What Is Agentic Engine Optimization (AEO)? Plain English

AEO is structuring the content on your site so AI agents can discover it, parse it without choking, and reason over it correctly when answering someone's question. It is not a separate game from SEO - it sits on top. The traditional SEO instincts (good headings, fast pages, descriptive URLs) still matter. AEO adds a handful of things that traditional SEO doesn't bother with because human readers don't need them.

The mental model that helps: imagine your most important page being read in 400 milliseconds, with no scrolling, no clicking, no animation, no JavaScript execution. The agent grabs the raw text, strips the HTML, breaks it into tokens (chunks of words), and tries to extract the answer to a specific question. Then it disappears. None of your analytics tools - Google Analytics, your scroll-depth heatmap, your A/B testing platform - record any of this happening.

Osmani lists five practical things that move the needle, and they are remarkably small:

1. robots.txt. The 1990s plain-text file that tells crawlers what they can fetch. Most site owners set this once for Google and never look at it again. The problem: many sites unintentionally block AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot) with overly aggressive defaults from CDN templates or WordPress plugins. Audit yours. Unless you have a specific reason to block AI bots (copyright concerns, paid-content protection), let them in.

2. llms.txt. A newer convention - a single Markdown file at the root of your site that gives AI agents a structured index of your most important pages with one-line descriptions and (ideally) token counts. Think of it as an XML sitemap rewritten for a reader that wants the summary before the detail. The official spec lives at llmstxt.org.

3. skill.md. For sites with APIs or developer tools, a declarative file that says "here's what this system can do, here are its limits". Useful for agents trying to figure out whether you're worth integrating with. Less relevant for content sites, very relevant for SaaS docs.

4. Clean Markdown structure. Consistent heading hierarchy, semantic HTML, no decorative nesting, predictable code-block formatting. This is the bit that overlaps most with SEO best practices - the difference is that agents are far less tolerant of structural noise than human readers.

5. Token efficiency. Surface the token count of each page somewhere (a tiny metadata field is enough). Osmani's suggested targets give a sense of scale: quickstart guides under 15,000 tokens, individual API references under 25,000 tokens, conceptual guides under 20,000 tokens. Most marketing-heavy pages run 2-5x that.

Three of the five levers do. The other two are niche.

Yes, for everyone:

• robots.txt audit - takes 60 seconds, free, removes the most common reason your site is invisible to AI search.
• llms.txt - takes a couple of hours for a small site, then ~10 minutes per major addition. Genuine upside even for personal blogs.
• Clean Markdown / heading structure - if your CMS is doing it right (Astro, Hugo, Eleventy, modern WordPress themes), you may already be fine. If you have a heavily-customised theme, worth a once-over.

No, unless you fit the niche:

• skill.md - only matters if you offer an API or developer tool. Skip if you're a content site.
• Token-count metadata - useful for documentation-heavy sites, overkill for blog posts.

Three steps, in priority order:

1. Check your robots.txt. Visit yoursite.com/robots.txt. If you see lines like User-agent: GPTBot followed by Disallow: /, you're blocking ChatGPT. Same shape for ClaudeBot, PerplexityBot, Google-Extended (Google AI Overviews), CCBot (Common Crawl). Unless you have a reason, remove the blocks. WordPress plugins like Rank Math and Yoast sometimes ship aggressive defaults; check there too.
2. Add an llms.txt. Create /llms.txt at your site root. Top-level lines for site name + one-sentence description. Then a list of your 10-30 most important pages with one-line descriptions each. The format is documented at llmstxt.org. Most static-site generators have plugins; on dynamic platforms you can hand-roll it in 30 minutes.
3. Stop publishing 5,000-word posts with three banner ads above the fold. The honest finding from talking to people running coding agents and AI assistants: they cope poorly with the standard "content marketing" style of long preambles, repeated keyword stuffing, and decorative section dividers. Tighter writing helps human readers anyway. AEO is the excuse to do it.

No - they are layers, not competitors. SEO gets you into search-engine indexes. AEO gets you usefully read once you're there. The exact mix will shift as AI search takes a bigger slice of total searches, but "rank in Google" remains the primary lever for both (see also the Ahrefs study showing 88% of cited ChatGPT pages came from the search index).

The realistic ratio for a small UK site over the next two years: spend 70% of your time on classic SEO + good editorial, 20% on AEO basics (robots, llms.txt, structure), and 10% on whatever the new entrant turns out to be. Anyone promising you a 50/50 split between SEO and AEO is either selling consultancy or hasn't checked the data.