# The AEO Checklist

> Twelve things that actually help a site get found by people and by models. Distilled from two Promptway field reports. One page if you print it.

Canonical: https://promptway.com/lead-magnets/aeo-checklist

### Make the page fast and small first

Weight, LCP, and a clean accessibility tree still decide whether anyone, human or agent, finishes the visit. A 14 MB homepage with perfect llms.txt is a roof with no foundation.

### Verify Search Console and ship a real sitemap

The verification record has to live on the DNS host that actually answers. The sitemap should list live URLs only, with a real lastmod.

### Ping Bing with IndexNow on publish

Bing feeds ChatGPT search, Copilot, and DuckDuckGo. A publish-time ping turns days of waiting into hours.

### Treat the accessibility tree as a public API

One h1, named links, real alt text. Agents walk this tree. An unnamed logo link cost 33 agentic-browsing points on the original audit.

### Allow the AI bots in robots.txt, on purpose

GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended, and the rest. Point them at llms.txt in a comment. Blocking them and then asking why you are not cited is a contradiction.

### Ship llms.txt and llms-full.txt as text/plain

The spec is explicit. Do not serve them as text/markdown. The index is a menu. The full file is the dump. Generate both from the same content as the HTML.

### Add a page-level script type="text/llms.txt"

Browsers ignore unknown script types. Models can read them. One block that says what the site is, who writes it, and where the machine-readable copies live.

### JSON-LD that matches the visible page

BlogPosting, Organization, WebSite, BreadcrumbList. Escape < in anything you dump into a script tag. Schema is not a ranking cheat. It is how a model knows what the page is.

### Full-content RSS plus a JSON Feed

Summary-only feeds starve every downstream reader, including the ones that are models. Register both in metadata so they auto-discover.

### Canonicals that match what the URL actually serves

If the tag says promptway.com and the host serves a different product, you have taught every crawler to ignore you.

### Write things a model can quote

Cite sources. Put numbers next to claims. Open sections with a crisp definition. That is the GEO evidence that actually moved citation rate, not a new file format.

### Keep the page fresh when the facts move

Update the date. Replace the stale statistic. Perplexity weights freshness hard. A cornerstone that still says 2024 is a citation you are donating to someone else.

## Skip

- Do not shard essays into FAQ pages just so a model can nibble them.
- Do not rewrite prose to sound machine-friendly. Google said not to.
- Do not pile on structured data that the visible page does not earn.
- Do not buy links. Do not expect llms.txt alone to produce citations.