So here's an attempt to sort what's actually been tested from what just sounds smart.

First, the stuff we can stop doing

A handful of tactics that used to feel like safe bets got quietly disproven this year. Worth clearing these out before anything else, because a lot of checklists still lead with them.

Schema markup was supposed to be the AEO silver bullet. Then a May 2026 study made the rounds showing that adding JSON-LD schema didn't measurably increase AI citations for pages that were already visible in AI Overviews. It's not useless - more on that below - but "add schema, get cited" just isn't holding up.

llms.txt is the other one. Google's own generative-search guidance flat out compares it to the keywords meta tag: nice to have, doesn't move anything. If you run docs or an API reference that developers might paste straight into a chat window, sure, add it - that's a real developer-experience win. As a citation lever, it's not doing anything.

And FAQ rich results are just gone. Google killed them in May. The markup won't hurt you sitting there, but it's not rendering anything visible anymore either.

If your technical SEO checklist still treats any of these three as the main lever, it's solving for a version of AI search that stopped existing months ago.

What's actually blocking a page from being readable by AI

Once you strip the folklore out, the real blockers turn out to be pretty unglamorous.

Crawler access is the one nobody checks first, and should. Half the "why aren't we cited anywhere" tickets we've seen this year trace back to a blanket robots.txt disallow someone added two years ago for a completely unrelated reason, and it's still quietly blocking GPTBot or ClaudeBot along with whatever it was actually meant to stop.

User-agent: GPTBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: Google-Extended
Allow: /

This is also the fastest fix on this whole list - removing a hard block tends to show up within days to a few weeks, since you're not competing for anything, you're just letting the crawler in.

Then check what the crawler actually sees, not what you see. If key content only shows up after client-side JS runs, plenty of AI crawlers never render it. Pull it yourself:

bash

curl -A "GPTBot" https://yoursite.com/your-page | grep -i "your key claim"

Empty result, content visible in your browser? That's the real problem. Not schema, not llms.txt.

Write the answer before the setup. This is the part that's actually named "answer engine optimization," and it's mostly a writing habit, not a technical one. Answer engines pull short, self-contained claims - so put the direct answer right after the heading, before you get into the caveats.

markdown

## How long does technical AEO take to show results?
Crawler-access fixes can register within days to a few weeks.
Content restructuring and entity work typically take one to
three months. Off-site corroboration is a six-month-plus effort.

Heading as the implied question, first sentence as the whole answer, everything after is just backup. It sounds almost too simple to matter, and yet most pages we audit bury the answer three paragraphs deep behind a story about why the question matters.

Keep schema, just for a different reason than you were told. Forget citations. Schema is useful for disambiguation - helping any parser, human search, AI crawler, your own site search, figure out what an entity actually is. Nesting an Article under an Organization with an author Person gives you a cleaner entity graph than three disconnected schema blocks sitting on the same page. And check it periodically. It fails silently. You genuinely will not notice it broke unless you go look.

The slow one: get corroborated elsewhere. Answer engines seem to weight a claim more when it shows up independently across a few credible sources, not just asserted once on your own domain. This one isn't really a dev task at all - it's closer to old-fashioned digital PR and link building. Worth remembering that AEO isn't really "a dev thing." It's dev work plus editorial plus outreach, all pointed at the same set of claims. We've written a bit more about how that split actually plays out across a delivery team, if that side interests you more than the technical half.

The order we'd actually do this in

Crawler access first - it's the fastest win and the easiest to overlook. Then confirm crawlers see what a browser sees. Then fix the answer-first structure on your highest-intent pages. Schema after that, for clarity, not hope. Corroboration last, because it's slow and compounding and nobody wants to start there, but it's the one that actually holds up over time.

Everything outside that order is optimization theater. Everything inside it, at least, has a mechanism behind it.