Key takeaways
- If a machine can't read, parse and trust your site, it won't recommend you — build for comprehension, not invisible rankings.
- Put key facts in clean, semantic HTML text; never bury them in images or late-loading JavaScript.
- Performance still counts: slow LCP can block your content from AI Overviews.
- Answer both agentic questions — who you are (LLMs.txt) and what you can do (WebMCP).
- Connect your ideas with structured data and meaningful internal links so agents can traverse your knowledge as a graph.
The way machines interact with your site has quietly changed. It's no longer just a search bot indexing pages for a ranking list. Today, AI agents read your content, try to understand what your business does, attempt to verify claims, and — increasingly — try to act on your behalf. When they get stuck, you lose visibility in AI Overviews, assistant answers and the emerging agentic web. The frustrating part: you often can't see it happening.
At Evocative, we treat this as an engineering problem with a business payoff. If a machine can't read, parse and trust your site, it won't recommend you. Here's where agents get stuck most often, and the practical fixes that move the needle.
1. The content is there, but machines can't parse it
Plenty of sites bury their best information inside images, JavaScript-rendered widgets or decorative layouts that mean nothing to a parser. The industry debate over HTML versus Markdown, or whether content sits in subdirectories, matters less than one principle: if it isn't in clean, structured text, it effectively doesn't exist to an AI.
You can't inspect how an AI ranks you the way you audit Google positions. There's no dashboard. So the only reliable strategy is to build what a machine can read: semantic HTML, clear headings, plain-language answers to real questions, and structured data that states your facts explicitly. Stop optimising for a ranking you can't see and start optimising for machine comprehension you can control.
Quick wins
- Put key claims, prices and specs in readable text, not images.
- Use one clear H1 and a logical heading hierarchy per page.
- Answer questions directly in the first sentence of a section.
- Keep critical content out of JavaScript that renders late or fails.
2. Performance quietly blocks discovery
Core Web Vitals didn't become irrelevant with AI. Slow Largest Contentful Paint (LCP) still signals a poor experience, and there's growing evidence that page performance influences whether content surfaces in AI Overviews at all. A machine that times out or renders a broken layout gets a degraded view of your page. Speed is no longer just a UX nicety — it's a prerequisite for being read correctly.
3. The machine doesn't understand how your ideas connect
Reading text is one thing; understanding relationships is another. Google's Open Knowledge Format points toward a future where your site's concepts become a linked graph that agents can traverse — entity to entity, claim to source. The takeaway for site owners is to stop thinking in isolated pages and start thinking in connected knowledge. Internal links with meaningful anchor text, consistent entity naming, and structured data (Schema.org) all help machines assemble a coherent map of who you are and what you know.
4. Agents don't know who you are — or what you can do
The agentic web is splitting into two distinct questions, and your site needs to answer both:
- Identity — who you are. Emerging conventions like LLMs.txt give AI systems a curated summary of your site: what it's about, which pages matter, how to describe you accurately. Think of it as a briefing document for machines.
- Capability — what you can do. Approaches like WebMCP expose your site's functions — booking, searching, purchasing — so an agent can act, not just read. This is where visibility turns into transactions.
Most sites answer neither today. The ones that do will be the ones agents actually use.
5. Editing your site for AI is harder than it should be
All of this requires frequent, safe content changes — and that's a bottleneck for many teams. New tooling is closing the gap: WordPress plugins can now connect AI directly to your site to edit content safely, letting you implement structured answers and metadata at speed. The point isn't to hand your site to a bot; it's to remove the friction that stops good ideas from ever getting published.
The business angle
None of this is about chasing a shiny format. It's about a shift in who your audience is. Increasingly, the first visitor to your site is a machine deciding whether to recommend you to a human. If it gets stuck — on unreadable content, slow pages, disconnected ideas, or unclear capabilities — you're invisible in the exact moment a decision is made.
Fix the readability layer first, then the connections, then the identity and capability signals. That sequence turns your site from something machines struggle with into something they trust, cite and act on.
Sources
- Goossips SEO : HTML, Markdown & Sous-répertoires — Abondance
- New WordPress Plugin Safely And Easily Connects AI To Your Website via @sejournal, @martinibuster — Search Engine Journal
- Google’s Open Knowledge Format Could Work For Websites, Too via @sejournal, @slobodanmanic — Search Engine Journal
- Goossips SEO : LCP & AI Overviews — Abondance
- You Can’t See How AI Ranks You, So Build What It Can Read via @sejournal, @slobodanmanic — Search Engine Journal
- The Agentic Web Is Splitting Into Two Bets: Identity And Capability via @sejournal, @slobodanmanic — Search Engine Journal
