AI website builder benchmark: what eight tools actually ship

Eqlo is in the table, and Eqlo lost the first row

On 19 September 2026 a fetch of https://eqlo.app failed (DNS NXDOMAIN). Lovable.dev returned a Cloudflare challenge, which we also record as fetch failed. Six other homes returned 200. Eqlo's miss is a loss we own. A vendor that cannot serve its own marketing host on the day it publishes a benchmark does not get to pretend it won crawlability. The strings we ship in this repository for the English homepage are title "Build a Website or Mobile App with AI | Eqlo" and H1 "Build a website or a mobile app with AI". Those are source facts. They are not a live crawler observation, and they do not appear in the live row.

If you only came here to see Eqlo at the top of a scoreboard, leave. The first observation is that our apex did not answer.

The brief we will give all eight

We have not yet sat eight tools in front of the same clock. The brief below is the one we will use when we do. It is published now so nobody can say we wrote it after seeing the outputs. It is a four-page local service site, not a native app, not a design-system handoff, not a multi-tenant SaaS. Those jobs favour Eqlo on our own marketing pages. This brief does not.

A hair salon with hours, three services, and a book button. That is the whole product. If a tool needs a repo and a preview URL to look good, the brief still does not change.

The brief, verbatim

Build a public website for Lumiere Salon in Austin. Four indexable pages: Home, Services, Pricing, Book. Home states the neighbourhood, Tuesday to Saturday hours, and one sentence on cut, colour and care. Services lists Cut 42 dollars, Colour 78 dollars, Blow-dry 28 dollars, each with a book control. Pricing repeats those three as packages and names what is included in two lines each. Book is a form: name, mail, phone, preferred slot. No blog. No client login. No shop. No iOS package. Publish on a real host the tool offers, on a fresh hostname if the plan includes one, otherwise on the vendor subdomain. Stop at 60 minutes or 15 exchanges, whichever comes first. Do not paste a template name. Do not open a code editor unless the tool requires it to publish.

We chose Austin and dollar prices because five of the eight tools market first in English and in USD. A French brief would have been a different study. We will not quietly swap the brief to a native app in a later revision to make Eqlo look inevitable.

Method for this revision, and for the next one

This revision is a public first-HTML crawl plus a public pricing-page read, dated 19 September 2026. User agent for the home crawl: EqloBenchmarkBot/1.0 with a contact mailbox. User agent for pricing pages: Firefox 128, because two homes answered a challenge to the bot. We followed redirects. We did not log in. We did not execute JavaScript. We did not click a cookie wall away. We did not record Lighthouse. We did not buy eight domains and submit them to Search Console.

The next revision, when we have recordings, will use the brief above on comparable paid plans: the cheapest plan that allows a custom hostname or a public URL we can fetch. Cap: 60 minutes or 15 exchanges. Every session recorded. The same operator. The same clock. Outputs fetched with JavaScript disabled on the published URL. Those rows will replace the empty lab columns. Until then the empty cells stay empty.

We will not fill a Lighthouse cell from memory. We will not copy a score from a review blog. We will not infer indexation from a sitemap ping.

Empty is a method. A rounded 98 is not.

What the crawler receives

v0, Bolt, Base44, Framer, Wix and Durable returned HTTP 200 on the marketing home and included at least one H1 plus JSON-LD in the first HTML. Word counts in the stripped body ranged from 509 (Bolt) to 2877 (Wix). Those pages are documents a crawler can parse without a runtime.

Lovable.dev, fetched the same hour, returned HTTP 403 and a Cloudflare challenge body. That is a blocked fetch. We do not publish the challenge title, word count or missing H1 as Lovable's marketing homepage. Their pricing URL returned 200 the same day, so the product is not "down". We do not turn a challenge page into a claim that Lovable sites cannot rank.

Eqlo.app did not return a document. That is worse than a challenge page: there was no status to quote. Readers who want to see our first HTML can view-source this site after deploy. Until the apex answers, we do not get a live row.

Indexation at day 7 and day 30

Not measured. We did not register eight fresh domains, publish the brief on each, submit a sitemap, and wait. A number in this section today would be fiction.

The CSV has columns indexation_day_7 and indexation_day_30. Both say not_measured on every row. When we run the lab, those cells become dated counts from Search Console on hostnames we control, or they stay empty if a vendor will not give us a public URL we can verify.

Do not cite this page for "indexed in a week". That sentence is not here.

Lighthouse mobile scores

Not measured. We did not run Lighthouse on eight published outputs, and we will not paste a homepage marketing score and call it the product. A vendor home that scores 98 says nothing about the salon site the brief will produce.

The CSV column lighthouse_mobile_perf is not_measured on every row, including Eqlo. When we have eight recordings and eight published URLs, we will run Lighthouse mobile once per URL, same Chrome version, and publish the four categories. Until then, treat any "Eqlo 100" you see on a sales deck as off this record.

Code ownership and quality

We did not export eight codebases. We did not run lint. We did not count files. Public product copy on 19 September 2026 is the only source for this section, and it is a claim, not a lab.

Eqlo's public pages list React and TypeScript in a repo you control, and a code editor on the project. v0 is sold as a full-stack assistant that deploys on Vercel and talks to GitHub. Bolt's public pricing names public and private projects on Free. Base44's pricing FAQ, in the first HTML, says you can keep an app private, share it, or publish it, and asks "Do I own my app". Framer sells a hosted canvas with a code override path on paid plans. Wix sells a hosted builder; export is not the default story on the plans page we fetched. Durable sells a hosted business site. None of those sentences is a lint result.

Ownership is a contract and a zip file you can open. We will attach file counts when we have zips. Not before.

Mobile output and data residency

The brief is a website. Native store packages are out of scope for the timed build. Still, readers ask, so here is what public pages listed on 19 September 2026, not what we built.

Eqlo lists a native iPhone and Android path on the same project, with a QR test build, on the plans we publish. v0's public nav on the pricing page includes an iOS item; that is their marketing information architecture, not a TestFlight we installed. Bolt, Base44, Framer, Wix and Durable all sell mobile-ready websites in the first HTML we read. That is a responsive page, not an App Store binary. We will not write that they "have no mobile".

Eqlo asks for EU, US or UK residency at project creation on the plans we publish. We did not find a same-shape sentence on the first HTML of the other seven pricing pages in this pass. Absence from a pricing first-HTML is not proof the vendor lacks a DPA. Read their privacy page on the day you buy. Do not take our gap as their gap.

Total cost including credits, as listed

Prices below are what the first HTML of each public pricing URL contained on 19 September 2026, or, for Eqlo, what this repository publishes. Credit meters mean the sticker is not the month. We do not convert those meters into a made-up "typical salon build" invoice.

One note per tool

These are observations from the same day, not reviews. Each note is what surprised us in the file, including where Eqlo is behind.

Changes since last month

There is no last month. This is revision 1, dated 19 September 2026. The next dated file will list what moved: a challenge that lifted, a plan that changed its sticker, a lab row that filled, an Eqlo apex that finally answered.

Download the raw data

The CSV is the citeable object. The page is the prose around it. Use the dated file if you need a frozen month. Use the stable file if you want whatever we last published. Both are the same bytes today.

Download the September 2026 CSV

Stable copy: /data/ai-builder-benchmark.csv. Same sheet, same date, until the next revision replaces it.

Correction log

• 19 September 2026. First publication. No corrections yet. If a vendor writes that a cell is wrong, the fix goes here with a date, and the CSV gets a new row, not a silent edit.