Free product is a media line with a hard ceiling
Lovable's growth lead says it flat out. Giving the product away is one of the biggest growth levers the company has, most of the revenue arrives self-serve, and a surprisingly large number of people using it inside real companies are still paying with a personal credit card. The line that sits next to it, that 60 to 70% of the growth tactics we all ran before AI no longer apply, is the kind of claim that usually means someone has stopped doing the work. Here it means something narrower. The free product is the top of the funnel, metered on the same credit ledger that bills paying customers.
So do the arithmetic. A free workspace gets 5 build credits a day, capped at 30 in a calendar month, plus a monthly grant of 20 Cloud credits and 4 AI credits. That is 54 credits. At Pro's own published rate of $0.25 a credit, Lovable hands every free user $13.50 of retail product a month.
The cap is the whole design. Six days of using the full daily grant exhausts the build ceiling for the month, and the daily refresh at 00:00 UTC stops arriving until the 1st. A user who shows up on the 3rd and builds hard is locked out on the 9th with three weeks of calendar left. That is a media line with a stop-loss on it, and it is the part almost nobody metering an AI product has actually set. Most of them shipped generous limits, watched inference eat the gross margin, and repriced under duress in front of an angry Reddit thread.
Two honest caveats. $13.50 is retail value; the cost to serve is model spend plus hosting, which is lower and which Lovable has never published. And a ceiling that tight keeps the free plan at weekend scale, which is the point, and is also the opening a competitor attacks with a fatter grant.
The curve that pays for the giveaway
This is a budget line. Lovable added $100M of ARR in February 2026 alone, across 146 people, which is $2.77M of ARR per head. Against that, 54 monthly credits per free user rounds to zero, and the three-month payback threshold they hold free acquisition to clears on the first $25 subscription. The giveaway is the cheapest line on the media plan, and it is underwritten by a revenue curve that went from $1M to $400M in fifteen months.
| Dec 2024 | 1M |
|---|---|
| Feb 2025 | 17M |
| May 2025 | 40M |
| Jul 2025 | 100M |
| Nov 2025 | 200M |
| Mar 2026 | 400M |
The limit that converts is not the credit limit
The meter is right there in the product. A credit balance in the dashboard bar, and if you click an individual chat message you can see exactly what that message cost you. That is unusually honest instrumentation for a metered AI product, and it trains a behaviour: users learn that simple styling runs about 0.50 credits and auth setup about 1.20, and they start budgeting their prompts. Plan Mode reinforces it at a flat 1 credit a message and never touches code, so thinking out loud has a posted price of 1 credit.
Then you hit zero and three different things break. Building stops behind a blocking dialog. AI features inside your deployed app fail, because runtime calls need available credits. And the built-in backend services, database, storage and auth, pause shortly after, with the data safe and unreachable.
The third one is the only real conversion event, and I would give it its own line in the dashboard. Build credits refresh at 00:00 UTC. A free user who hits the dialog on Tuesday afternoon comes back Wednesday morning with five more credits, and does that for six days until the 30-a-month cap lands. A wall you can wait out buys a delay. A live app whose auth stopped working forces a decision, and it arrives at people who have already shipped something they care about.
Which is why I think the limit actually doing the converting is the one nobody counts. Free projects are public only, and they carry a Lovable badge on lovable.app. The trigger is the moment you send the link to a client, a cofounder or a customer and the artefact is visibly the free version of itself. That state is attached to the thing you built and survives every midnight refresh.
For teams the trigger is different again. Free workspaces cannot set a per-member credit limit and cannot switch on auto top-up, both of which arrive on Pro at $25 a month. The second person in the workspace is the upgrade event.
What I cannot tell you is how the upsell is worded at each of those moments. It's not public and i'm sure Loveable test multiple variants and likely personalise heavily per user
Five to ten, or forty to sixty
Same free product, two different businesses. The open tier converts at 5-10%, which works because the cost of serving it is capped at 54 credits. Credits handed to a partner's audience, people who already trust the sender and already have the problem, convert at 40-60%, somewhere between four and twelve times better depending on which end of each range you land on. So stop budgeting them together. The open tier is a media line you judge on volume; the partner giveaway is a partnership line you judge on payback, and Lovable holds both to three months.
Public by default is the referral programme
Free projects on Lovable can only be public. They sit on a lovable.app subdomain with a Lovable badge on them, and that one product constraint is doing more acquisition work than the affiliate programme. Every free user who ships something is a placement, the inventory grows with usage, and the unit cost falls as the base grows: 25 million projects in the first year, 100,000 new ones a day.
Compare that to what they actually pay for referrals. The affiliate programme runs through Impact and pays up to $100 per first-time subscriber, one-time, with no published recurring rate. Against a $25-a-month Pro plan that is four months of revenue handed over before a single dollar of gross profit shows up, and the company's own stated threshold for free acquisition is a three-month payback. The arithmetic only clears if the affiliate is landing Business seats at $50 a month, where $100 is two months. So either the top rate is reserved for Business conversions and the public number is a headline, or the programme is running outside the rule the rest of the acquisition stack is held to. I would tier it explicitly, $40 on Pro and $100 on Business, and publish the cookie window, because affiliates price their effort off certainty and an unpublished window prices it at zero.
The launch was a one-off and it does not transfer. Lovable v1 went out as a soft release through the founder's X account in November 2024, with no press cycle, no Product Hunt run and no paid acquisition, on top of roughly 27,000 people who had starred or installed gpt-engineer in a repo that reached 52,000 stars. That is a warm list built over eighteen months of open source, and you cannot buy the equivalent.
The buyable substitute is the hackathon. Lovable runs public ones and helps enterprise accounts organise internal ones on the platform in place of a demo, which is how a staffing marketplace like Nursa, connecting 5,000+ facilities to 500,000+ nurses, ends up with its CEO handing access to all 200+ employees. What nobody has published is attendance, cost per event or conversion off them, so treat it as a motion with no unit economics in public.
The debut brand campaign runs the same logic. The band app in the ad is a live, working app built on the platform, and the marketing team has said future campaigns will keep turning the concept into a shipped product. The same move runs on their own blog, in a register the AI-tooling feed rewards this year: a post titled "$85,000 in tokens later", agent swarms shipping 150+ pull requests a week, and a route-by-route migration of lovable.dev itself, 42 million monthly visitors, 400 routes and 850,000 lines of code, off Next.js and onto TanStack Start, with the one 11-minute incident admitted in the write-up. The format is the receipt, including the number that makes you look reckless, and what gets screenshotted is the working UI at the end of it.
The personal card is the enterprise motion
The single most useful line anyone at the company has said out loud is that a surprisingly large number of people using Lovable inside companies are still just swiping their credit card. That is the funnel. Pro at $25 a month sits under every expense threshold in every company on earth, so the buying decision meets a person who wants a working prototype by Thursday, and it meets them before it meets procurement, legal or a security questionnaire.
Then look at how the container is built. The workspace owns the plan and the credits, every plan takes unlimited members, and price follows the credits included. Upgrade one workspace, invite everyone into it, and the whole group builds from one subscription on one invoice. Owners and admins set a default monthly credit limit with per-member overrides so one person cannot drain the shared pool, a finance control shipped as a product feature, and the first thing the person consolidating the personal cards asks for.
Seat-free pricing kills the expansion trigger most growth teams are built around.
There is no seat count to watch, so the signal that a team has outgrown its plan is the pool running dry. Pro buys 100 credits at $0.25 each and a full landing page costs about 1.70, so a dozen people each shipping one page a week burn roughly 82 credits of building a month and fit inside the plan, right up until the apps they have deployed start drawing on the same balance for hosting and runtime AI, which since the 2026 billing merge they do. The binding constraint on a paid workspace is everything the built thing consumes after it ships, and that consumption scales with the customer's traffic. So I would set the expansion trigger on projected days-to-zero off the credit ledger, which already retains twelve months of activity, and route any workspace under seven days with three or more active builders to a human.
Stripe carries the rest: payments in more than 150 countries across more than 125 local payment methods, and token-metered charges once the plan's free thresholds are crossed. Klarna and HubSpot are customers, and the company claims more than half the Fortune 500 are inside somewhere.
What no source carries is a single word of the lifecycle messaging on this transition. No admin email, no in-app copy, no nudge that fourteen people on your email domain are each expensing $25. That is the highest-leverage unbuilt sequence in the business, and the join key is the email domain.
Free is cheap next to what they pay Google
The company calling free product its biggest growth lever bought 2,632,482 paid search visits in June 2026, at $2,871,043. Organic, on the same domain, was worth about $113,218, roughly 4% of what they handed Google that month.
So when someone asks what a free user is allowed to cost, the honest comparator is the click price, and at $1.09 a visit free product is cheaper by an order of magnitude. The other thing I would read off that line: $2.9M a month on search, for a product whose own pricing page already pulls 56,238 organic visits from 47 keywords, is mostly defence. I would geo-holdout the branded portion for four weeks before touching anything else on the plan.
| Organic traffic value | Paid search spend | |
|---|---|---|
| May 2026 | 140865$ | 1497537$ |
| Jun 2026 | 113218$ | 2871043$ |
| Jul 2026 | 114179$ | 2411170$ |
How I would bound the number nobody publishes
Lovable has never published a cost per free session, and nothing in front of me carries one. The only handle is a line in their own docs: AI gateway usage rates are based on the underlying provider model costs. So the pass-through is knowable in principle and undisclosed in practice, which tells you the gross margin on a free user is a number they can see hourly and will not put in a blog post.
The three-month payback threshold on free acquisition, against a $25 Pro plan, gives you $75 of allowable cost per paying customer. At $13.50 of retail grant a month the giveaway runs past that ceiling inside six months, which is why the building dialog stops the session outright.
Here is the arithmetic I would run on Monday. Their paid search ran about $1.09 a visit in June. To land inside a $75 ceiling you need a paying customer out of every 69 clicks, a visit-to-paid rate above roughly 1.5%. Free converts at 5-10%, and it keeps clearing that bar while the grant stays capped at 54 credits.
So budget the free tier the way you budget a channel: a cost per activated free user, a kill threshold written down in advance, and a monthly read on grant consumption per cohort.
Then test where the existing gate sits. Lovable has three: public projects only, the 30-credit monthly cap, and the 4 AI credits. Move the binding constraint between them one at a time, split on new signup cohorts with a fixed two-week read so the 00:00 UTC refresh cycle lands inside every arm, and take paid signups as the primary metric with activation as the guardrail. My bet is the public-project restriction converts harder than the credit wall, because embarrassment is a stronger trigger than effort. Read the expiry terms while you are in there: monthly credits die two months after issue, annual credits one month after the annual period ends, and top-ups last twelve months from purchase.
Capability gets metered next, credits get cheaper
Here is the call, and it is specific enough to check. Within twelve months the meter moves off generic credits and onto capability. The evidence is already sitting in Lovable's own documentation: the /goal command is billed on the work done but costs noticeably more, because the agent keeps working until the goal is achieved. That is an uncapped compute commitment dressed as a chat command. And runtime AI in deployed apps is explicitly priced off the underlying provider model costs, which makes the four AI credits a free workspace gets each month the one grant on the sheet with a real marginal cost attached to every call.
So that is what gets gated. The four AI credits go, or get replaced by a capability wall: a model picker, a rate limit, a paid-only toggle. Long-running agentic commands become a Pro-and-above feature or get their own price. Meanwhile the cheap surface loosens, with build credits on the plan ladder getting cheaper per credit or the five daily build credits going up, because build credits are the ones that manufacture public URLs and public URLs are the distribution. The tell is that Lovable already labels the monthly Cloud and AI grants a temporary offering subject to change, and there is one recorded pricing change on the free tier, logged on 3 September 2026.
The precondition is the whole thing, and it does not travel. Lovable's free output is a public project, hosted on a lovable.app subdomain, carrying a Lovable badge. The free user produces an ad while they consume a sample. Free stays generous as long as generosity manufactures inventory.
Which is why I think the companies that quietly shut their free tiers over the last eighteen months were mostly right to. If your free output stays private, a document nobody sees, a query nobody shares, an internal dashboard behind SSO, then the free tier is a cost line with a conversion rate on it and you should price it like one. Lovable's giveaway is a media buy where the creative is the user's own project, and the reason it survives contact with a CFO is the 30-credit ceiling.
