer is advancing Expert AI as a cross-vertical direction. The business model here is fundamentally more stable and higher-margin than open-ended answer advertising.

How AI Search and AI Browsers Make Money

Search advertising has the advantages of large demand, low advertiser education cost, and the ability to capture commercial intent; its disadvantage is that AI answers compress clicks, reduce explicit ad slots and raise inference cost. Both Google and Microsoft are currently making "ads" a natural action after the answer rather than discrete ad blocks in a traditional SERP.

Subscriptions are the realistic solution for AI-native companies. OpenAI, Perplexity and Claude all prioritize using subscriptions to cover inference cost, in exchange for an ad-free experience and stronger model capability; the upside of this path is direct cash flow and a direct user relationship, while the downside is a ceiling capped by the scale of users willing to pay.

Enterprise licensing is one of the highest-quality revenue models. Enterprise search and professional databases naturally require permissions, auditing, knowledge graphs and workflow integration, with a willingness to pay markedly higher than mass search. Glean went from 100 million dollars in ARR in February 2025 to over 200 million dollars in ARR in December 2025, showing that an "enterprise answer engine" forms high-quality revenue more easily than "consumer general search."

API/developer platforms are another underestimated revenue curve. Perplexity's Sonar Pro API, Azure AI Search's agentic retrieval and the Brave Search API all show that "search capability itself" is being turned into software and platforms. For many developers and enterprises, buying an answer-engine API is more practical than buying a consumer search entry.

Shopping commissions / product discovery will be the earliest high-value transaction layer to prove out in AI search. Google is already testing sponsored-retailer formats in AI Mode; both OpenAI and Perplexity are working on product discovery, because high-commercial-intent queries naturally turn "answers" directly into "purchases." This is also why the future profit pool need not land only in general search advertising but may land in the "search-to-checkout" commerce layer.

Three Scenario Forecasts

The key divergence in these scenarios is not "how much smarter the model gets" but whether ads can load, whether browser Agents can be safely granted authority, and whether users really migrate high-value commercial queries to AI entry points. Google publicly says AI Overviews drives query growth; Microsoft signals growth in the composite Copilot+search journey; meanwhile Pew and Similarweb show that AI summaries do significantly compress external clicks. In other words, the future question is not "whether users will come" but "once they come, whether money will still flow the way it used to."

Advertising, SEO, GEO and Publisher Economics

The economics of traditional search advertising rest on three pillars: clear intent, low fulfillment cost and measurable clicks. AI search may be stronger on the first, because it encourages longer, more complex, more contextual queries; but it is currently weaker on the latter two, because answers that satisfy needs directly reduce clicks, while inference and content costs raise fulfillment cost. Neither Google nor Microsoft plans to abandon search advertising; instead they make ads "embedded in answers" — which forms new inventory but raises the bar on both advertiser attribution and platform margins.

Google's public framing leans toward "AI expanding search" rather than "AI replacing search." On one hand, Google says AI Overviews brought over 10% usage growth on queries that showed it in markets like the U.S. and India, and that overall query volume, including from platforms on Apple devices, is also growing; on the other hand, Apple executive Eddy Cue testified in the DOJ litigation that Safari search declined year over year for the first time in April 2025, attributing it to users shifting to AI. Putting these two sets of information together, the most reasonable reading is not that one side must be wrong, but that AI is pulling away a portion of high-information-density queries while also creating new complex-query demand on the original platform.

For publishers, this change is highly asymmetric. Pew's data shows that result pages with an AI summary send fewer users to external sites; Similarweb's report notes that since Google launched AI Overviews, organic traffic to news sites has fallen 26%. At the same time, AI platforms are creating new referral traffic: Similarweb says referral traffic from ChatGPT to publishers grew 25x, with Reuters and NY Post among the leaders. The problem is that this new kind of referral currently looks more like a "high-intent but low-base" supplement than a fill for Google's organic-traffic gap.

This is why publisher strategy splits into three lines: license, block and sue. News Corp and OpenAI have signed a multi-year partnership letting OpenAI display News Corp content in answers and use it to enhance products; Perplexity first launched a Publisher Program, then expanded to more media partners, but still faces a Dow Jones/News Corp copyright suit, and its motion to dismiss/transfer was denied; Google's AI Overviews has also drawn an antitrust complaint in Europe, with one core dispute being that publishers find it hard to opt out of AI Overviews use alone without exiting search results entirely.

Therefore, GEO will not replace SEO but become a higher-layer budget on top of SEO. What used to be optimized was "ranking and clicks"; what is newly added is "which models cite you, under what prompts you are mentioned, the sentiment and position within the answer, and whether you can be parsed with machine-trustworthiness." Semrush, Similarweb, Ahrefs and Yext have all launched products or research frameworks around this. The budget will not explode overnight, but it has begun shifting from "keyword tools" toward an "AI visibility stack."

By order of sector impact, the most fragile are sites that depend on search-traffic arbitrage; next come general informational media and shopping-guide/affiliate marketing; only above them sit strong-brand news, communities and databases. A very important example is Reddit: it may have "part of its page visits eaten" by AI summaries, yet because its community corpus is unique, real-time and structured, it becomes a content layer that AI platforms must license. Reddit has reached a Data API partnership with OpenAI, and in 2025 grew DAUq to 121.4 million and full-year revenue to 2.2 billion dollars, with ad revenue growing strongly in 2025 and "Other revenue" increasingly absorbing new revenue such as data licensing.

AI Browsers, Agents and Security/Compliance

The core value of a traditional browser is open, render, save and sync. What AI browsers want to sell is understand, remember, execute and act on your behalf. Chrome has publicly extended Gemini in Chrome from web summarization gradually to cross-tab understanding and on-behalf-of-user navigation; Safari has placed web summarization and stronger privacy into Apple Intelligence; Edge/Copilot emphasizes cross-file, cross-tab and system-level coordination more. This shows the giants do not see the AI browser as "building a new browser brand" but as the next step in turning the browser into an operating system.

The opportunity for standalone AI browsers is that they can rewrite the interaction paradigm more aggressively. Perplexity Comet is positioned for research, shopping, comparison and complex workflows; Opera Neon writes "plan, purchase, create web apps" directly into its product narrative; Dia tries to turn the browser into a more natural AI working environment. The problem is that browsers are a market of strong distribution, high switching cost and strong default power — Chrome, Safari and Edge together already cover the vast majority of global browser usage. AI features alone may not be enough to break such a distribution moat.

The bigger real-world obstacle comes from security. OWASP listed Prompt Injection as the top risk in its 2025 generative-AI security project; NIST explicitly treats agent hijacking as an indirect-prompt-injection issue requiring focused assessment; Google Security has also publicly noted that threat actors can plant prompt injection in public web pages, and in the real world one should expect someone to exploit this to cause harm. In other words, as long as an AI browser reads web pages, email, PDFs and social content and holds the power to click, copy, fill forms or grant OAuth, it is exposed to an attack surface entirely different from "traditional web content."

Comet's public audits illustrate this well. Brave disclosed Comet's indirect-prompt-injection risk in August 2025; Trail of Bits disclosed in February 2026 that, based on threat modeling and adversarial testing against Comet, four prompt-injection techniques could be used to extract private information from highly sensitive environments such as Gmail. Opera itself also publicly summarized its prompt-injection vulnerability response for Opera Neon in October 2025. This does not mean AI browsers have no future; it means they must either be restricted in permissions or develop an entirely new security architecture to move from "impressive demo" to "enterprises dare to procure."

AI browsers are therefore more likely to take hold first in three kinds of scenarios. The first is research-intensive personal productivity, such as investing, legal work and market research. The second is enterprise intranets and controlled environments, because permissions and content sources are controllable. The third is the search-to-transaction high-commercial-intent path, such as price comparison and shopping. The hardest, by contrast, is "fully autonomous general browsing," because it simultaneously stacks the highest permissions, the highest uncertainty and the most complex liability boundaries.

Master Table of Investment Targets and Key Public Companies

Master Table of Investment Targets

The table below prioritizes public companies "with clear public evidence." Within the very large company pool the user provided, many Chinese, A-share, Hong Kong and some European/Japanese/Korean companies are indeed worth including in follow-up themes, but where current public materials cannot directly verify their AI search/AI browser revenue path, this report marks them "needs further verification" and does not force a factual buildout.

Table note: A = core direct beneficiary; B = clear benefit but needs continued verification; C = defensive AI; D = narrative exceeds evidence; E = higher risk of being hit by AI search. Financial snapshots and commercialization evidence are based mainly on official filings/IR/product documentation, Reuters and company blogs.

In-Depth Look at Key Public Companies

Alphabet This is the most central platform-type winner in AI search today. The key is not only the product lead of AI Overviews and AI Mode but that Google has connected search, shopping, advertising, Chrome, Android, maps, local, product feeds and model capability. AI Overviews already has over 2 billion monthly users and AI Mode keeps expanding; this means Google is embedding "AI answers" into its largest existing commercial search asset rather than building a separate operation. The main public risk comes from antitrust — the DOJ's remedies discussion includes a Chrome divestiture and search-data sharing — and from the possibility that Apple/Safari grants AI search more default opportunities at the entry point. The current valuation is not cheap, but it still does not fully reflect the upside case of "AI reshaping search while also holding the profit pool."

Microsoft Microsoft is the clearest acquirer of the "second profit pool." On one hand, Copilot Search, Bing and Edge help it win more complex queries in consumer search; on the other, Azure AI Search and Microsoft 365 Copilot Search embed search and Agents deep into enterprise software budgets. Compared with Google, Microsoft's consumer entry is weaker but its enterprise distribution is stronger, which suits the commercialization of high-value, low-fault-tolerance knowledge search. Search and news advertising ex-TAC grew 20% in FY25, showing AI has at least partly converted into monetization rather than pure defense. The risk is that consumer query share is still insufficient and Copilot's multi-product form easily dilutes mindshare.

Apple Apple is the "gatekeeper" that decides the reallocation of the industry profit pool, but it is not today's clearest AI-search beneficiary. Safari is still the world's second-largest browser, and Apple Intelligence has brought web summarization and privacy enhancements into Safari; meanwhile Eddy Cue admitted in court that Safari search declined for the first time in April 2025 and proactively mentioned AI search options such as OpenAI and Perplexity. This shows Apple has the ability to change the industry's competitive landscape through default distribution and system UI. The issue is that Apple's public materials do not yet show intent to turn AI search directly into a standalone profit pool. In other words, Apple looks more like the platform that "decides who makes money" than the "pure AI-search beneficiary" most worth betting on today.

Reddit Reddit is one of the very few companies in the AI-search era whose "content gets summarized, yet whose content asset appreciates." The reason is that what it sells is not traditional media traffic but a globally scarce, continuously updated, community-generated natural-language long-tail corpus. Reddit has partnered with OpenAI, and litigation filings show that large AI companies such as OpenAI and Google have obtained Reddit data access via licensing; the company's 2025 DAUq reached 121.4 million, revenue 2.2 billion dollars, with Other revenue staying in the tens of millions. Its risks are: first, the valuation already clearly reflects AI data-licensing expectations; second, both Google and ChatGPT may "digest" Reddit content and reduce direct on-site visits; third, community quality and platform governance are critical to the long-term data moat.

RELX RELX is one of the high-quality assets in this AI-search shift most easily overshadowed by the consumer-internet narrative. The truly high-willingness-to-pay scenarios for AI search are not in daily Q&A but in legal, risk, research and professional decisions. RELX reported 2025 revenue of 9.59 billion pounds and adjusted operating profit of 3.34 billion pounds, with the Legal segment launching Protégé. RELX's core is not "making answers cooler" but making answers more usable, more trustworthy and more embeddable into workflows in accountable scenarios. Compared with open-ended answer platforms, it is almost naturally better suited to be the high-margin retrieval of the AI era. The main risk is that the valuation is already not low, and that AI growth will likely be partly masked by the traditional "information-services company" label.

Thomson Reuters TRI is similar to RELX, both on the "high-value vertical search" path, but TRI's CoCounsel and Westlaw narrative is more easily recognized by the market as AI growth. The company repeatedly listed Westlaw, CoCounsel and Practical Law as growth drivers in 2025 and the second half of 2025. Unlike open-ended search, TRI's customers need auditable, traceable AI answers that can flow directly into legal/tax/compliance workflows, so its unit revenue and renewal quality are both higher. The risk is that once the market treats TRI as a "core AI beneficiary," the valuation premium rises, and product-liability risk is always higher than for ordinary SaaS.

Wolters Kluwer Wolters Kluwer's position is very important in this research because it proves a structural conclusion: the truly high-certainty profit pool of AI search likely comes from professional answers rather than open answers. WKL has made Expert AI one of its core directions in its 2025 annual report and launched UpToDate Expert AI. Healthcare and compliance are textbook industries where "users value accuracy, explainability and liability boundaries more than free," so WKL's AI shift is not about traffic but about ARPU and workflow upgrades. The risk still lies in a not-cheap valuation, and in public markets often pricing it as a steady information-services provider rather than a frontier AI name.

Baidu Baidu is the AI search target in the China market most worth considering separately. The reason is not its traditional ad business but that it already shows signs of "AI-related business growth taking over from traditional search advertising." Reuters reported that in Q1 2026, Baidu's core AI-related businesses (cloud, AI applications, Robotaxi) reached 13.6 billion yuan, exceeding half the total for the first time, while online marketing revenue fell to 12.6 billion yuan. In other words, Baidu is not waiting for AI search to monetize someday; the switch of "old profit pool under pressure, new profit pool growing out" has already occurred. The biggest risk is that China's ad macro environment is still weak, open-source model competition is fierce, and the sustainability of the AI revenue mix still needs more quarters of verification.

Alibaba Alibaba is not the most direct beneficiary in general web AI search, but it has long-term optionality in AI cloud + product discovery + Quark/ecosystem entry. Reuters reported that Alibaba Cloud revenue grew 38% year over year in the relevant quarter of FY2026, with AI products already about 30% of external cloud revenue and expected to exceed 50% within a year. This shows Alibaba's AI profit pool is more likely to appear first in infrastructure and the enterprise side, then spill over to e-commerce discovery and consumer entry points. The risk is that the market easily stuffs all of Alibaba's AI expectations into capex and cloud valuation, while evidence on the consumer side of "AI search as a transaction entry" is still insufficient.

Semrush Semrush is one of the most direct public-market mappings of GEO/AEO budgets. The company no longer positions itself merely as an SEO tool but directly as a brand-visibility platform that "covers AI search, SEO, PPC, social," and in 2025 launched enterprise optimization capabilities for AI search. The benefit logic is clear: if AI search reduces clicks and increases citations and mentions, brands need a new generation of visibility tools. But the biggest question is equally clear: how much AI-visibility budget will be net-new versus how much will replace old SEO budget is still in the validation phase. Its investment profile is more "high-elasticity pick-and-shovel" than high-certainty platform.

Similarweb Similarweb is another textbook "pick-and-shovel" name, but its advantage is owning an independent measurement framework for external traffic and AI referrals. The company has turned AI Search Intelligence, AI Share of Voice, Citation Analysis and Prompt Analysis into products; meanwhile, its report on AI's impact on publishers shows it is converting AI search from a "trend" into "a sellable data product." Q4 2025 revenue was 72.8 million dollars, up 11% year over year. If AI discovery truly becomes a new traffic source, Similarweb's value will rise in brand measurement, competitor monitoring and investment research. The risk is that the company is still small, and more and more AI-visibility tools will appear in the market.

Yext Yext's story is not "building a general answer engine" but managing structured, machine-readable, cross-location consistent factual-layer data for brands and local merchants. The more AI search relies on citations, the more valuable consistent location, service, inventory, FAQ and brand facts become. The conclusion Yext drew after publicly studying 6.8 million AI citations essentially tells advertisers and brands: in the AI-search era, the value of brand owned sites and brand-management data rises. FY26 revenue was 446.6 million dollars and ARR 444.3 million dollars, with improving profitability. The risk is that its demand release may be slower than the market imagines.

News Corp News Corp's investment logic splits two ways: professional information assets like Dow Jones benefit from AI licensing and professional queries, while the News Media traffic side faces the problem of AI summaries weakening clicks. The formal partnership with OpenAI proves high-quality news can be licensed for a fee; the lawsuit against Perplexity shows the tug-of-war between platforms and content owners will not end automatically. For investors, this is neither a pure "AI beneficiary" nor a pure "AI casualty" case but a case of layered re-rating of content assets: professionalized, subscription-based, B2B content is stronger, open news traffic is weaker.

Opera Opera is a textbook case of "a narrative that needs a strict discount." It certainly has more browser installs and brand foundation than a startup, and Neon has advanced from announcement to public early access; but to date, public materials provide no evidence of payment, retention, advertiser adoption or enterprise procurement from AI browsers. Opera therefore looks more like an "observable option" than a validated AI-browser winner.

Company Tiering and Investment Priority

• Tier A: core direct beneficiaries Alphabet, Microsoft, RELX, Thomson Reuters, Wolters Kluwer, Reddit. They either directly control the search/advertising/browser entry, or control the scarcest high-value content and professional workflows of the AI era.

• Tier B: clear benefit, but with valuation, copyright, competition or cost risk Baidu, Alibaba, Semrush, Similarweb, Yext, News Corp. The commercialization logic exists, but the pace and sustainability of profit realization still need more quarters of verification.

• Tier C: AI is mainly a defensive tool Apple, Atlassian, ServiceNow, Salesforce. Their AI search/browser capabilities matter, but more to prevent their core platforms from being marginalized than to form new standalone profit pools in the near term.

• Tier D: narrative exceeds evidence Opera Neon, Dia/The Browser Company, some standalone AI browsers/GEO niche tools. The product direction holds, but evidence on distribution, retention, revenue, ads and enterprise adoption is still weak.

• Tier E: higher probability of being hit by AI search Low-quality media, traffic-arbitrage affiliate-marketing sites, content farms, traditional SEO providers that only sell rankings rather than visibility, and some media companies heavily dependent on external search referrals. Evidence from both Pew and Similarweb shows zero-click pressure is already rising.

Private Opportunities, the Scoring Model and Final Conclusions

Important Private Companies and Primary-Market Opportunities

Table note: Except for OpenAI, Perplexity, Glean, Anthropic, Brave, DuckDuckGo, The Browser Company and xAI, this report deliberately does not elaborate on unverified data for the other companies.

The Scoring Model

This report uses the following positive scoring model, out of 100:

• Direct AI search/browser revenue exposure: 20

• User entry, search data and ecosystem moat: 25

• Advertising/subscription/enterprise commercialization ability: 15

• Content, publisher and copyright moat: 10

• Security, privacy and regulatory capability: 10

• Financial quality and margins: 10

• Valuation reasonableness: 10

It also builds a reverse "commercialization risk score," where a higher score means higher risk:

• Insufficient user adoption and retention: 20

• Insufficient search-ad monetization: 20

• Inference-cost and licensing-cost pressure: 20

• Copyright and antitrust risk: 15

• Risk of being built into the OS/browser platform: 15

• Overvaluation: 10

Based on the framework above, here is a rough ranking that leans toward "research priority" rather than "trading advice":

The most important thing here is not the decimal points of the scores but the logic behind the ranking: who has already turned AI search into revenue versus who has merely put AI into a product; who controls the user entry versus who is merely attached to it; who can price content versus who is merely dragged down by content cost.

Final Conclusions

The AI-search and browser rebuild is one of the few mega-themes in the AI value chain that simultaneously affects traffic acquisition, ad pricing, content distribution, browser permissions, Agent execution, enterprise knowledge work and copyright relationships. Its importance lies not in "whether it will change search" but in the fact that it is changing "who controls the information entry, who takes the commercial intent, who pays for content, and who bears liability for errors."

The five sub-tracks most worth watching are: AI search advertising, enterprise search, professional vertical search, GEO/AI visibility tools, and AI browser security. Among them, the first four are closer to profit realization, while the last is closer to a key constraint.

The ten public companies most worth deep study, in my priority order, are: Alphabet, Microsoft, RELX, Thomson Reuters, Wolters Kluwer, Reddit, Baidu, Semrush, Similarweb, Alibaba. The first five lean high-certainty, while Reddit/Baidu and the tool-chain companies lean high-elasticity.

The ten private companies most worth tracking are: OpenAI, Perplexity, Glean, Anthropic, Brave, DuckDuckGo, The Browser Company, xAI, You.com, and Exa/Tavily-type Search API companies. Among them, OpenAI/Perplexity decide the ceiling for consumer answer engines, Glean/Anthropic more decide the enterprise-side landing direction, and Brave/DuckDuckGo decide whether the independent-search/privacy path can keep existing in the AI era.

The five points the market most easily misreads are: First, AI search is not an "ad terminator" but an ad-format reconstructor; Second, an AI browser is not equal to "having a sidebar assistant"; Third, publishers will not disappear wholesale but will clearly stratify; Fourth, SEO will not die; it will only upgrade toward GEO/AEO; Fifth, the truly high-certainty AI-search profit pool may land first in professional databases and enterprise search rather than mass free answers.

The metrics most worth tracking over the next 6–12 months include: the ad-load rate of Google AI Mode/AI Overviews, whether ChatGPT Search starts advertising, Perplexity's API/enterprise/browser revenue, changes in the default entry points of Safari/Chrome/Edge, the share of AI referral traffic, progress on publisher licensing/litigation, and prompt-injection incidents in browser Agents. These metrics matter more than the timing of any single model release, because they directly determine the speed of the profit-pool migration.

If we compress the companies into the four clearest investment roles:

• AI search platform companies: Alphabet, Microsoft, Apple.

• AI-native answer-engine challengers: OpenAI, Perplexity, Brave, Anthropic, xAI.

• AI search pick-and-shovel providers: Glean, Semrush, Similarweb, Yext, Azure AI Search, Elastic, Search API companies.

• Traditional participants at higher risk of being hit by AI search/browsers: low-quality media, affiliate-marketing sites, content farms, SEO providers that only do traditional ranking optimization, and some media businesses overly dependent on open-web search referrals.

For narrower follow-up research worth starting immediately, I suggest converging on five themes: AI search advertising, GEO/AEO tools, AI browser security, publisher content licensing, and enterprise search. These five directions correspond respectively to: the profit pool, budget migration, landing constraints, content cost, and high-certainty paid scenarios.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.