Research note · Research note 03 — current edition
Why Clean Content Beats Raw HTML for Most AI Tasks
Written by Aaron Grainger
Independent Content Strategist & Product-Marketing Writer · Published Jun 21, 2025
- Primary audience
- Product managers building AI features
- Also useful for
- AI engineers and technical founders
- Tone
- Authoritative
- Reading time
- 2 min
- Published
- Jun 21, 2025
Abstract
Raw HTML is a rendering instruction set that happens to contain text. This note examines what is lost and gained when a page is normalized to structured Markdown, and identifies the narrow set of tasks where the markup itself is the signal worth keeping.
Key questions
- Which parts of HTML carry meaning a model can use?
- What does normalization discard that later turns out to matter?
- When is raw markup genuinely the right input?
How this note was developed
- — Side-by-side comparison of raw and normalized versions of documentation, article, and listing pages.
- — Review of what downstream tasks needed from each version.
- — Analytical note based on prototype observation, not a measured benchmark.
Central arguments
01Structure survives normalization; presentation should not
Headings, lists, tables, code blocks, and link targets all carry meaning and all translate cleanly. Grid classes, inline styles, and wrapper divs carry none and cost tokens. The normalization question is not how much to strip but which structures to protect while stripping.
02Tables and code are the usual casualties
Aggressive readability extraction commonly flattens a pricing table into an unordered run of numbers, or drops a code block because it lacks prose density. For technical content these are the highest-value elements on the page, so they need explicit preservation rules rather than generic heuristics.
03Metadata is worth more than the last five percent of body text
Canonical URL, title, publication or update signals, and heading path do more for downstream retrieval and deduplication than marginal recall of trailing paragraphs.
04Markup matters when the task is about the page, not the content
Technical audits, structured-data validation, and template detection all need the original document. Those are real tasks, and they are the exception rather than the default.
Limitations
- — Normalization quality varies enormously by extractor and by page template.
- — No benchmark figures are presented here; the observations are qualitative.
Related reading
Related content
Foundational guide · 4 min
How to Turn a Website Into LLM-Ready Markdown
Why clean, structured text often matters more than raw HTML in AI workflows.
Research note · 2 min
The Web Context Problem
Most AI products that reason about the outside world fail for an unglamorous reason: the material they reason over is incomplete, stale, or structurally unusable. This note argues
Research note · 2 min
The Anatomy of a Source-Linked AI Answer
A source-linked answer is not an answer with links appended. This note breaks a well-formed answer into its parts — claim, support, provenance, confidence, and refusal — and argues
Research note · 2 min
The Hidden Maintenance Burden of DIY Web Data Pipelines
The first version of a web-data pipeline is usually a weekend. The cost arrives afterwards, in silent breakage, template drift, and the operational question of who notices when a s
Version history
Current: 1.0 · Published
- 1.0Jun 21, 2025First published.
Was this useful?
Sourceframe is an independent product concept created for research, product-design, and technical-content exploration. It is not an operating company, and nothing here describes a live commercial service. All examples, schemas, and code are illustrative unless a page says otherwise. No client data, customer outcomes, performance results, or partnerships are described anywhere on this site.