Lovable, Replit and five other big AI app builders all lost traffic this summer

Lovable, Replit and Base44 drew fewer estimated visits in September than in May, while Netlify’s rose 18%. Where our histories say to look next.
Table of Contents
If you are building an AI app builder, or a product that depends on one, our May-to-September histories are worth a look.
The seven large AI app builders we could measure all had fewer estimated visits in September than in May. Lovable fell 14%, Replit 20% and Base44 22%. Rork lost more than half.
Some of that is ordinary. Most of the sites we track were smaller in September than in May. But six of the seven fell further than three-quarters of sites their size, while two of the hosting platforms their apps can be deployed to grew: Netlify’s estimated visits rose 18% and Vercel’s 3%.
These figures end in September. They estimate visits to each company’s website. They do not count customers, apps built or revenue.
All seven fell, and six fell further than three-quarters of sites their size
We looked for products whose main offer is turning a prompt into a working app, with at least one million estimated visits in May or September and an estimate for both months. Seven qualified. Broader site, hosting and design products that have added AI builders, such as Wix, Webflow, Figma and Hostinger, are not in the group.

| Tracked domain | May 2026 visits | September 2026 visits | Change | Median for sites that size |
|---|---|---|---|---|
| lovable.dev | 34,814,456 | 29,902,065 | −14.1% | −2.9% |
| base44.com | 17,995,491 | 13,964,588 | −22.4% | −2.9% |
| replit.com | 13,589,900 | 10,850,320 | −20.2% | −2.9% |
| emergent.sh | 8,162,250 | 5,869,454 | −28.1% | −8.0% |
| v0.app | 4,111,359 | 3,236,707 | −21.3% | −8.0% |
| bubble.io | 2,184,728 | 1,648,654 | −24.5% | −8.0% |
| rork.com | 1,050,537 | 503,001 | −52.1% | −8.0% |
The last column matters. Across the 8,708 sites we track with at least 20,000 visits in May and estimates for both months, the median change was −11%. A falling line on its own says little.
Size changes the picture. Among the 331 sites with more than 10 million visits, the median change was −3%, and only a quarter lost more than 12.5%. Lovable, Base44 and Replit each lost more than that.
Among the 946 sites with one to ten million visits, a quarter lost more than 22.6%. Emergent, Bubble and Rork are in that bottom quarter. v0.app, at −21%, sits just above it.
The timing differs by company. Base44, v0, Bubble and Rork had done most of their falling by August. Lovable and Emergent were down only 3% and 5% through August, and most of their decline came in September alone. Their figures depend heavily on that one month.
Bolt.new has no September estimate in our data, so it is not in the chart. Its estimated visits fell 5% between May and August.
Three platforms that apps get deployed to held up
Generated apps have to live somewhere. Lovable’s documentation describes connecting a Supabase project as an app’s backend. Bolt’s describes publishing to Netlify. v0’s describes one-click deployment to Vercel.
Between May and September, Vercel’s estimated visits rose 3% and Netlify’s rose 18%. Supabase slipped 3%, which matches the median for a site its size.
Read that as context, not as cause. Most visits to those three platforms have nothing to do with AI-generated apps, and not everything downstream held: Clerk, a sign-in service, fell 14.5%.
Several explanations fit, and our data cannot separate them. Fewer new people may be trying the builders while the apps already built keep running. The work may be moving to tools that never need a visit to a builder’s website. Or the two groups may simply be unrelated.
Two builder searches peaked in April
Search demand tells a similar story with a longer memory.

- “ai app builder” reached an estimated 18,100 monthly searches in April and was down to 8,100 by September, a 55% fall. That is exactly where it stood a year earlier.
- “lovable ai” peaked at 90,500 in April and ended at 33,100, below its level a year earlier.
- “no code app builder” fell from 18,100 to 2,400 in a year, down 87%.
- “claude code cli” rose from 3,600 to 14,800, up 311%. “codex ai” rose from 2,900 to 14,800. Both are off their own 2026 peaks of 22,200.
The site data does not show a simple move from builders to coding tools. Of the seven coding tools we compared, Cursor’s estimated visits rose 3% and opencode.ai’s rose 36%. Kiro fell 16%, Google’s Antigravity 31%, Zed 38%, Warp 58% and Trae 67%.
Nor is this a collapse. Searches for “base44” were 83% higher in September than a year earlier, and “replit” was level. Those brand searches peaked in February and March. What the histories show is interest coming down from a high earlier in the year.
What we would look at if we were building here
The general-purpose builder is a hard pitch right now. Seven large competitors each drew fewer visits in September than in May, and the generic search for the category is back to last year’s level. At least one incumbent says it is moving upmarket. Replit’s chief executive told TechCrunch in October that enterprise contracts had gone from about 2% of revenue to about 15% in a year.
We would look at what happens after the app is generated. Two hosting platforms grew while all seven builders we measured lost visits. That is thin evidence, and it is enough to ask a question: what do people need once the app exists? Keeping it running, fixing it when it breaks and moving it somewhere else are all candidates.
That is a direction to investigate, not a finding. Before building anything, we would find ten people who shipped an app with one of these tools more than three months ago and ask what happened next. Is it still live? Who maintains it? What did they pay for after launch, and what did they give up on?
If the same problem comes up in most of those conversations, test whether solving it earns a paid pilot. If the answers scatter, move on.
To keep watching this group, our ranking of AI developer-tool companies by monthly growth is public, and Replit’s profile collects what it has said about valuation, funding and revenue, with sources.
How we checked the numbers
We retrieved stored records from Night Watcher on November 16, 2026. The traffic comparison uses estimated monthly visits for May and September 2026. The seven builders are the tracked products we classified as prompt-to-app builders, with at least one million estimated visits in either month and an estimate for both. We chose them by hand, so this is the group we could measure and not a census. The data is modeled domain traffic across all channels. It is not first-party analytics, and it does not measure use inside desktop apps, code editors or command-line tools.
The size comparison uses every tracked site with at least 20,000 estimated visits in May and an estimate for both months: 8,708 sites, grouped by their May visits.
For search demand, we inspected 148 keyword histories from a targeted search and compared four phrases from September 2025 to September 2026. This is a selected comparison, not a survey of every related keyword. We excluded duplicate spellings and did not sum overlapping queries.
Nothing here measures revenue, retention or how many apps were built. We have not interviewed customers of these products.
Jordan Cole
Creator of NightWatcher AI. Specializes in data-driven insights for AI product development, market validation, and competitive analysis.