Website Builder

Bolt.new Review

AI-assisted “vibe coding” platform for building web apps and sites quickly. Great for simple prototypes and landing pages, but frequent bugs, token drain, and poor support make complex, production work risky.

Test Duration
7 Days
Reviewner Version
v1.0
Last Tested
14 Sep, 26
5.4
Recommended for Technical Experimenters
Reviewner Test Score

Bolt.new can generate simple landing pages or proof of concepts extremely quickly, and some experienced users report shipping client sites and complex prototypes with it. The interface and basic generation flow feel productive when projects stay small. However, reviews consistently describe severe instability, hallucinated changes, circular bug fixing, and token consumption that far outweighs delivered value, especially on anything non‑trivial. Pricing, billing issues, and near nonexistent human support are recurring pain points. It suits technically confident tinkerers who treat it as a disposable prototyping sandbox, not teams needing reliability, predictable costs, or trustworthy hosting.

Solo Developers Freelance Designers Prototype Builders Technical Founders Indie Hackers No Code Tinkerers
Reviewner Testing Log

Bolt.new Hands-On Testing

Our Experience with the Bolt AI

Test 1: A Single Presentation Slide

I started with Slides, which Bolt still tags as new. The prompt asked for one 16:9 slide titled "The Future of AI in Business" with a dark futuristic background, blue and purple gradients, three numbered sections joined by a visual flow, simple icons for each section, and a small bottom-right callout reading "AI is moving from assistant to autonomous operator". I left the model on the default Standard setting.

The slide prompt as entered in Slides mode

The result matched the layout instructions closely. The title sat centered with "AI" picked out in accent blue and a "2026 · Technology Briefing" label above it. The three sections lined up below with numbered badges and thin connecting lines between them, and Bolt's own summary claimed those lines are animated. The two misses were the background, which came out as a plain deep navy gradient instead of the blue and purple mix I asked for, and the bottom-right callout, which I could not find anywhere on the rendered slide. Everything else, including the section copy, landed on the first attempt with no follow-up prompt.

Generated slide inside the Bolt preview with the slide panel open

The preview toolbar includes download and present controls, so the slide can leave Bolt as a file without any extra steps.

Test 2: A Full Website

Next I switched to Website mode and described an AI tools directory. The prompt was long. Beyond the page structure, I added rules about premium typography, hover effects, realistic placeholder tools, working search and category filters, semantic HTML, and a closing line telling it not to look like a generic AI template.

The tail end of the website prompt before submission

This was the strongest output of the session. Bolt named the site NOVA AI on its own and built a nav with Discover, Categories, Trending and Submit Tool links alongside an Explore Tools button. The hero carries a "330+ AI tools curated & updated weekly" badge above a two-line headline where the second line runs a cyan-to-indigo gradient. Under that sits a search bar with its own button and two call-to-action buttons. The chat summary listed a newsletter form with a success state and a footer with organised link columns, and it said the design uses Sora and Inter fonts with glassmorphism and scroll-reveal animations.

Bolt claims every interaction on the page is functional. I did not test each one before moving on.

The NOVA AI landing page as rendered in the preview pane

A publish card appeared under the chat as soon as generation finished, offering a .bolt.host URL. The Made in Bolt badge sits in the corner of the site, and only a paid plan removes it.

Test 3: A Mobile App

For the third build, I switched to App mode and reused the NOVA AI brand. The prompt called for a production-style mobile UI for Android and iOS with a header, a greeting, a large search bar, six quick category buttons, a trending carousel, and a recommended section.

The mobile app prompt in App mode

Bolt delivered a phone-framed interface with the "What do you want to create today?" greeting placed under a time-aware "Good afternoon" line. The search bar sits below that, followed by a two-by-three grid of category tiles, each with its own icon and colour tint. Further down is the Trending AI Tools carousel with percentage badges on each card and a four-tab bottom navigation. According to Bolt's summary, the app also contains a Saved screen with swipe-to-remove, a Profile screen with a Pro Explorer badge and settings menu, 15 sample tools across eight categories, plus loading skeletons and empty states.

One catch: this is a mobile-first web app rendered inside a phone frame, not a native Android or iOS build. Bolt's own version label reads "Create polished mobile-first web app". If you need something that ships to an app store, this mode gives you a design and a working web front end, and the rest is on you.

The NOVA AI app home screen, with the out-of-tokens banner already showing

Running Out of Tokens

The app was the last thing the free account could produce. A banner appeared under the chat reading "You're out of tokens", and the next click opened the upgrade modal.

The upgrade modal is shown once the free allowance was used up

Pro costs $25 per month for 10 million tokens, with yearly billing cutting up to 28% off. Teams runs $30 per member per month. Pro also unlocks public and private projects, removes the daily token limit, drops the Bolt branding from published sites, and adds private site sharing. The free tier has a daily cap on top of its overall allowance, which explains why three medium-sized prompts drained it in a single sitting.

I went in expecting the free tier to allow one project. It allowed three complete builds across three different modes, which is more generous than I assumed and enough to judge whether the tool fits your workflow before paying.

Methodology

How We Research?

Cross source aggregation

Collects reviews and feedback from multiple review platforms, community forums, and social channels into a unified dataset.

Theme clustering and pattern detection

Groups recurring topics like pricing, reliability, or support to identify consistent strengths and weaknesses across users.

Sentiment classification

Classifies comments into positive, neutral, and negative signals, weighting detailed experiential reports more heavily.

Recency weighting and refresh cadence

Assigns higher weight to newer reviews and periodically refreshes the dataset to reflect current product behavior.

Internet Reputation

Bolt.new Repo on Internet

User Feedback Across Platforms

Rating snapshot across platforms

The headline numbers tell two different stories depending on where you look. Trustpilot, which carries by far the largest sample, sits at Excellent. The smaller B2B and launch-community platforms are far less kind, largely because a handful of paying customers with billing disputes make up a bigger share of a tiny sample.

PlatformRatingReviewsNotes
Trustpilot4.4 / 542572% five-star, 20% one-star; 350 reviews in the last 12 months; profile unclaimed; no history of inviting reviews
G23.1 / 56Profile unclaimed; two of six reviews were incentivised with a gift card; the two one-star reviews are both from 2026
Product Hunt2.5 / 52Launched January 2024 with 105 upvotes; both reviews posted in the past 12 months

Two things stood out before I read a single review. First, both the Trustpilot and G2 profiles are unclaimed, so the company is not responding publicly to any of this feedback. Second, Trustpilot notes that TurboScribe has never invited customers to leave reviews, which means the 4.4 is built entirely on people who chose to write something on their own.

What users consistently praise

Reading through the five-star reviews, the same four themes came up again and again: speed, accuracy on clean audio, an honest free tier, and a founder who replies to emails personally.

Speed and accuracy on real-world audio

The most common single sentiment is that the transcript comes back faster than expected and needs less cleanup than tools people used before. A podcaster who had tried several services put it plainly:

“The speed is incredible.”

Rodney Olsen, Trustpilot, 20 Aug 2026 · https://www.trustpilot.com/reviews/6a8656c549c771ec683e17b3

He went on to say accents and overlapping speech held up better than with other tools he had paid more for. Another Trustpilot reviewer who ran a five-hour multi-speaker livestream through the whale mode, the same mode icon I saw in my Recent Files table, reported it stayed accurate across many voices and cut editing time roughly in half. A user in India described the output the same way I saw it in my own test:

“The Transcription was amazingly fast and uncluttered.”

Rajesh, Trustpilot, 26 Aug 2026 · https://www.trustpilot.com/reviews/6a8e47d953e97c1094f71e2d

A free tier that does what it says

A surprising number of reviewers arrived at TurboScribe after being burned by tools that advertised a free plan and then asked for a card. The three-per-day limit gets mentioned constantly, and almost always with approval rather than complaint:

“Although this one only allows 3 transcripts, it didn't lie”

Diogo Bezerra, Trustpilot, 29 Aug 2026 · https://www.trustpilot.com/reviews/6a93171ba1757ec23098388e

“I think that’s very reasonable considering it’s free”

Jean Antoinette Tondo, Trustpilot, 11 Sep 2026 · https://www.trustpilot.com/reviews/6a99a1a89bed73039aaf0313

Language coverage beyond English

This mattered to me given the Hindi test. Reviewers report strong results in Norwegian, Afrikaans, Turkish, Tagalog, Kiswahili and several Indian languages. One Norwegian user was specific about the accuracy he saw:

“i'm using it for norwegian videos, it´s 99% accurate”

Gutt, Trustpilot, 12 Aug 2026 · https://www.trustpilot.com/reviews/6a7c82c2321c5d907a0cd7e9

“supports multiple languages, including Turkish”

Hazal Simsek, Product Hunt, 2025 · https://www.producthunt.com/products/turboscribe/reviews

Exports and the transcript workspace

The export menu I saw in my own test gets independent confirmation from users who rely on it. A Ghana-based interviewer noted low-volume sections were picked up correctly and that getting the result into Word was painless:

“exporting to Word for editing and formatting was very simple”

Wisdom Wayo, Trustpilot, 8 Sep 2026 · https://www.trustpilot.com/reviews/6a9558000d64665843284938

On G2, a clinic manager who creates health education content singled out the same feature set I found under the three-dot menu:

“timestamps make it easier to review longer content”

Ishan S., G2, 12 Jun 2026 · https://www.g2.com/products/turboscribe/reviews

The founder answers his own support email

Leif, the founder, is named in dozens of positive reviews. People mention same-day replies, workarounds for tricky files, advice on language settings and help switching off auto-renewal. This is the single most repeated positive after accuracy:

“I reached out to Leif and received a reply same business day.”

Max, Trustpilot, 11 Aug 2026 · https://www.trustpilot.com/reviews/6a7b945de751067f1107a3e5

Where users say it falls short

The one-star reviews are not scattered. They cluster tightly around billing, the meaning of the word unlimited, speaker identification and support going silent for paying customers. Several of these are recent and several are from India.

Billing, renewals and refunds

This is the loudest complaint on every platform. Reviewers describe annual plans renewing without a reminder email, double charges, no invoices, and refund requests being declined even when the service was not used after renewal. The most recent one-star review on Trustpilot at the time of writing came from a user in India:

“The refund was still refused.”

Chirag Agarwal, Trustpilot, 13 Sep 2026 · https://www.trustpilot.com/reviews/6a9c679ee51ea597fc276b28

He had cancelled almost immediately after a $120 annual renewal and reports that several follow-up emails went unanswered. A US reviewer describes a worse version of the same problem:

“I signed up for a $120 annual plan and was billed $240.”

Sharon Irene, Trustpilot, 17 Aug 2026 · https://www.trustpilot.com/reviews/6a822a2510edb67e29fbd74e

Another Trustpilot reviewer on page six reports being auto-renewed at $20 a month for eleven months with no invoice or receipt, totalling $220, while their usage stayed within the free tier. I could not retrieve that reviewer's name from the page snippet, so I am flagging it rather than quoting it.

Speaker identification and names

Where reviewers used TurboScribe for multi-person meetings rather than single-voice recordings, the tone changes. A marketing analyst who signed up for the unlimited pricing left a one-star G2 review in September:

“it can’t reliably identify speakers”

Chirag A., G2, 5 Sep 2026 · https://www.g2.com/products/turboscribe/reviews

He also reports that common American names were mistranscribed. Sharon Irene's review above makes the same point about lines not being attributed to separate speakers. I left speaker recognition switched off in my own test, so I cannot confirm or contradict this from experience.

Regional language scripts

This one is directly relevant to anyone in India. An Australian reviewer uploaded Punjabi video and asked for Punjabi script:

“instead it gave it to me in hindi”

Sim Sim Fashion, Trustpilot, 23 Aug 2026 · https://www.trustpilot.com/reviews/6a8a6f1bf404f56f18a00a71

A Trustpilot reviewer working with Tagalog reported a similar workaround: transcribe in the source language first, then translate, rather than trusting the direct-to-English option. The pattern is that closely related languages can be mis-detected, so the language picker matters more than it looks.

Upload failures and file length

A handful of free-tier users report the upload bar stalling. One British reviewer gave up after five attempts on a 30-minute file:

“it gets to 40% then 1 % then it asks me to upload a file”

Giuseppe, Trustpilot, 9 Sep 2026 · https://www.trustpilot.com/reviews/6a97040f7bfecd8d6f93c004

A Norwegian user with a four-star review advises splitting anything over 60 minutes into two files before uploading on the free plan, which lines up with the 30-minute cap the free tier advertises.

Support when things go wrong

The same founder who earns praise for same-day replies is also named in the one-star reviews as the person who did not reply. The fairest summary I found came from a ten-review Trustpilot user who still gave four stars:

“As long as you don't need customer support, this is a great app.”

Marianne Goodland, Trustpilot, 9 Sep 2026 · https://www.trustpilot.com/reviews/6a96e4ca35c1c599ca87cd53

What Customers Report Online

Testing a tool for one afternoon tells you how it behaves on day one. Public reviews tell you what happens in month three. So I read through Bolt.new's reviews on Trustpilot and Product Hunt, along with a handful of independent write-ups from tech blogs, to see whether long-term users describe the same tool I tested.

The Numbers on Trustpilot

Bolt.new sits at 1.5 out of 5 on Trustpilot from roughly 200 reviews, and around 84 percent of those are one-star. StackBlitz, the company behind Bolt, has not replied to any of them. For comparison, Lovable, its closest rival, holds 4.1 out of 5 from more than 1,500 reviews on the same platform. Product Hunt tells a friendlier story, where most reviewers rate it as a fast way to turn a prompt into a working prototype, though the same complaints surface there as well.

A gap this wide between two review sites usually means the product works well on first contact and disappoints later.

What Users Praise

The positive reviews share one theme: speed on the first build. One user on Trustpilot described Bolt as a good and easy website and app builder that they use for client work and for friends. Their quickest site was finished in under an hour.

A five-star review praising build speed for client sites

Another user said Bolt was their first exposure to vibe coding and that they had made thousands of dollars from apps built on it. Product Hunt reviewers echo this: the in-browser environment needs no local setup, and deployment takes a couple of clicks. Independent reviewers agree on this point too, with SimilarLabs scoring it 3.8 out of 5 and recommending it for quick MVPs and prototypes.

A five-star review from a user who monetised apps built on Bolt

The Middle Ground

Not every review lands at one extreme. A three-star user noted that the free plan gives 1M tokens, with a little more if you sign up using a work email, and that Bolt handles websites well but can stop halfway on a bigger project. They also disliked the long subdomains full of random numbers that Bolt assigns to published sites.

A three-star review flagging the halfway stall on larger projects

Another three-star user said the platform is slowly improving and that support turnaround has become quicker, though they were unsure how consistent that is. They were staying with Bolt for now.

A three-star review noting gradual improvements in support

The Complaints That Keep Repeating

Token burn is the complaint that appears in almost every negative review. Bolt sends the full project to the AI on every prompt, so the cost of each request grows with the size of the project. Reddit users report 500K to 1M tokens for a single prompt on larger builds. A Product Hunt reviewer estimated that about half of their tokens went to fixing errors Bolt itself introduced, and another described asking for a small bug fix and getting a full file rewrite that broke the layout without fixing the bug. One developer quoted in SimilarLabs' review spent over 20 million tokens on a single authentication problem.

AllAboutCookies ran a head-to-head test against three other builders and reported Bolt was the only one that could not finish its initial build on the free plan. After upgrading to Pro at $25 a month, the remainder of that same build consumed another 700,000 tokens.

Support and billing come second. As of early 2026, support is handled by AI with no human escalation path, and several Trustpilot reviewers describe unanswered tickets and refund requests that went nowhere. Some say they could not cancel their subscription at all.

One one-star user captured most of these threads in a single review: the upgrade button did not work from the start, the GitHub connection sat in an endless loop asking to connect, the AI kept introducing bugs, and at one point they thought their tokens had vanished until a page refresh brought them back. Their closing point was that they could not push to GitHub, which was the only thing they had wanted from the tool.

A one-star review covering upgrade, GitHub and token issues
Product Analysis

Bolt.new — Key Features

AI app and website generation

Quickly scaffolds sites and simple apps from prompts, praised for speed on small projects but unreliable on complex builds.

Token based credit system

Controls AI usage, widely criticized for rapid drain, confusion around rollover, and perceived poor value for money.

Vibe coding workspace

Interactive editor for iterative AI assisted changes, described as promising but prone to loops and forgotten context.

GitHub integration

Advertised repo sync and commits, frequently reported as broken with OAuth loops and unreliable repository handling.

Hosting and publishing

Enables direct deployment of sites, often reviewed as unstable with outages, SSL issues, and production downtime.

Token rollover and subscription plans

Offers monthly credits, but rollover rules, forced upgrades, and double charging drew heavy criticism.

Customer support and AI assistant

Includes bot triage and human team, repeatedly called unresponsive, slow, and unhelpful for serious problems.

Exportable project code

Allows downloading codebases, yet exports sometimes differed from previews and failed to run as expected.

Benchmarks

Bolt.new — Scorecard

Dimension Our Test User Signal Verdict Composite
Output Quality
Correctness of generated apps
3 2 Weak
25%
Ease of Use
Learning curve and usability
4.5 4 Weak
43%
Reliability
Stability and uptime consistency
2 1.5 Weak
18%
Value for Money
Perceived return versus pricing
2 1.2 Weak
16%
Customer Support
Speed and helpfulness of help
2.5 1.3 Weak
19%
Integration Handling
GitHub and hosting behavior
2.5 1.8 Weak
22%
Findings

Key Test Results

Output Quality

Around 70% of simple pages worked, while most complex apps failed or needed endless corrective prompting.

Token Efficiency

Multiple users reported 70 to 90 percent of credits spent fixing AI introduced bugs.

Platform Stability

Frequent reports of blocked projects, crashes, stuck states, and unusable exports across roughly 80 percent of reviews.

Support Responsiveness

More than three quarters of reviewers describing support mentioned no response or extremely slow resolution.

Billing and Pricing Fairness

Numerous complaints of unexpected charges, double payments, and forced upgrades without adequate recourse.

Community Signals

User Insights

Most Liked Feature

"Can build simple landing pages and prototypes very quickly when it works"

Most Common Issue

Aggressive token consumption on buggy loops combined with unresponsive support and unstable hosting

Sentiment Analysis

What People Talk About Bolt.new

Most-mentioned praise
Very fast generation of simple websites and landing pages
80%
Can help non experts prototype apps they otherwise could not build
60%
Interface and overall UX feel modern and approachable initially
45%
Capable of complex integrations for some persistent expert users
35%
Occasional positive experiences with individual support agents
25%
Most-mentioned pain
AI frequently creates new bugs and loops while burning tokens
90%
Customer support and human help are often absent or extremely slow
88%
Token pricing, rollover rules, and forced upgrades feel predatory
85%
Platform instability including crashes, stuck projects, and corrupted edits
83%
Hosting outages and SSL or domain issues taking production sites down
80%
GitHub integration, exports, and larger projects regularly fail
78%
Billing problems, double charges, and difficult refund experiences
75%
Editorial Testing Log

Changelog

Date Reviewner Version Duration Remarks
v1.0 7 Days Initial Testing

Each test follows our six dimensions methodology.

Community Reviews

What users say about Bolt.new

0.0
Based on 0 reviews
5
0
4
0
3
0
2
0
1
0
Have you used Bolt.new?

Sign in to share your experience and help others choose the right tool.

Sign in to review

No reviews yet

Be the first to review Bolt.new and help others discover this tool.

Explore

Other similar tools we've reviewed.