notreally.work
$menu[+]

Claude vs Copilot: same prompt, different outputs

#ai-comparison#claude#copilot#biotics#vibe-coding#experiment

Ran the same prompt cold through Claude and GitHub Copilot. No system instructions. No context. Just:

Build me an interactive explainer on biotics for a general audience.

Hosted both outputs at biotics.notreally.work. No cleanup, no cherry-picking.

What Claude produced (v1)

Four-tab interactive component covering probiotics, prebiotics, postbiotics, and synbiotics. Each section structured identically: formal definition (WHO/FAO 2001 for probiotics, ISAPP 2017/2020/2021 for the others), a plain-language analogy, three mechanism bullets, an evidence grid, a myth-busting callout, and a gut-brain axis section.

The evidence grid is the most considered design decision — three tiers (strong/RCTs, emerging, early/preliminary) color-coded per biotic type. Not binary pass/fail. Not a vague "research suggests." An actual epistemic stance per claim.

The garden metaphor threads all four sections: probiotics are seeds, prebiotics are fertiliser, postbiotics are fruit, synbiotics are seeds planted with their matched fertiliser. Chose a unifying metaphor and committed to it.

The file is named v4_reprise. This was not the first attempt. The first three versions got iterated out of existence — presumably progressively stripping down, restructuring, or rebuilding the component until this one was right. The "reprise" suggests a restart after a direction change, not a linear refinement.

Technical notes: Pure CSS + HTML + JS, no framework. Designed as an embed — uses CSS custom properties (--color-background-primary, --color-text-primary, --font-sans, --border-radius-*) inherited from the host page rather than defining its own. ~20 lines of JavaScript. Clean.

What Copilot produced (v1)

A standalone scroll page. Full HTML document — DOCTYPE, head, body, footer, all included. Self-contained, no CSS variables, no external dependencies. Deploys and renders without a host.

Three sections on probiotics, prebiotics, postbiotics. Synbiotics gets one paragraph. Interactivity is <details>/<summary> — native HTML disclosure elements, no JavaScript. One hardcoded quiz question with the answer already showing.

Garden analogy also present — seeds, fertilizer, harvest. Independent convergence on the same metaphor.

Content is food-examples-first: yogurt, kefir, onions, bananas. No mechanism description, no clinical definitions, no evidence grading. Written for someone who has never heard the word "biotic."

The comparison (v1)

ClaudeCopilot
FormatComponent snippet (no html/head/body)Full standalone document
NavigationTabbed, JS-drivenSingle scroll
InteractivityCustom JS, animated card transitions<details>/<summary>, no JS
Content depthWHO/ISAPP citations, 3-tier evidence grading, mechanismsFood examples, simplified definitions
CoverageAll 4 types in equal depth3 types + synbiotics as footnote
Iterations visiblev4 reprise — at least 4 refinement roundsv1, single pass
Audience assumptionCurious adult who wants the real scienceComplete beginner
Garden analogyYes — extended, each type mapped deliberatelyYes — listed, not developed

Claude interpreted "general audience" as: intelligent adult, no prior knowledge, deserves accurate information. Copilot interpreted it as: simplify everything, keep it short, assume low engagement.

Neither is wrong. They're different editorial decisions about the same three words.

$ diff claude.html copilot.html 190 lines vs 314 lines. Different in almost every other way.

Round two

Follow-up prompt, same cold context:

Make it more visually engaging. Add animations, improve the design, and make it feel more intuitive for a general consumer audience.

What Claude produced (v2)

Complete redesign. Dropped the botanical specimen aesthetic entirely and built a full-screen scroll experience: hero with large Fraunces display type, then four chapter slides — each a full viewport, color-coded, with the biotic name at ~7rem. Detail sections animate in on scroll via IntersectionObserver. Evidence chips stagger in with 60ms delays.

The content is preserved exactly — same WHO/ISAPP citations, same evidence tiers, same gut-brain axis sections. The redesign is structural, not editorial. It reads like a premium wellness app. Dark background, minimal chrome, typography doing the heavy lifting.

Technical notes: Fraunces + DM Sans replacing Playfair + Libre Baskerville. Four distinct radial gradient backgrounds per chapter. All entrance animations via CSS transitions + a single IntersectionObserver loop. No framework. ~20 lines of JS — same budget as v1.

What Copilot produced (v2)

Refined the existing green card layout rather than rebuilding it. Added @keyframes fadeUp to each <section>, staggered via animation-delay. Rounded corners, larger card shadows, slightly improved typography spacing. Kept the same scroll structure and <details> interactivity. Still standalone HTML.

The content is identical to v1. The redesign is purely presentational — same information, softer packaging.

The comparison (v2)

Claude v2Copilot v2
Redesign depthFull architectural rebuildVisual polish on existing structure
NavigationFull-screen chapter slides, scroll-drivenSame single scroll
AnimationIntersectionObserver entrance, staggered chipsCSS fadeUp on all sections simultaneously
TypographyFraunces display + DM Sans bodySame system font stack
Dark/lightDark-first, rich color per chapterLight background, green accent unchanged
Content changesNone — same citations and evidence gradingNone
File size~600 lines~285 lines

Claude treated "more engaging" as a reason to rethink the entire experience. Copilot treated it as a reason to add animations to what was already there.

Both are defensible responses to the same instruction. One is more transformative. The other is more conservative. Neither asked a clarifying question.

user@notreally:~$ _