Technical SEO
Rich Semantic HTML5 vs. Div-Soup: How Clean DOM Structure Improves Crawl & AI Extraction
, 3 minute read
Semantic HTML5 uses elements such as main, article, section, nav, dl and table to state what each part of a page is. Div-soup wraps the same content in anonymous containers. Crawlers and language models extract headings, answers and relationships more reliably from semantic markup, because the structure itself separates content from layout.
Read your page the way a machine does
View the source of a site built with a visual page builder. Around a single paragraph you will often count fifteen or twenty nested div elements, each with a generated class name. To a browser this is fine. It paints the same pixels either way.
A parser is not looking at pixels. It is trying to answer questions. Where does the main content start? Which text is a heading, and which is a styled paragraph pretending to be one? Is this list of items a navigation menu, a set of steps or a pricing table? In div-soup the honest answer is that it cannot tell, so it guesses.
Google has spent two decades getting good at guessing. Newer systems that retrieve passages for AI answers are less forgiving, and they are increasingly the ones reading.
What the elements are for
The HTML standard defines sectioning and grouping elements precisely so that software does not have to guess:
mainholds the content unique to this page. One per page.articleis a self-contained piece that would make sense on its own.sectionis a thematic grouping, and it should have a heading.navmarks navigation, so link lists are not mistaken for content.asideis related but not essential.headerandfooterintroduce and close their nearest section.
Inside those, a few underused elements carry a great deal of meaning. A dl with dt and dd states that these are terms and their definitions. A table with th scope states which cells are labels. figure and figcaption tie an image to its description. details and summary make an accordion whose answer is in the document whether or not it is open.
Before and after
Here is a question and answer as a typical builder emits it:
<div class="e-con-inner"><div class="widget-wrap">
<div class="accordion-title" role="button">How long does a roof last?</div>
<div class="accordion-content" style="display:none"></div>
</div></div>
The answer is not in the HTML at all. A script inserts it when someone clicks. Now the same content written properly:
<section id="roof-lifespan">
<h2>How long does a roof last?</h2>
<p>An asphalt shingle roof typically lasts 20 to 30 years...</p>
</section>
The second version has an addressable section, a real heading and an answer in the first paragraph. A crawler can index it, a link can point at #roof-lifespan, and an answer engine can lift the paragraph. This is the pattern behind direct-answer engineering.
Rendering is not guaranteed
Google processes JavaScript, but crawling and rendering are separate queues. A page can be fetched today and rendered later. Many other crawlers, including several used by AI products, do not execute scripts at all. Content that exists only after JavaScript runs is content some readers never see.
The test is simple. Fetch the raw HTML, without a browser, and search it for a sentence from the middle of the page. If it is not there, it is not reliably there for crawlers.
Server-side rendering fixes this at the source. It also fixes the performance problems that come with shipping a framework to render a paragraph, which is part of our work on crawl budget and Core Web Vitals.
Semantics and schema work together
Structured data describes entities. Semantic HTML describes the document. Google's guidelines require that markup reflect what is visible on the page, so the two have to agree: the FAQ in your JSON-LD should be the FAQ in your details elements, and the sections listed in a WebPage node should be real sections with those IDs. We build them from the same source, as described under JSON-LD architecture.
A short audit you can run today
- Count the
h1elements. There should be one. - Check that headings descend in order, with no jump from
h2toh4. - Confirm the page has a
main, and that navigation sits innav. - Find every question on the page and confirm its answer is in the raw HTML.
- Give each major section an
idand link to it from somewhere.
On a desktop screen, the Inspect Live Schema & Semantics button on this site runs the same check against whatever page you are reading. The full method is on our semantic HTML5 page. It matters most where pages have to be found by specification, as in RFQ engineering for manufacturers.
Questions this article raises
Does Google penalize div-heavy pages?
No. There is no penalty. The cost is indirect: weaker extraction of headings and answers, slower pages, and content that some crawlers never see.
Is semantic HTML a ranking factor?
Not as a direct signal. It improves how accurately your content is understood and how often passages are selected for snippets and AI answers.
Can I fix this without rebuilding the site?
Often yes. Most builders let you set the HTML tag for a container and output real headings. The biggest wins are usually in the main templates, not on every page.
Talk to the person who will actually do the work
No account coordinator, no discovery deck. Call and describe what the phone should be ringing for, or send the site and we'll trace where the signal is leaking.