FOWL AI Tests · Monthly · Test #2

We asked four AI tools for one HTML file. Only three gave us that.

Claude Code, Gemini, ChatGPT, and v0 — same fictional SaaS product, same prompt, word for word, explicitly asking for a single working HTML/CSS file. We opened every real result in a browser ourselves. Here's what actually showed up.

Tools testedClaude Code, Gemini 3.1 Pro, ChatGPT (GPT-5.6), v0 Max
Test caseOne-page landing page for a fictional meeting-notes SaaS tool
What we checkedThe real rendered page, not the chat reply
Key Findings — 15-Second Version
  • v0 ignored the "just one HTML file" instruction and built an entire Next.js project — you have to download a ZIP and dig out the real file to get what was actually asked for.
  • ChatGPT was the only tool that nailed the brief exactly: one portable file, no invented product details, best visual polish of the three that followed instructions.
  • Gemini and v0 both invented specific technical claims — a "no visible meeting bot" mechanism and calendar-based call detection — that were never in the brief.
  • All four reproduced the pricing ($0 and $12/user/month) correctly. Nobody got the numbers wrong.
The Setup

We invented a fictional SaaS product — Recapp, a meeting-notes tool that posts call summaries to Slack automatically — and gave all four tools the exact same one-paragraph brief and the exact same instruction: build a one-page marketing landing page with a hero, a 3-step "how it works," 3-4 features, both pricing tiers, and a footer, delivered as working HTML and CSS, not a description. No follow-up prompts, no edits. We took the first thing each tool produced and opened the real result.

ProductFictional, invented for this test
PromptIdentical across all four, word for word
Output checkedThe real rendered page, not the chat reply
VerificationOpened live in a browser, scrolled the full page
Show the exact prompt we used
Build me a one-page marketing landing page for my SaaS product. Recapp is a Chrome extension + web app that automatically joins your Zoom or Google Meet calls, transcribes them, and posts a one-paragraph summary with action items to Slack within 5 minutes of the call ending — no bot has to be manually invited, and it never stores raw video. It's built for small remote/hybrid teams (5-50 people) who are drowning in meetings and don't want a dedicated notetaker. Pricing is free for up to 10 meetings/month, or $12/user/month for unlimited. The main pitch versus competitors like Otter and Fireflies is that it posts to Slack automatically — nobody has to remember to share notes. I want a hero section with a clear headline and CTA, a "how it works" section with 3 steps, a features section covering 3-4 real value props from what I just told you, a pricing section with both tiers, and a footer. Give me working HTML and CSS, not a description — I want to actually open the file in a browser.
The Scoreboard

Five categories, scored from the actual rendered pages.

Every category is scored 1–5, dot for dot, based only on what we found in the real result — never the chat reply.

5 = Excellent 4 = Good 3 = Acceptable 2 = Weak 1 = Poor
CategoryWhat it measures
Brief HonestyWhether the tool invented a product mechanic, feature, or technical claim that wasn't in the original brief.
Structural CompletenessHero, 3-step how-it-works, features, both pricing tiers, and a footer — all present, as asked.
Visual PolishTypography, spacing, and design choices in the real rendered page.
Technical ExecutionWhether it delivered exactly the portable, single-file HTML/CSS format explicitly requested, and whether it renders without bugs.
Conversion ClarityHero message strength, CTA hierarchy, and how clearly the two pricing tiers are distinguished.
Category
Claude Code
Gemini
ChatGPT
v0
Brief Honesty
Structural Completeness
Visual Polish
Technical Execution
Conversion Clarity
Total (out of 25) 22 21 24
Winner		 20
Summary Showdown

Same product, four completely different headlines.

The hero headline each tool wrote from the identical brief — verbatim, no edits.

Claude Code

"Meeting notes that post themselves to Slack."

Gemini

"Stop losing meeting notes in the void."

ChatGPT

"End the meeting. Recapp posts the recap."

v0

"Your meeting notes, posted to Slack before you close the tab."

Sample View

The actual rendered pages — open any of them yourself.

Claude Code Single file
Open full page →
What it did differently — Plainest of the four, but exactly what was asked: one clean file, no invented features, verified to render correctly since it could check its own output directly.
Gemini Single file
Open full page →
What it did differently — Solid single-file output, but added a "no awkward meeting bot, runs via browser extension" mechanic that was never in the brief — a specific technical claim, invented.
ChatGPT Single file
Open full page →
What it did differently — Best polish of the three that followed the brief exactly: a realistic Slack-message mockup (clearly illustrative, not a factual claim), working anchor nav, no invented product mechanics.
v0 Full project, zipped
Open full page →
What it did differently — Sharpest visual design of the four — but instead of one HTML file, it built a full Next.js project with a package.json and a build step. Getting to this page meant downloading a ZIP and extracting it ourselves. It also added a "learns your calendar" claim not in the brief.
What We Found
Only three tools gave us the file format we actually asked for
We were explicit: "working HTML and CSS... I want to actually open the file in a browser." Claude Code, Gemini, and ChatGPT all delivered exactly that — one self-contained file, openable immediately. v0 built an entire Next.js application instead: a real project with its own package manager lockfile, environment config, and app router structure. It's genuinely more capable software — but it's not what was asked for, and getting the actual page meant downloading a ZIP and finding the right file inside it. That's a real cost if you just wanted a file to hand to a client today.
Two tools quietly invented how the product works
The brief never said whether Recapp shows up as a visible participant in a call. Gemini decided it doesn't ("no awkward 'Recapp Bot is recording this meeting' participant... runs via your browser extension") and built marketing copy around that specific technical claim. v0 separately added "Recapp learns your calendar so it knows which calls to join" — again, a mechanic never described in the brief. Neither claim is wildly implausible for a product like this, which is exactly what makes it worth flagging: invented specifics that sound plausible are the hardest kind of AI output to catch before they end up on a real marketing page.
Every tool got the numbers right
Free tier at $0 for up to 10 meetings a month, paid tier at $12/user/month for unlimited — all four tools reproduced both figures correctly, with no rounding errors, no invented discounts, no fabricated "limited time" framing. Whatever else went differently, pricing accuracy wasn't where any of them slipped.
🏆 The Verdict
Best overall
ChatGPT

Followed the brief exactly, delivered the single-file format asked for, and added no invented product claims. Best visual polish among the tools that actually complied with the request.

Best visual design, if format doesn't matter
v0

The sharpest-looking result of the four — but be ready to unzip a whole project, and double-check anything it adds beyond your brief.

The honest finding here isn't "which tool writes the prettiest landing page" — it's that half of these tools quietly went beyond what was actually asked, either in format (v0) or in invented product detail (Gemini, v0). If you're going to use AI to draft real marketing copy, read every claim it adds that wasn't in your own brief — that's where the risk actually lives, not in the design quality.

Should you switch tools?
Use ChatGPT if you want a real single file today, on-brief, with solid design.
Use v0 if you're already building a real Next.js app and want a deployable starting point — but don't ask for "just HTML" and expect it to listen.
Use Gemini if you're already in Google's ecosystem — but read every feature claim it adds before you ship it.
Use Claude Code if your workflow is already code-first and you're comfortable being the one who checks the copy.

We test a new AI tool matchup every month.

Real files, real numbers, no sponsorships. Subscribe to get the next one first.

Subscribe free →
All tests Weekly briefings Browse AI jobs