By the PrototypeVerdict Test Team
Our Base44 and Lovable links are referral links: if you sign up through one, Base44 or Lovable may reward us, at no extra cost to you. How we make money.
Yes, AI can rebuild a recognizable version of an existing website. That is not the same as moving a working website. Lovable kept our page layouts and tables more faithfully in these tests. Base44 reproduced much of the content but also invented pages and badges, reversed lists and replaced many destinations with # links.
Neither gave us a verified replacement for the live site. The homepage idea matcher did not carry over as the same working feature. Newsletter forms varied from dead copies to a more surprising result: one Lovable copy retained a live external form destination. Do not submit a form in a clone until you have checked where it goes.
We ran six imports: the same homepage, review and guide in each tool. We used our existing plans, stopped at preview and did not buy anything. Only import a site you own or have permission to copy; our tests used PrototypeVerdict.com, our own site.
New to these tools? Start with vibe coding for beginners. Moving an app’s code is a different job, covered in our Bolt and Base44 app import test.
Quick verdict
| Question | Lovable website import | Base44 Start from URL: Content & Design |
|---|---|---|
| How close was it? | Closer to the source in our checks; homepage card order and inner-page table shapes preserved | Recognizable styling and content, but more rewriting and additions |
| What came over? | Headings, navigation, footer, review and guide tables, many images | Much of the text, figures, navigation and tables |
| Biggest problems | Missing destinations; inconsistent form behavior; homepage matcher inert | Many # links, invented pages and badges, reversed card/list order, forms posting nowhere |
| Phone result | Homepage stacked at 393px; inner-page table views had awkward scrolling | Homepage menu worked around 390–400 CSS px, but horizontal overflow remained |
| Credits measured | Review 1.4; guide 3.2; homepage UNMEASURED | 4.00 message credits for each page; 12.00 total |
| First preview | About 3, 4 and 8 minutes | About 8, 9 and 8 minutes |
| Plan used | Pro, as recorded in the test plan | Builder, as recorded in the test plan |
Our pick for a faithful starting point: Lovable. That verdict is about these three pages, not every website or every feature. Base44’s result is better treated as an adaptation to audit than a copy to trust.
Is it an import or a rebuild?
From the results, treat both as a rebuild from the public page, not a migration of the original system. The copied review became the home page of its own project. Navigation could look complete even when the pages behind it did not exist.
Lovable’s import dialog explicitly warned: “Only what visitors can see is read. Logins, databases, and other backend features aren’t included.” Our homepage looked familiar, but its “Match my idea” button did nothing. Base44 substituted a keyword check rather than our matching logic.
Vendor wording:
- Lovable: the free-import page says the first website or project import is free until you see a working preview, then standard credit usage applies. You get one free import per account and per workspace, shared across both import types; the Free badge in the import dialog shows whether the import is covered. The website-import docs add that website import is available on all plans; the free period ends once the first page is imported and covers at most the first five project-chat messages; the workspace still needs credits available to start any import, including the free one; each project imports pages from one site; one request adds up to 10 pages (you can keep adding later); and an account can start at most 10 website imports and read at most 100 pages per hour. No per-project page cap is stated. Do not use the conflicting figures from the older idea-brief mirror.
- Base44: the test dialog offered “Content & Design — Build a similar site” and “Design only — Create a new site in the same style.” Its Start from URL docs say the feature is available on all plans, including Free, and that starting from a URL uses credits during generation. Free-plan message credits, per the Credits docs, are up to 5 a day and 25 a month on your personal Free workspace (other Free workspaces get none). No separate Start-from-URL allotment is stated.
We tested the interface and output, not those offers on a new Free account. A visible preview is also not proof that every original feature works.
What we tested
| Test page | Why we chose it | Runs completed |
|---|---|---|
| Homepage | Navigation, builder cards, forms and idea matcher | LOV-P0 and B44-P0 |
| Bolt review | Long article, four tables and 11 figures including a GIF | LOV-P1 and B44-P1 |
| Landing-page guide | How-to text, four tables, four screenshots and FAQs | LOV-P2 and B44-P2 |
Each URL was imported separately. These were not six stages in a complete multi-page migration. Lovable’s deep-URL imports each created a new project; Base44 also made separate builds from the supplied pages.
The optional Base44 vs Lovable comparison was not imported. Base44’s Design only mode and the informal Bolt URL-paste check were not run. The plan proposed more than the completed tests, so we do not present those proposals as results.
We inspected preview output, counted elements and checked selected text and layouts. We did not make a full word-for-word or table-cell diff. For some original-page screenshots, the test agent rendered logged-out HTML locally with scripts stripped and the site’s CSS referenced; those are layout references, not live functional tests. See how we test for our broader approach.
Step by step: Lovable’s website import
These are the labels recorded in our runs, not a promise that the menu will always stay the same.
- Open the dashboard’s + menu. Choose Import project to Lovable. The next dialog was titled “Import a project.”
- Choose Website. Under “Everything else,” the option said “Paste a website URL.” That opened “Import website,” with a Website URL field and Import website button.
- Enter a URL you can copy. No prompt field, page picker or mode settings were offered in this dialog. We started with the homepage, then made separate imports of the review and guide URLs.
- Confirm permission. In the new editor, Lovable asked whether we owned the site or had the owner’s permission. We answered Yes.
- Wait for the preview and read the report. The homepage took about three minutes. Its report admitted that the idea matcher did not work and that only the homepage had been imported.
- Audit before asking for more. Lovable offered to rebuild the matcher or import more pages. We did neither. We also did not record a separate user-triggered “compare with original” step, even though Lovable described its own desktop and phone checks.
No free-import badge or credit-cost note appeared in the homepage flow. That observation does not establish whether the account qualified for a free import.
Step by step: Base44’s Start from URL
- Choose Start from URL. The mode picker offered Content & Design and Design only.
- Select Content & Design. For the homepage we entered: “Recreate this website as faithfully as possible.”
- Add the URL. The dialog warned, “Only use URLs where you have rights to the content.” After Add, the homepage chip showed the domain and “Content & Design.”
- Let the build finish. The homepage took about eight minutes. Base44 said it created Home, Reviews and Compare, but also admitted that the extra pages were new and that it had not previewed the result itself.
- Check the output, not just the report. We examined desktop and narrow-window previews. We ignored domain suggestions and did not enable sign-ups or add search.
The review and guide are logged as Content & Design runs, but their individual reports did not restate the chosen mode and prompt. We do not assume every homepage setting was independently re-confirmed in those reports.
What came over: element checklist
| Element | Lovable | Base44 Content & Design |
|---|---|---|
| Headings and text | Review: 16 H2s; guide: 15 H2s. Limited checks found no invented content | Guide: 15 matching H2s; review heading presentation changed; additional labels and content appeared |
| Menu and footer | Visually present, but many destinations absent | Visually present, but many links were # placeholders |
| Homepage cards | Six cards in the source order | Six cards in reverse order; free-tool and comparison lists also reversed |
| Images | Homepage: one image vs four live; review: 11 including GIF; guide: four | Review: all 11 figures; guide: all four screenshots; homepage image detail not fully checked |
| GIF | File present in the review; animation not checked | Review GIF rendered; animation behavior not separately documented |
| SVGs | Homepage: six vs seven live; missing item unidentified | Logo/name reproduced; SVG detail not audited |
| Tables | All four review tables and all four guide tables retained their recorded shapes | Review: three real tables, verdict converted to div rows; guide: four tables, but verdict had one fewer row |
| Forms | Homepage lacked a newsletter form; review forms had no action; guide retained a live external form action | No live newsletter destination found; homepage/guide newsletter behavior was thank-you only |
| Analytics | No GA/GTM reported in the copies | No GA/GTM reported in the copies |
| Search metadata and JSON-LD | Not independently logged; review builder claimed it added a title and description | Not reported in the completed inspections |
| Fonts and colors | Homepage serif/green styling recognizable; not numerically measured | Homepage styling similar; review gained gradients, glows and motion |
Homepage: the shell survived better than the behavior
Lovable preserved the hero, navigation, Start here section, free tools, comparisons, FAQs and footer. Its six builder cards stayed in the original order. The phone view stacked properly with a hamburger menu at 393px.
That is not a pixel-perfect pass. The copy contained one image against four on the live page, and six SVGs against seven. We did not identify each missing asset. The newsletter form was absent, and the idea matcher was inert. About 64 links were relative rather than #, but destinations beyond the imported homepage were missing.
Base44 copied much of the homepage text nearly verbatim, with two FAQ source/more lines missing. It reversed the builder cards and two other lists. It also added motion and created its own Reviews and Compare pages rather than importing ours. About 50 of roughly 70 links pointed to #.
That difference matters: a plausible new page may be harder to notice than an obvious missing-page error. A redesign audit must check whether the destination is the right content, not merely whether something opens.
Review: tables and figures mostly survived, destinations did not
Lovable retained all four review tables with matching recorded shapes and all 16 H2s. Its inspected output had 11 images, including the GIF, re-hosted under site assets. Lovable’s own summary claimed 16 images downloaded; our element count was 11. We use the inspected count rather than treating its summary as proof.
The copy had about 2,965 words against about 3,000 live. That and the heading/first-screen checks suggested a close copy, not a complete text verification. The phone menu initially failed; Lovable said it fixed it during the import. The two newsletter forms became inert forms with no action.
Base44 retained all 11 figures and three of the four tables as actual tables. The quick-verdict table became div rows. Its token-log rows were reordered, and the last column was clipped in the phone view. Of 39 links, 31 were plain # placeholders.
It also wrote new How We Test and Reviews pages, described some reviews as “upcoming” and added decorative review labels. Those were generated additions, not evidence imported from our site.
Guide: the most important difference was the form
Lovable retained the guide’s 15 H2s, seven FAQ headings, four table shapes and four screenshots. Its text count was about 2,939 words against about 2,982 live, again without a full diff. One external link was missing, and we did not identify which one.
The footer form retained a live external subscription action. We did not submit it, and we are not publishing the destination. A clone can look unfinished while still posting to a live service.
Base44 kept the guide’s 15 H2s, seven FAQs and four screenshots, but the verdict table had five rows against six live; the missing or merged row was not identified. The newsletter form only showed a thank-you, with no email saved or sent according to its report.
Base44 also invented a Guide badge, a tested/date tag and three stat cards. Those added performance and cost claims were not on our source page and are not our redesign-test results. It fetched a generated team avatar from an external avatar service, adding a third-party request too.
Mobile: a phone toggle is not enough
Lovable’s homepage capture was 393px wide and showed a stacked layout. The inner-page phone screenshots showed horizontally scrolling tables; the review token log had stacked cells in its first row, while the guide table had tall padded rows. The guide’s phone menu was not tested, so we cannot give all three pages a navigation pass.
Base44’s first in-editor “mobile” frame was about 768px wide. That was not a phone test. A later standalone-preview check used a narrow real window at approximately 390–400 CSS px. The homepage hamburger opened a right-side menu with a close control, resolving the earlier menu uncertainty. Horizontal overflow still produced a scrollbar, and most menu links remained placeholders.
The Base44 review token log scrolled sideways with its last column clipped. The guide tables also scrolled sideways. These results are about the observed narrow views, not a complete device/browser matrix. For more on that distinction, see our mobile-app-builder tests.
What it cost: credits and time
These are observed credit movements, not dollar quotes or interchangeable units between vendors. We used existing Pro and Builder plans; current subscription pricing was not re-read and is not quoted here.
| Run | Balance or usage reading | Credits used | Approximate time to preview |
|---|---|---|---|
| Lovable homepage | No clean before/after pair attributable to this run | UNMEASURED | 3 min |
| Lovable review | 75.7 remaining → 74.3 remaining | 1.4; Details also showed 1.40 | 4 min; builder showed 3m 57s |
| Lovable guide | 74.3 remaining → 71.1 remaining | 3.2 | 8 min; browser hung during the run |
| Base44 homepage | 11.21 used → 15.21 used | 4.00 message credits | 8 min |
| Base44 review | 15.21 used → 19.21 used | 4.00 message credits | 9 min |
| Base44 guide | 19.21 used → 23.21 used | 4.00 message credits | 8 min |
Base44 used 12.00 message credits total, moving from 11.21 to 23.21 of its 250 allowance. Integration usage stayed at 8.2/10K throughout: zero integration credits used in these runs.
The two clean Lovable readings total 4.6 credits, not the cost of all three imports. A stray parallel project started during the homepage run, so we cannot attribute the shared balance movement to that homepage. We will not estimate a P0 figure or call it free. The daily-credit pool was already empty before the clean review measurement.
Lovable’s credit counts have not always matched its usage page in our tests, so we report the observed balance readings; see our AI app builder usage study.
No separate follow-up fix messages were logged for these runs. Lovable’s reported menu repair happened during its import flow; we do not assign it an invented separate cost. Times are approximate, especially the guide run affected by the browser hang, and exclude a complete migration audit.
If these results fit your needs, the tested options are Lovable and Base44. Start with one page, not a domain switch.
Before you turn a copy into a redesign
The safest next step is a small audit, not “make it live.”
- Keep the original untouched. Work from a preview and preserve the source content. A successful first screen is not a rollback plan.
- Make a page-and-link list. Check every menu item, article link and important button. Look for both missing routes and
#placeholders, plus generated destinations that contain the wrong copy. - Disable copied forms before interacting. Inspect the action and any submission code. Our guide copy shows why a retained external destination needs as much attention as a dead form.
- Compare factual content. Check badges, dates, tables, ordering and new pages against the source. Remove generated assertions you cannot support.
- Audit mobile and search setup separately. Test menus and wide tables at real narrow widths. Inventory URLs, titles, descriptions and structured data before considering a move.
Only after that would we request a visual redesign. Changing the design while missing content and behavior are still unresolved makes it harder to tell an intentional change from an import error. We did not run that second redesign phase here.
When NOT to use this
- You do not own the site or have permission. Copying a public URL does not give you rights to its text, branding or images. Both test flows displayed ownership or rights requirements.
- You only need a fresh look. If the current system works, changing its existing theme, layout or CSS avoids rebuilding the content pipeline just to alter presentation.
- Your site depends on a store, membership system or frequent CMS updates. We did not migrate any of those systems. A copied public page is not evidence that its records, checkout or login moved.
- Your forms matter. Our copies ranged from inert forms and thank-you-only behavior to a live mailing-list destination. None was validated by submitting it.
- Search traffic is business-critical and you lack a migration plan. We did not independently verify copied titles, descriptions or JSON-LD, and did not test redirects or rankings.
- You need an exact replica immediately. Missing images, reversed ordering, altered tables and generated content all required attention. Neither tool earned an exact-copy claim from this test.
Free alternatives
Stay in the editor you already use. Try changing spacing, typography or an existing layout before changing platforms. This can avoid new builder-credit usage, although your current hosting or subscription still costs what it costs. We have not checked a specific theme catalog or promised that every redesign option is free.
Make a one-page mockup by hand. Sketch the new structure, reuse content you own and rebuild only the parts that need changing. Our landing-page walkthrough can help you scope that experiment. A manual draft is also a useful reference for spotting an AI-generated addition you never requested.
Check free-import eligibility before starting. Lovable’s free-import page still offers the first website or project import free until a working preview, one per account and workspace and shared across both import types. The English docs match that eligibility and add that the free period ends once the first page is imported, covers at most the first five project-chat messages, and still needs credits available in the workspace to start. Base44’s docs still include Start from URL on Free; it uses message credits during generation, drawing from the Free personal-workspace pool of up to 5 a day and 25 a month. We did not test either Free plan here.
What we didn’t test
- Base44 Design only, or an informal URL paste into Bolt.
- The optional P3 comparison page or a completed multi-page site migration.
- Publishing a copy, connecting a domain or replacing the original site.
- Any form submission in a clone, including the retained live external form action.
- Shops, CMS migration, membership accounts, login/data transfer or paid checkout.
- Dark mode, tablet layouts, a full device matrix or a complete accessibility audit.
- Full text/table-cell diffs, every asset, copied search metadata or structured data.
- Free-plan imports, a clean Lovable homepage credit measurement or credit scaling on larger sites.
- A separate visual-redesign prompt after import, or search-ranking effects.
Redesign website with AI: FAQ
Can Lovable clone a website?
It rebuilt recognizable copies of our three public pages, preserving the review and guide table shapes in our checks. It did not produce a verified working migration: the homepage matcher was inert, linked pages were missing and forms behaved inconsistently. Only copy a site you own or have permission to copy.
Can Base44 clone a website?
Its Content & Design mode reproduced much of our content and styling, but also reversed homepage lists, created new pages and badges, and left many links pointing to #. Treat the result as a rebuild to audit, not an exact clone. We did not test Design only.
Is website import free?
We cannot call these tests free. Lovable’s measured review and guide runs used 1.4 and 3.2 credits; its homepage cost was unmeasured. Base44 used 4.00 message credits per page. Lovable still advertises one free first import (account and workspace shared) until preview, with the docs’ first-page and five-message boundary; Base44 Start from URL remains available on Free and uses generation credits. We did not run either Free offer in these tests, so check the Free badge or your remaining credits before you start.
Does my blog, shop or login come over?
We did not test a CMS, shop or login migration. The public-page copies did not preserve our original homepage matcher, and newsletter forms were inconsistent. Do not assume a familiar page means the system behind it moved.
Will it hurt my Google rankings?
We did not publish the copies or measure ranking changes. Our completed logs do not independently verify copied titles, meta descriptions or JSON-LD. Check those, your page URLs and redirect plan before replacing a live site; these tests cannot predict a ranking outcome.
Related guides
- Import a Bolt or Base44 app into Lovable
- Build a landing page with AI
- Base44 vs Lovable
- Lovable review · Base44 review · Bolt review
Sources
Our completed test evidence is recorded in tests/redesign-website-with-ai/log-lovable.md, log-base44.md and TEST-PLAN.md. The parked-results note and screenshot inventory support the scope and media selections; the completed logs take precedence over planned runs and older brief figures.
Vendor pages: Lovable free import, Lovable website-import docs, Base44 Start from URL docs and Base44 Credits.
Last tested: Nov 1, 2026. Vendor free-import / Start from URL docs re-read Thu Nov 5, 2026, about 9:05–9:10 PM CT. No subscription prices quoted. Published: Nov 5, 2026. Updated: Nov 16, 2026.