Google Killed FAQ Rich Results. AI Engines Still Want Them.

Trust Breaks Differently Now

Fake TLS certificates for Google and other major services have been circulating for years, and the whole point is to exploit a shortcut in how people decide what to trust. You see the padlock, you assume the connection is legitimate, you move on. Browsers eventually catch most forged certs, but the underlying weakness never goes away: trust online has always depended on signals that are easy to fake and expensive to verify.

Answer engines run into the same problem, just with content instead of certificates. ChatGPT, Perplexity, and AI Overviews are not reading your FAQ section the way a person does, skimming for tone and giving you the benefit of the doubt. They are scanning for a clean, self-contained unit of information they can lift and cite without introducing an error. Sloppy FAQ content, where the schema says one thing and the visible page says another, or where the answer rambles past the question, fails that test the same way a forged cert fails a browser check. It looks like the real thing from a distance. It does not survive verification.

That is the shift worth sitting with before getting into what actually changed on Google's end.

What Google Removing Rich Results Actually Changed

On May 7, 2026, Google fully removed FAQ rich results from Search. The little expandable dropdowns that used to eat up half the search results page, gone. If you built your FAQ strategy around winning that real estate, you lost the reason you built it.

A lot of marketers read that as a death notice for FAQ schema generally. Why mark up questions and answers in JSON-LD if Google is not going to reward it with a rich result anymore? Pull the schema, simplify the page, move on.

That read misses what the schema was actually doing the whole time. FAQPage markup was never just a styling trick for Google's results page. It was a way of telling any machine reading your page: here is a question, here is where it ends, here is the answer that goes with it, and nothing else is mixed in. Google stopped using that structure to decorate search results. ChatGPT, Perplexity, and AI Overviews never stopped using it to decide what to extract.

The numbers back this up. Q1 2026 benchmarks reported in August showed FAQPage schema still producing real citation lifts, 1.9x for ChatGPT, 2.3x for Perplexity, 3.1x for AI Overviews. Google closed one door. The schema just walked through a different one that was already open.

The Anatomy of a Quotable Answer

So what does a schema entry that survives extraction actually look like? September and October 2026 playbooks keep landing on the same three traits, and none of them are complicated.

The question has to sound like something a person typed into a search bar at eleven at night, not something a marketing team workshopped. "What is the best CRM for small business" beats "CRM Selection Considerations" every time, because that is the actual query an answer engine is trying to match.

One idea per entry. Not three. If your answer starts with "it depends" and then branches into five scenarios, you have written a paragraph, not a quotable unit. Answer engines want something they can lift whole, not something they have to edit down first.

And the answer itself needs to land the direct response in the first sentence, then wrap up inside 40 to 90 words total. Lead with the fact. Follow with just enough context to make it self-contained. No throat-clearing, no "great question" filler, no circling back to the point three sentences later.

Think of it less like writing copy and more like writing a caption someone else will read out loud without editing it first.

Where People Still Get the Schema Wrong

Most FAQ schema failures have nothing to do with the question or the answer. They come from the gap between what the JSON-LD says and what the page actually displays.

Developers update the visible FAQ accordion, fix a typo, reword an answer for clarity, and forget the schema block sitting in the page head still says the old version. Engines check for that mismatch. When the schema text and the rendered HTML do not match exactly, word for word, the entry gets ignored. Not flagged, not partially credited. Ignored. You did the work of writing a clean, quotable answer and the markup never gets read because it failed a basic consistency check nobody on the team was watching for.

This is also where the "FAQ schema is dead" misconception does real damage. Teams hear that Google pulled rich results and conclude the markup itself stopped mattering, so they stop maintaining it. The visible copy keeps getting edited. The schema freezes in place. Six months later every entry on the page is out of sync, and the pages that still bother to keep them aligned are the ones showing up in ChatGPT and Perplexity citations.