Google Just Killed FAQ Rich Results: What It Means for Your Schema

Google retired FAQ rich results in May 2026, but FAQPage schema is still valid and there is no reason to strip it from pages whose questions and answers are real. Fast Frigate Digital Marketing, a Vermont SEO agency, reads the evidence this way: AI-cited pages carry JSON-LD roughly three times as often as uncited ones, and no engine has ruled schema out as an LLM reference point. Accurate FAQ markup costs nothing to keep, may still help earn a citation, and only truly fails when it stops matching what readers can see.

Last Updated: September 10, 2026

We were running a routine audit for a political campaign client when we saw it: a small, unceremonious banner at the top of Google’s Rich Results Test in early May 2026. FAQ rich results were going away. Rich Results Test support would follow in June.

We felt the usual spectrum of SEO emotions. Betrayed. Bewildered. A brief moment of panic about client reporting. Then we read the notice again.

Google had taken away a display treatment, not made FAQPage invalid. If a page still has a useful FAQ section and the code matches it, leave the code there. Just do not treat it like a ticket to an AI citation. That is not what Google promised, and it is not what the evidence can support.

Removing a clean FAQ section because Google took away the decoration would be like piercing the toast. The Birdcage reference remains apt. The decision feels dramatic, it makes a mess, and it misses the actual problem.

What Lies Beneath…

• The Display Layer Versus The Data Layer
• What AI Systems Actually Read on Your Page
• What The Evidence Says So Far
• The Dual-Layer Model Needs A Caveat
• What To Do With Existing FAQ Markup
• An FAQ On FAQ Schema?
• The New Rules Of FAQ Engagement

The Display Layer Versus The Data Layer

Search has always had two jobs. One job is to crawl, organize, and index information. The other is to decide what that information looks like on a results page. FAQ rich results sat in both places for a long time. Mark up a real question-and-answer section, and Google could recognize the relationship in the page data and, for eligible sites, render expandable questions under the blue link.

Google retired the second job. Its May 2026 notice said FAQ rich results would no longer show in Search. Search Central later dropped the old documentation. The earlier Google guidance is still useful: site owners can remove unused FAQ structured data, yet Google said there is no need to remove it just because Search is not using it for a visible result.

That is not a claim about rankings, the Knowledge Graph, or AI citations. It is much more limited. Google removed the display feature, and it gave site owners no reason to spend time deleting accurate markup from pages that still contain the matching FAQ.

Most sites had already lost ordinary FAQ rich-result eligibility in 2023, when Google limited the feature to well-known government and health sites. The 2026 change finished the job. According to the deprecation notice, Google removed the search appearance and Rich Results Test support in June, with Search Console API support scheduled to end in August. That is a reporting and display change. It is not a ranking penalty.

What AI Systems Actually Read on Your Page

A useful FAQ begins with a visible question and gives a direct answer before the explanation starts wandering. That format helps a person scan the page. It can help an AI retrieval system find a relevant passage too. The useful object is the answer a reader can see, quote, challenge, and click through to inspect.

This is where the old article needed a correction. FAQPage JSON-LD is not a private hotline to ChatGPT, Claude, Gemini, Perplexity, or Google AI Overviews. No public documentation from those systems says that adding FAQ markup earns a citation. Anyone selling that certainty is selling a story, not an established result.

In a December 2025 experiment, Search VIU’s test placed product facts in visible HTML, JavaScript-rendered content, JSON-LD only, hidden Microdata, and hidden RDFa. During the direct-retrieval scenarios it tested, none of the five systems surfaced content that appeared only in JSON-LD. Visible HTML did the work. That experiment does not settle what happens during indexing, model training, or a search engine’s internal retrieval pipeline. It does settle one practical question: do not hide your best answer in markup and assume a machine will rescue it.

So write the answer for the page first. Put the question in language your customer would use. Answer it in the first sentence. Add the fact, caveat, or qualification that makes the answer honest. Stop. That is better publishing even if no AI system ever cites the page.

For a broader look at the technical side, Fast Frigate’s guide to schema for GEO is a useful companion. Just keep the order straight: real content first, markup that accurately describes it second.

What The Evidence Says So Far

A cited page is usually not a random page. It may have stronger links, a real editorial history, a cleaner technical setup, a known brand, and JSON-LD. Looking at that pile of traits and assigning the win to the markup alone is where people get into trouble.

Ahrefs’ schema study is useful because it did not stop at the snapshot. In its six-million-URL scan, AI-cited pages were almost three times as likely to carry JSON-LD. Then Ahrefs followed 1,885 pages that added JSON-LD between August 2025 and March 2026 and compared them with 4,000 matched controls. The follow-up did not show a dependable citation gain: AI Mode moved +2.4%, ChatGPT +2.2%, and neither result cleared statistical significance. AI Overview citations fell 4.6% relative to the controls. Ahrefs could not tie that decline to the markup, which is the point. Lots of page changes travel together.

Evidence	What It Supports	What It Does Not Prove
Ahrefs’ 1,885-Page Study	Adding JSON-LD alone did not create a measurable citation lift in the studied group.	Schema never helps an uncited page, or every schema type behaves the same way.
Search VIU Test	Visible HTML should carry the answer you want retrieved.	JSON-LD is ignored at indexing or in every search-integrated AI system.
Three-Page Schema Experiment	Schema quality may be worth testing when all other page work is sound.	FAQPage markup causes AI Overview visibility.

There is a counterweight. A small Search Engine Land experiment put strong, weak, and absent schema on three near-identical pages. The page with the strongest implementation was the only one that appeared in an AI Overview and it earned the best organic position. The authors called the result promising, not conclusive. Fair. A three-page experiment and a large observational-plus-controlled study do not cancel each other out. They tell you to avoid sweeping claims in either direction.

The sensible position is less exciting than “schema is dead” or “FAQ markup is an AI cheat code.” Schema may help some systems understand a page in ways we cannot isolate cleanly. It is not evidence-backed as a standalone citation tactic. Visible answers are the part no study in this group asks you to ignore.

The Dual-Layer Model Needs A Caveat

The Fast Frigate model still holds up, with one correction. The page has reader-facing material and machine-readable material. A person sees the question, the answer, the example, and the caveat. The code describes some of that same material in a formal format. Keep those two things in sync. Once the markup starts claiming more than the page delivers, it has become a liability.

The reader-facing material has the better evidence behind it. A plain answer gives a visitor something useful and gives a retrieval system text it can find. Accurate markup may still help a crawler make sense of a page or support a rich result that remains eligible elsewhere. Fine. That does not create a guaranteed path from JSON-LD to an AI citation.

Organic rankings still matter. They just do not explain every citation anymore. In March 2026, Ahrefs found that 37.1% of AI Overview-cited URLs sat in the organic top ten for the same query. That is overlap, not causation. Google can fan a query out into related searches and pull from video, forums, or pages that do not rank for the original wording. Our AI citation guide explains why citation work needs more than a rank tracker and a prayer.

And citations matter even when clicks get thin. Ahrefs’ April 2025 study of 300,000 keywords linked AI Overview presence with a 34.5% lower average position-one click-through rate. Its February 2026 update used another 300,000-keyword comparison and reported a 58% reduction for position one. The studies used different before-and-after windows, and Google’s AI Overview coverage, presentation, and user behavior changed between those periods. Neither number is a universal forecast for your site. Together they point to a less comfortable fact: when Google answers more of the question on the page, visibility and traffic stop being the same measurement.

That does not turn citations into a vanity metric. It means the work has to be measured in more than one place. Brand recall. Branded search. Assisted conversions. Referral behavior. The answer will not fit in one chart. It never did.

What To Do With Existing FAQ Markup

Leave valid existing markup in place. Just do not consider it a shortcut to guaranteed AI citations.

Audit the pages that already have FAQPage markup and ask a boring question first: does the schema match the visible page? If a visible FAQ section has been removed, the answers have changed, or the markup was bolted onto content that never contained real questions, fix it. Or remove it. A stale JSON-LD block is not a clever technical tactic. It is a maintenance problem waiting for someone else.

For new pages, add FAQPage markup only when the page earns it. The questions need to exist on the page. The answers need to answer them. No synthetic sales questions, no awkward keyword exercises, no five versions of “Why choose us?” stacked under a service page. Your audience can smell that stuff. So can an editor.

Do not use Google’s Rich Results Test as a FAQ validation step anymore. Google removed FAQ support from that product because the Search appearance is gone. Use the Schema.org Validator to catch syntax and property errors, then check the page itself. Does the schema match what a reader sees? That is the test that survives a feature retirement.

There is one housekeeping task for teams with automated reporting. Google said FAQ rich-result data would leave the Search Console API in August 2026. Check reporting jobs, Looker Studio connectors, and BigQuery workflows that used the old search-appearance dimension. Do not wait for a dashboard to go blank before finding out somebody built a dependency around a retired feature.

If you want a method for monitoring the broader result, our resource on AI visibility tracking is a better starting point than counting FAQ dropdowns that no longer exist.

An FAQ On FAQ Schema?

The New Rules Of FAQ Engagement

Google owns the display layer. It can add a feature, narrow it, or remove it when the feature stops serving the result page. Building a strategy around rented pixels was always a short lease.

The part you still own is the page. Make the question real. Make the answer useful. Put the answer where a customer, a crawler, a skeptical editor, and an AI retrieval system can all see the same thing. Then use structured data to describe that material accurately, not to smuggle thin copy past a machine.

Keep the markup if it matches. Fix it if it does not. Write better answers either way.

And do not pierce the toast.