Summarize this article with AI
Send it straight to your assistant and get the short version first. The link travels with the prompt, so the model reads the real page.
Opens in a new tab with the prompt already written. Nothing is sent until you press enter.
The short answer
Machines read structure before style. Headings, structured data, link relationships and plain factual statements are what survive the trip into an answer. Your layout does not.
What actually gets read
Your visible text, your heading hierarchy, your structured data, your links and the relationships between your pages. Not your layout, not your scroll animation, and not what you meant to say.
This is why a beautiful contractor site can be functionally invisible. The craft went into the parts a machine discards.
Headings are an outline, not decoration
An H2 is a claim about what the section beneath it contains. When headings are chosen for visual weight rather than meaning, the outline a machine builds from your page is wrong, and everything downstream inherits that error.
- One H1 per page, stating what the page is
- H2s that name real sections, ideally in the words a buyer would use
- H3s only underneath a related H2, never for styling
- No skipping levels to get a font size you liked
Structured data is a verification aid
Schema does not get you into answers by itself. What it does is let a machine confirm that the entity, location, service and price on the page mean what you intended. That reduces ambiguity, and ambiguity is what gets you skipped.[1]
For a construction company the minimum useful set is an organization or LocalBusiness block with address and service area, a service list, and FAQ markup on pages that carry questions.[2][3]
The extractability test
Take any important fact on your site and ask three questions. Could a machine lift this sentence out? Would it still make sense alone? Could it be attributed back to us?
The trade-off nobody mentions
The structures that maximise extraction, answer-first summary blocks and rows of parallel FAQ answers, are also the strongest signals a page was machine-written. We have had a piece scored as fully AI-generated by a detector when a human wrote every argument in it.
The way through is to keep the inline citations and the clean structure, which is where the value is, keep your own headings and flowing paragraphs, which is where your voice is, and put the extraction scaffolding into the structured data where no detector is looking.
Aligning to how machines read should never mean handing your voice over to them.Jonathan Sterritt
Questions builders actually ask
Does my page builder handle this?
Some do partially. Most generate a heading structure driven by styling choices rather than meaning, which is exactly the failure described above.
Do images matter?
Alt text does, because it is text. The image itself carries far less weight than most people assume.
Is more structured data always better?
No. Accurate structured data that matches your visible content is better. Markup that claims something the page does not say is worse than none.
Sources
Every claim above that could be checked links to the primary documentation rather than to another agency blog. That is the standard we hold client content to as well.
Share this article with AI
Finished it? Hand the whole thing to a model and ask what it means for your business specifically, including anything in it you think is wrong.
Opens in a new tab with the prompt already written. Nothing is sent until you press enter.
