May 12, 2026 · SEO

Google killed FAQ rich results last week. I had 231 pages of it. Here's how I stripped them in one pass.

On May 7, 2026, Google quietly retired the last visual FAQ rich result in search. Not "for most sites" — for every site, including the government and health domains that were the last holdouts after the August 2023 deprecation.

Here's the official notice, quoted verbatim from Google's own docs (page updated 2026-05-08):

As of May 7, 2026, FAQ rich results are no longer appearing in Google Search. We will be dropping the FAQ search appearance, rich result report, and support in the Rich results test in June 2026.

If you've been carrying FAQPage JSON-LD on your site for years for the SERP visual benefit, that benefit is gone. The schema still works for semantic understanding, but the rich-result accordion is dead.

I had 231 pages of it

I run an AI tool called HookGenius — a prompt engine for Suno music creators. Over a year of SEO buildout, the site accumulated 231 pages with FAQPage JSON-LD plus 48 pages with HowTo JSON-LD (deprecated even earlier, September 2023). Some of those FAQ blocks were 200+ lines of inline JSON. Across the site, that's roughly 14,000 lines of schema that was no longer doing anything useful.

The temptation was to leave it. It doesn't hurt anything, right? Almost true, but not quite. Three reasons to strip:

Page weight. Each FAQ block adds 2-16 KB of inline JSON-LD. On 231 pages, that's real bytes on the wire. Mobile-first indexing cares about page weight; Cloudflare bandwidth caps care about it; LCP cares about it.

Schema clutter. Every JSON-LD block on the page competes for parser attention. The SoftwareApplication, Article, Organization, and Person markup you actually want AI engines to read works better when the page isn't drowning in deprecated cruft.

Maintenance debt. Schema you stopped caring about becomes invisible. The visible FAQ HTML on the page keeps getting updated; the JSON-LD copy of it gets stale. Within a year you have a machine-readable version that doesn't match what users see — which is exactly the trust signal Google's helpful-content guidance penalizes.

The script that did it

I wrote a small Python script that walks every .html file in a public/ directory, finds <script type="application/ld+json"> blocks whose top-level @type is FAQPage or HowTo, and removes the entire script tag — while leaving every other piece of HTML (including the human-readable Q&A on the page) completely untouched.

The tricky case is @graph arrays — a JSON-LD block can be a graph containing both an Article entity (which you want to keep) and a FAQPage entity (which you want to strip). My script intentionally doesn't auto-mutate those — it leaves them for a manual decision so you don't accidentally lose your Article markup. I had exactly one such case on HookGenius; I fixed it surgically by hand.

After the run on HookGenius:

234 files changed, 256 insertions(+), 14,333 deletions(-)

Live-verified across 5 sample URLs post-deploy:

• 0 FAQPage JSON-LD blocks remaining
• 0 HowTo JSON-LD blocks remaining
• Visible <h2> + <h3> tag count identical before and after
• All 11 visible Q&A phrases I spot-checked still rendered on the page

It's open source

I put the script on GitHub here: github.com/meriabroz/strip-deprecated-schema. MIT licensed. If you maintain a site with FAQPage or HowTo markup that's no longer earning its keep, run it — it's idempotent, so safe to test against your build directory first with --sample 3 before --apply.

The README walks through what it does and doesn't touch. Pull requests welcome if you find an edge case the regex doesn't handle.

What I'm doing with the breathing room

The clean-up was the entry point for a broader audit at HookGenius: stripping legacy schema, expanding SoftwareApplication markup from 33 pages to 221, adding visible "Last updated" stamps across 200+ pages, enriching the Person author schema with knowsAbout fields, allowing 10 more AI-bot user-agents in robots.txt, and a few other tightenings aimed at being correctly cited by ChatGPT, Claude, Perplexity, and Bing Copilot.

The TLDR for any SaaS maintaining structured data in 2026: strip the deprecated stuff, but invest harder in what's left. SoftwareApplication, Article with rich author.knowsAbout, Organization with sameAs, BreadcrumbList — these are what AI engines and Google AI Overview are actively reading. Spend the cycles there.