If you have spent the last few days reading the reactions and wondering whether to ship a Monday-morning fix, here is the calm version. Keep your schema. Audit one specific kind of page. Ignore the rest. The sections below argue both halves and end with the audit that is actually worth running this week.
Google removed the rich result, not the schema
Google’s Search Central documentation confirms FAQ rich results are no longer appearing in Search as of May 7, 2026. The deprecation is staged. The rich result vanished from SERPs immediately. The Search Console FAQ report and Rich Results Test support both retire in June 2026. API support for FAQPage structured data ends in August 2026. After August, third-party tools hitting Google’s structured data testing API get a deprecation error. The Schema.org spec itself is untouched.
Eligibility had already been narrow since August 2023, when Google restricted FAQ rich results to government and health authorities. For the rest of the web, the rich result has been functionally absent for nearly three years. The May 7 notice is the formal end of a feature that most sites lost in 2023, not a new restriction.
Don’t remove your FAQ schema
Removing FAQPage schema from a site that still has visible Q&A is the wrong call. The rich result has been gone for non-gov/health sites since 2023, the markup still ships to LLMs and Bing, and the engineering work to strip it is better spent on something the user can see.
FAQPage JSON-LD is small. A typical block is one to three kilobytes, served inline with the HTML. There is no rendering cost in the browser, no Core Web Vitals impact, no separate request, no measurable effect on crawl budget. Removing it costs a developer ticket. Keeping it costs nothing. The math is not interesting.
The one exception predates May 7. FAQPage schema must mirror visible Q&A on the page. Schema-only FAQs (markup describing questions that don’t appear to the user) have always been outside Schema.org’s FAQPage definition and Google’s guidelines. The May 7 announcement does not change that rule, it just removes the SERP reward for following it.
Visible FAQs are a separate decision
If there is one thing to keep from this post, keep this. Visible FAQs are a content question, not a schema question. The trigger is user intent: are people asking the question? If yes, the answer belongs on the page. May 7 does not change user intent.
A product page answering “How long is the trial?” or “Does it integrate with Salesforce?” earned its place on the user signal, not on the SERP rendering. The signal is unchanged. Keeping the section is correct. Marking it up with FAQPage JSON-LD is also correct, for the LLM-side reasons in the next section.
FAQs that were padding (questions invented to push a page past a word count target without anyone asking them) should come down. They should have come down before May 7. The deprecation is a useful prompt for that audit, but the underlying reason is unchanged: the user wasn’t asking.
Industry advice was partly right about the destination
Search Google for “FAQ schema for GEO” and you get over 130,000 results. Most argue, fairly, that FAQ schema matters for AI search visibility. The general direction is right, even if it isn’t quite the lift the strongest pieces claim: FAQ schema does feed LLMs. The rich-result framing was the weaker leg, because the rich result had been fenced off to gov and health since 2023, but the part of the argument about AI retrieval is unaffected by May 7.
Two stories were being told as one. Google rich results were one channel. AI retrieval is another. FAQ schema was always more useful for the second than the first, and the same SEO foundations carry over to both. The rich result going away is news for SEO dashboards. It is not news for AEO.
The 130,000 figure is a soft number. It is a Google search result count, not a survey, and it drifts day to day. The order of magnitude is the point. Industry consensus that FAQ schema matters for LLMs is well placed, and May 7 does not invalidate it. Whether it measurably lifts AI citations is a separate question, taken up below.
LLMs still read FAQ schema
Common Crawl ships JSON-LD blocks intact in its WARC extracts WARC (Web ARChive, ISO 28500) is the standard format Common Crawl publishes the web in. Each record is a single HTTP response with metadata, raw HTML and head tags included, which is why JSON-LD survives the trip from page to model. , which is the corpus most foundation models trained on. GPTBot, ClaudeBot, PerplexityBot, and OAI-SearchBot all fetch the full HTML of indexed pages, schema tags included. None of them announced a feature change last week. Whether each system uses FAQPage as a feature in retrieval specifically is undocumented, but the data reaches them either way.
Think of FAQPage JSON-LD as pre-labeled storage. The model doesn’t have to guess which sentence is the question and which is the answer, the markup says so. Three concrete benefits follow. It groups question and answer pairs explicitly, which makes passage retrieval cleaner. It binds each answer to a single question, so the model doesn’t have to infer the relationship from heading proximity. And it survives the HTML-to-plaintext conversion that most extraction pipelines run before tokenization, where unmarked Q&A formatting often loses its structure. None of those properties depend on Google rendering anything.
Adding schema doesn’t measurably lift AI citations
Ahrefs published a difference-in-differences experiment on 11 May 2026 measuring whether adding JSON-LD schema increases citations in AI systems. They tracked 1,885 pages that added schema between August 2025 and March 2026 against 4,000 matched controls. Citation changes after schema was added landed between -4.6% and +2.4% across Google AI Overviews, Google AI Mode, and ChatGPT, statistically indistinguishable from zero.
Schema doesn’t push a page into more AI answers. The content and the signals around it, visibility, authority, topical fit, do that work. Schema is the markup that labels that content so the model can parse it cleanly, useful at the moment of reading, not at the moment of selection. Ahrefs measured selection and schema didn’t move it. That doesn’t disprove the parsing benefit. It rules out the stronger claim that adding schema is itself what gets a page cited. The study has limits the authors flagged (only pages with 100 or more existing AI citations, all schema types pooled, a 30-day window), but those narrow the result rather than overturn it.
For the question on the table, removing FAQ schema you already have, the answer is still no. Adding schema doesn’t lift you. Removing it doesn’t drop you either. Keeping markup that mirrors visible Q&A costs nothing. Removing it leaves the structural rule violated for no measured gain. The action list is unchanged.
The action list is short
Most of the action list is “don’t touch the working schema”. The deprecation timeline gives you months on the measurement side, and the only real work is the audit you should have been running anyway, namely pages with FAQPage schema and no visible Q&A.
- 01
Don't strip your FAQ schema
If a page has visible Q&A and matching JSON-LD, leave both alone. Removing schema that still ships to LLMs has no upside on the AI side of retrieval. - 02
Audit pages with schema-only FAQs
Pull a list of URLs with FAQPage JSON-LD and crawl for visible Q&A. Where the markup describes questions that don't appear to the user, fix it: add the visible Q&A, or remove the schema. The schema-only state has always been outside Schema.org's definition. - 03
Keep visible FAQs that answer real intent
Sections answering 'Does it work with X?' stay. FAQ blocks invented to push a page past a word count target come down on their own merits, not because of Google. The trigger is user intent, not the May 7 notice. - 04
Adjust your dashboards before June
After June 2026 the FAQ Search Console report is gone. If your monthly client deck has a row for FAQ rich result impressions, replace it now with LLM citation tracking. Better to update the metric in May than to discover the old one is empty in your July report.
The May 7 announcement is a SERP feature change, not a content rule change. Google chose to stop showing FAQ rich results. They did not say the schema is wrong, did not say to remove it, did not say it would harm rankings. The rest of the retrieval stack (LLMs, Bing, Perplexity, the open web) kept reading the markup without comment. The rule is unchanged. Visible FAQs that answer real questions, JSON-LD that mirrors them. The work was always for the user and the model. The SERP feature was the bonus, and bonuses come and go.
If you wrote your FAQ blocks for users and your schema for the model, today changes nothing for you, you can close this tab. If you wrote them for the rich result, today is the audit you have been postponing for three years, and the deprecation gives you cover to run it. Either way, the answer to “should I remove my FAQ schema” is no.