The 2026 AEO Playbook

AI referral traffic reached 1.13 billion visits in June 2025 alone, a 357% jump year-over-year. ChatGPT alone now drives 87.4% of that volume. And here's the structural shift most operators haven't internalized yet: the overlap between Google rankings and AI-search visibility has dropped from 70% to under 20%.

Translation: good SEO no longer guarantees you'll be cited by AI search engines. It's a different game with different rules. This post is the playbook we run across our own properties (HBOT Finder, World Wellness Guide, max-ship.com itself) and, in adapted form, on client work.

What AEO actually is

Answer Engine Optimization (AEO) is the practice of structuring content so AI-powered search platforms select it as a cited source when generating answers. Every time someone asks Claude "what's the best HBOT chamber for home use" or asks Perplexity "compare GLP-1 medications", there's an answer engine deciding which sources to cite.

The data on what AI models prefer to cite:

• 74.2% of all AI citations come from structured "Top N" / listicle content
• Pages using 3+ schema types show ~13% higher citation likelihood
• Content older than 3 months sees significantly fewer citations
• AI-driven visitors convert at 4.4× the rate of standard organic visitors

The 5-pillar AEO framework

We've consolidated the framework across our ventures. Five pillars, each one a concrete intervention you can run on your site this week:

1. Schema markup at multiple types per page

The single highest-leverage move. Pages with 3+ JSON-LD schema types get cited 13% more often. For a directory page, that's ItemList + BreadcrumbList + FAQPage. For a product page, that's Product + Offer + BreadcrumbList. Easy lift, real signal.

2. Structured data for extractability

AI models pull data into their answers. Format your specs, prices, hours, addresses, and conditions in tables and structured lists. Avoid burying facts inside prose paragraphs, AI extractors miss them.

3. Comparison tables on every list page

ChatGPT explicitly prefers comparison tables. Perplexity rewards structured comparisons. If your content covers multiple options (chambers, vendors, services), a comparison table at the top is the highest-citation-probability format.

4. Freshness signals

Add visible "Last updated" dates on every page. Update your sitemap with lastmod on real changes. Re-publish key pages quarterly. AI engines weight recency more heavily than Google does.

5. AI crawler welcome in robots.txt

Explicit User-Agent entries for GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, anthropic-ai, cohere-ai, etc. Most sites block these by default with overly broad Disallow rules. Welcoming them is a 30-second fix that compounds over months.

What changed between 2024 and 2026

Two major shifts. First, AI search platforms diverged from Google's ranking signals, the ranking/AI-visibility overlap dropped from 70% to under 20%. Second, AI traffic became conversion-grade traffic. Visitors arriving from ChatGPT or Perplexity convert 4.4× higher than standard organic visitors because they've already been vetted by the AI for relevance.

That changes the math on AEO investment. Even if traditional SEO still drives most of your visits, the AI slice converts so much better that ranking for AI citations becomes the highest-ROI optimization on your site.

The MaxShip implementation pattern

We use the same stack across every venture: Astro for static-rendered pages (fast, schema-friendly), Cloudflare Workers for edge delivery, and a content-hub for centralized authoring. The AEO checklist is embedded in our open-source command folder as a reusable template, every new venture inherits it on day one.

If you want the operating model behind this, including how to set up your own Command Kit and run the Captain's Flywheel, read The Captain's Flywheel. It's the foundation tier of our curriculum.

What changed since I wrote this (updated August 2026)

I published this playbook in May 2026 and then did something slightly embarrassing for an article about measurement: I ran it for two months without measuring it. Here is what happened when I finally did.

What held up

All five pillars survived. Multiple schema types, extractable structured data, comparison tables, visible freshness dates, and explicitly welcoming AI crawlers are all still the right first moves, and they are still cheap. Nothing below contradicts them. The correction is about sequence and expectations, not technique.

1. Build the measurement first. Ours started at zero.

At the end of June 2026 we wired up a citation grid: a bank of specific questions we wanted to win, run on a schedule against each engine, storing whether we were cited. Brand questions ("what is MaxShip") plus category leaderboard questions ("best agentic marketing agency").

The first honest baseline was zero citations out of 36 prompt-and-engine combinations. Two months of doing the right things, correctly, on a young site. Not one citation.

The lesson is not that AEO does not work. It is that a number you can be wrong about is worth more than a checklist you feel good about. Zero is a useful reading. It told us the problem was not on-page technique, which sent us looking somewhere far more productive.

2. The biggest problem was the name, and no schema fixes a name

When we asked the engines our most basic brand question, they answered confidently about an unrelated industrial company that happens to share our name. Not a partial match. A different business in a different industry, described in detail.

That reframed the whole project. We had been treating citation as a markup problem when the binding constraint was an entity problem: the engines did not have a stable idea of what the name referred to. Schema types, comparison tables, and freshness dates do nothing about that. What does: consistent naming everywhere, unambiguous "X is a Y that does Z" descriptions in the places engines actually read, and enough third-party mentions tying the name to the right thing.

3. Narrow and old beats broad and new

On the same day the main brand measured zero, a small directory site of ours was already being cited in its category by the same engines. One narrow subject, a year of depth, nothing clever in the markup that the parent site did not also have.

If you are choosing where to spend the next quarter, depth in one category will get you cited before breadth across five. That is the opposite of how most content plans are drawn up, and it is the single most useful thing the measurement told us.

4. Know where your measurement is blind

One major engine returned empty results through our data provider for the entire first run. If we had not noticed, we would have quietly recorded structural zeros for it and drawn conclusions from a column that was never populated. Any measurement you automate needs a "did this source actually answer" check, separate from "did we score." A blind spot that reports zero looks exactly like a real zero.

5. Where the playbook did pay off immediately

The pillars work best pointed at a specific, checkable failure. The first concrete thing this produced for us was unglamorous: our own site was not the top result for its own name. Fixing that (answer-format headings for the brand question, organization and FAQ schema, an unambiguous definition above the fold) is exactly the kind of narrow, verifiable job the five pillars are good for. Start there, not with a site-wide schema project.

What I would tell someone starting today

1. Write the ten questions you want to win before you touch any markup. If you cannot name them, you cannot measure and you are optimizing blind.
2. Check what the engines currently say about your brand name. Four minutes, and it occasionally changes the entire plan.
3. Then run the five pillars, on your most important page first, not everywhere at once.
4. Re-measure monthly and expect months of nothing. The first non-zero is the signal that the approach is compounding.
5. Go narrow. Own one category completely before you touch the second.