Google killed FAQ rich results: your FAQ schema now only works for AI

What changed in August 2023

Google announced in August 2023 that it would reduce the visibility of FAQ and HowTo rich results, and limit FAQ snippets mainly to well known, authoritative government and health sites. For most websites that ended the visible accordion block in search results. The FAQPage schema did not disappear, but its job changed from triggering a SERP feature to acting as a machine readable signal.

Search Engine Land reported the change as a broad reduction in FAQ rich result visibility, and Google's own liaison replies indicated it was global and applied across site types. Independent visibility datasets from Sistrix and Semrush then recorded sharp declines in FAQ rich result occurrences in the weeks that followed.

So the practical position in 2026 is settled. A typical Australian business site will rarely see its FAQPage markup rendered as a visible FAQ result, and planning content around that pixel is no longer sound.

Why FAQ schema still earns its place

Google's FAQPage documentation, updated after the announcement, is explicit that sites can keep using the structured data because it helps Google understand the page and may enable future features. The markup still has value, just not as a guaranteed rich result.

The larger reason to keep it is AI search. Google's own guidance on AI Overviews states that the systems draw on underlying web content and structured data as inputs when they build a summary. Clean FAQ markup that matches the visible text gives those systems a clear, labelled question and answer to quote. That is the mechanism that matters now, and it works whether or not a rich result ever appears.

Our view at Bushnote is straightforward. Treat the FAQ block as an extraction surface for assistants first, and a search feature a distant second. Build it for the model that will quote it, and the human reader is served at the same time.

How to write FAQs that AI can use

The change in purpose changes how the answers should be written. An answer built for an accordion could lean on the question for context. An answer built for extraction cannot, because an assistant may lift it on its own.

Write each answer so it stands alone. Open with the direct response, keep it accurate, and make sure the on page text matches the markup exactly. Avoid rhetorical phrasing, vague qualifiers and anything that depends on the surrounding section to make sense. Short, plain answers are easier for a model to identify and safer to reuse, which also reduces the consumer law risk that comes with vague or outdated claims.

Apply the schema only where genuine question and answer content appears on the page. Markup that does not match what a reader sees is a liability, not a signal.

What to measure instead

If FAQ rich result impressions have collapsed, they cannot be the measure of success. Chasing them back is not a goal a team can act on.

Measure outcomes that reflect real value. Organic sessions, time on page and on page engagement show whether the content does its job for readers. Assisted conversions and task completion connect it to commercial results. Support deflection matters where the same answers feed a help centre or chatbot. Those figures tell you whether the FAQ work is paying off, in a way a deprecated SERP feature never could.

Build answers that travel

The durable play is to make each answer portable. A canonical question and answer with a stable URL can serve the website, the support desk, an internal knowledge base and the assistants that now summarise the web, all from one source. That lowers the cost of keeping content current and raises the chance it is reused accurately wherever it appears.

For an Australian business deciding where to spend effort, the priority order is clear. Fix the pages with real, repeated questions first, keep the markup honest, write the answers to stand alone, and judge the result by reader and commercial outcomes rather than by a search feature Google has already retired.