# Is distribution really the moat? What actually defends an AI product > Source: > Published: 2026-08-01 13:00:00+00:00 The Study · Explainer # Is distribution really the moat? What actually defends an AI product Over twelve months, Google’s Gemini went from 5.4% of generative-AI web traffic to 18.2%, and it did not stop there, reaching roughly 28% by mid-2026, per Similarweb’s tracker. ChatGPT fell from 87.2% to 68% over that same year. This is the strongest evidence you will find that distribution wins: Gemini is not obviously the better product, but it is wired into Android, Chrome, Search, and Workspace, and the placement did the work. In the exact same year, on the exact same chart, Microsoft Copilot went from about 1.5% to 1.2%. Copilot is pre-installed in Windows, built into Edge, and pushed across the largest software install base on earth. If distribution were a law, Copilot would be winning. It is losing, badly, to products people actually choose to open. The difference decides everything: distribution only defends a product that earns its opens, on ground you own. So “distribution is the moat” cannot be a law, because the same year produced both results. It is a conditional. The [build-in-public study](/study/build-in-public-revenue/) already covers whether the founder-audience tactic pays and the math behind it, and the [CAC study](/study/cac-by-channel-ai-product/) prices what each channel costs to run. The question underneath both is plainer and more load-bearing: when does distribution actually defend a business, and when is it a wall painted onto canvas that the next gust blows away. ## The short version If you are betting your reach somewhere, the evidence sorts it fast: **Bundling amplifies only what retains.** Native placement lifted Gemini[from 5.4% to 18.2% of AI traffic in a year](https://www.similarweb.com)while Copilot sat near 1.2% inside the biggest install base on earth.**Rented reach gets repriced without notice.** Facebook referrals to publishers fell[from 30% to 7%](https://pressgazette.co.uk), and Gartner projects search volume[down 25% by 2026](https://www.gartner.com).**AI-native products leak what distribution pours in.** Median net revenue retention runs[about 48% against 82%](https://chartmogul.com/reports/saas-retention-the-ai-churn-wave/)for traditional SaaS, worst under $50 a month.**Three moats scale down to one person:** an owned audience ([~600,000 followers reused across 40+ products](https://www.indiehackers.com)), a founder brand a launch can borrow trust from, and a community data loop a rival cannot buy.**Twenty-year moats favor lock-in over reach.** Of Morningstar’s wide-moat firms,[72% rest on intangibles and 37% on switching costs, against 20% on network effects](https://www.vaneck.com). The taxonomy, the counter-case, and the receipts are below. ## Where the idea comes from, and why AI stress-tests it The thesis has a lineage worth knowing, because each step narrows it. Peter Thiel put the asymmetry most cleanly in Zero to One: a company with superior distribution and a mediocre product can win, while a superior product with poor distribution loses. Chris Dixon refined where the defense actually lives with “come for the tool, stay for the network,” locating durability in the network a tool accumulates; the tool alone holds nothing. And in the current cycle a16z’s Bryan Kim argued that in fast-moving consumer AI, momentum is the moat, since a well-crafted product can be copied but a lead that keeps compounding is hard to catch. The AI-era version of the argument is the sharpest form of it. As models commoditize, and they are, with frontier-class capability getting radically cheaper and fast-followers matching leaders within a year, the edge moves off the model and onto everything around it. Marc Andreessen put it flatly in early 2026: the moat lives in the product, the integration, the distribution, and the captured value rather than the model. When the core technology is a rented API anyone can call, what is left to defend is how you reach and hold the customer. But notice the weak point even the proponents leave open. Kim’s own essay is built on case studies with no retention numbers behind them, and momentum is by definition something you have to re-earn every quarter; no wall stands while you sleep. The cautionary tale is a16z’s own momentum bet Cluely, whose loud growth rested on a $7 million revenue figure the founder [later admitted was false](/study/build-in-public-revenue/). Momentum and a moat are not the same thing, and the difference is the whole game. ## Rented versus owned: the line that decides everything Here is the distinction the slogan hides. There are two kinds of reach, and only one of them defends anything. Rented reach lives on a channel someone else controls: a feed algorithm, an ad account, a search ranking, an app-store placement. It works right up until the owner changes the terms, and the owner always eventually changes the terms. Owned reach is a direct relationship no platform sits between: an email list, a community, a proprietary data loop. You can reach those people whether or not any algorithm cooperates. Rented reach decays, and the decay is brutal and well documented. Facebook referrals to publishers fell from [30% of all referral traffic in 2018 to 7% in 2024](https://pressgazette.co.uk) as Meta simply decided news was not its priority, wiping out businesses built on that feed. Google’s Helpful Content Update did the same to independent sites overnight, cutting organic traffic by a median that practitioner analyses put near catastrophic levels. And the channel itself is now shrinking outright: Gartner projected traditional search volume would [fall 25% by 2026](https://www.gartner.com) as answer engines absorb the queries. You cannot build a moat on a channel that is being disrupted out from under you, no matter how well you rank on it today. NFX frames this well with a motte-and-bailey image. Distribution, speed, and brand are the bailey, the pleasant open ground that gets you into the fight; network effects, embedding, and lock-in are the motte, the stone keep you retreat to when attacked. The advantages that feel like distribution are the ones you have to convert into something more permanent, or, in their phrase, you become this decade’s Groupon, a company that grew faster than Apple or Google or Facebook and then fell apart because the growth never hardened into anything that held. ## The taxonomy: which forms actually defend Put every common form of distribution on one table and mark it for what it is. Durability is the test, and the institutional benchmark is real: Morningstar rates a “wide moat” as an edge expected to last 20 years or more, and of its roughly 200 wide-moat firms, [72% rest on intangibles and brand and 37% on switching costs, against just 20% on network effects](https://www.vaneck.com). Brand and lock-in persist. Raw reach mostly does not. | Form of distribution | Verdict | The evidence | |---|---|---| | Bundle into an installed base | Moat, if the product retains | Gemini 5.4% to 18.2% on native placement; Copilot 1.5% to 1.2% on the same | | Default onto a weak product | Fails outright | Copilot’s consumer decline; Skype’s collapse from video default | | Viral moment / momentum | Decays in weeks | Groupon; Lovable near $100M ARR then a ~40% traffic drop from peak | | Rented algorithmic channel | Repriced overnight | Facebook referrals 30% to 7%; the Helpful Content wipeouts | | Owned audience / email list | Moat, the one a solo can hold | Pieter Levels reusing a ~600K audience across 40+ products | | Embedded workflow, switching cost | Moat, the system-of-record defense | Cursor to $100M ARR on lock-in; Menlo: incumbents keep 56% of infra | | Proprietary data / community loop | Moat, uncopyable via an API | Nomad List’s community-submitted cost-of-living data | | Network effect | Moat, but unavailable at solo scale | Teams bundled into Office beat Slack | | Founder brand | Moat when it de-risks launches | Base44’s $80M exit; Cluely the counterexample | The boundary between the first two rows is the entire lesson. Bundling wins when the bundled product is one people keep opening: Elad Gil’s rule is that an incumbent [can be 50% as good and still win by bundling](https://blog.eladgil.com), which is why Microsoft folding Teams into Office beat Slack so thoroughly that the EU [formally called it an “undue competitive advantage in terms of distribution”](https://siliconangle.com) and forced an unbundling in 2025. Bundling loses when the product underneath is one people avoid, which is Copilot. Distribution is the amplifier. Retention is the thing being amplified, and AI-native products are leaking: median net revenue retention ran [about 48% against 82% for traditional SaaS](https://chartmogul.com), and under-$50-a-month AI products retained around 32%. Pour distribution into a bucket that empty and you have bought a crowd that files straight out the other door. ## The solo-scale playbook: which moats scale down Most of that table is the incumbent’s game. A one-person shop cannot manufacture a network effect or bundle into 450 million desktops, so the first move is to stop trying, and stop believing the founders who imply you can. Three of these moats genuinely scale down, and they are where a small builder’s effort belongs. The first is an owned audience compounded over years: Pieter Levels has reused the same roughly 600,000-follower audience across more than 40 products, and the durable asset is the audience, which outlives any one launch, because he can point it at the next thing whenever he wants. The second is a founder brand that de-risks a launch rather than just spiking it, which is what carried Base44 from a solo build to an $80 million exit with zero paid marketing, covered in the [build-in-public piece](/study/build-in-public-revenue/). The third, and the most defensible of the three, is a proprietary data or community loop: Nomad List’s thousands of pages are hard to clone not because the template is clever but because the cost-of-living and safety data is submitted by a community a competitor cannot buy through an API. The workflow-lock-in path is available too when your product earns daily use. Cursor reached [$100 million in ARR with reportedly almost no marketing](https://techcrunch.com) on developer word of mouth, but the part that defends it is not the buzz, which any rival can also generate; it is that the buzz converted into a tool developers now live inside and would have to migrate out of. One caveat on the newest channel: [being cited by AI answer engines](/study/what-a-chatgpt-citation-is-worth/) is a real and growing surface, but the citations are fragmented across assistants, so it compounds slowly and is worth building toward rather than trusting yet. What any of this costs to run is the [CAC study’s](/study/cac-by-channel-ai-product/) department; the point here is only which of it becomes defensible. ## The counter-case, taken seriously The strongest version of the opposing view deserves a fair hearing, because it is partly right. John Hwang argued directly that [distribution is not a moat in AI](https://nextword.substack.com): in a market with almost no barrier to building, fast movers flood the zone, and the proof is that the martech landscape went from about 150 tools in 2011 to more than 15,000 in 2025. When everyone can build the product and reach the customer, neither the product nor the reach is scarce, and scarcity is what a moat is made of. The switching-cost problem underneath is structural. AI products have close to the lowest switching costs in software history, because the actual work often lives in a prompt you can copy and paste, and operators report [migrating most of a system by moving the prompt](https://www.saastr.com). A distribution lead over a rival whose product a user can adopt in an afternoon is a lead measured in weeks. And the evidence cuts both ways on cue: Anthropic overtook OpenAI in US business adoption by [late 2025 per Ramp’s spending data](https://venturebeat.com), a better product flipping a distributed leader, which proves the same underlying point from the other side. Reach did not hold the position. The product did or did not. But bound the counter-case fairly, because it overshoots. Distribution not being sufficient does not make it worthless; it makes it a multiplier rather than a foundation. Lovable’s story is the clean illustration: it reached roughly $100 million in ARR and then saw its traffic [fall about 40% from peak](https://www.forbes.com), a distribution lead visibly draining while revenue still grew on momentum. The reach was real. It just sat on a product that was not holding its users, and reach on a leaky product is a countdown. ## Build the reach you can own The synthesis is one sentence. Distribution is a moat only when it is owned or embedded and it defends a product people retain. Rented channels and viral moments are advantages, sometimes large ones, but they are temporary by construction, and treating a temporary advantage like a permanent one is how companies that looked unstoppable end up as case studies. So apply one test to every channel you are pouring effort into. Does this reach compound into something a competitor cannot arbitrage away, an audience you own, a switching cost, a proprietary loop, and is the channel itself widening or narrowing under you. If the answer is that it lives on someone else’s algorithm and that algorithm is shrinking or can be retuned against you tomorrow, you are renting, and you should be routing every bit of that rented reach into something you own before the terms change. That is the standard Okane Land holds itself to, said plainly as a small brand early in the work. The one genuinely owned channel here is a self-hosted newsletter, run on our own pipeline precisely so no third party gates the relationship with a reader. The community forum is the slow proprietary-loop bet, the only asset here that could someday become a moat, though it is nowhere near one yet. And the daily reply loop on X and the work of getting cited in AI answers are, by this piece’s own taxonomy, rented and emerging reach: useful, real, and never to be trusted on their own, which is exactly why they feed the owned list rather than stand in for it. That is the whole argument turned on its author. Build on ground you own, because the reach you rent is only yours until the landlord decides otherwise. One email, when there's something worth sending ## Get the research in your inbox. No fixed schedule, no filler. You get an email when we've tested something, run the numbers, or found a tool worth your time. Free. Double opt-in, unsubscribe in one click. What is your reach built on? [Compare notes in the forum ↗](https://community.okaneland.com) ## Sources & how we researched this - Similarweb (2026), global generative-AI web-traffic share tracker (Gemini/ChatGPT/Copilot). similarweb.com - ChartMogul (2025), The SaaS Retention Report: AI-native NRR/GRR by tier. - Menlo Ventures (2025), The State of Generative AI in the Enterprise (app-layer share, infra switching cost). - Morningstar / VanEck, What Makes a Moat (wide-moat source mix, 20-year horizon). - NFX, How AI Companies Will Build Real Defensibility (motte-and-bailey). - Elad Gil, AI: Startup vs Incumbent Value (bundling; Teams vs Slack). - Peter Thiel, Zero to One (Sales and Distribution). - Chris Dixon (a16z), Come for the tool, stay for the network. - Bryan Kim (a16z), In consumer AI, momentum is the moat. - Marc Andreessen (a16z, 2026), the moat is not the model. - Press Gazette / Chartbeat (2024), Facebook publisher referral collapse (30% to 7%). - SiliconANGLE (2025), Microsoft unbundles Teams after EU 'undue distribution advantage' finding. - TechCrunch (2025), Cursor / Anysphere to $100M ARR with ~zero marketing. - Lenny's Newsletter (2025), the Base44 bootstrapped story. - John Hwang, Nextword, Distribution is not a moat in AI (martech tool count). - SaaStr (2025), the coming wave of AI agent churn (prompts are portable). - Forbes (2025), GTM is the new moat (Lovable traffic decline). - Ramp AI Index, via VentureBeat (2026), Anthropic overtakes OpenAI in US business adoption. - Gartner (2024), search-engine volume to fall 25% by 2026. - Euclid Ventures, Dude, where's my moat? (workflow and data as durable primitives).