Blog · Machine readability

FAQPage and HowTo after the rich-result changes

Published 1 October 2026 · Updated 12 October 2026 · 8 min read

FAQPage and HowTo after the rich-result changes
Photograph: Mitchell Luo / unsplash

The short answer: Both types are still valid schema.org markup and both are still parsed. What changed is the payout. Google removed the How-to rich result in September 2023 and stopped showing the FAQ rich result on 7 May 2026, so neither one earns you extra pixels in the results page any more. Keep the markup if it describes the page honestly. Stop treating it as a visibility tactic, and move the effort to the visible question and answer text, which is the layer Google's own AI features documentation points at.

That is the whole story in a paragraph. The useful part is what follows from it, because most sites did the wrong thing in both directions: some stripped the markup out in a panic, and some kept adding it to pages that never had a question on them.

What was actually removed, and when

The dates are checkable, and the distinction between them matters.

How-to went first. On 14 September 2023 Google removed the How-to structured data documentation outright, saying the rich result was no longer shown on desktop or mobile. The same documentation update narrowed FAQ rich results to well-known, authoritative government and health sites, which quietly ended the feature for almost everyone while leaving the page in place.

FAQ then finished dying in two steps. On 8 May 2026 Google added a deprecation notice saying the feature would no longer appear in Search from 7 May 2026. On 15 June 2026 the documentation itself was removed. Neither FAQ nor How-to appears in Google's structured data gallery today, checked on 12 October 2026.

Neither type was deprecated at schema.org. HowTo is still a current type in schema.org version 30.1, with step, tool, supply, totalTime and the rest intact, and the schema.org page for the type reports usage of 100,000 to a million domains, a figure it attributes to Google's August 2026 aggregations. A Google feature ending is not the same event as a vocabulary term being retired. People conflate the two constantly.

The failure mode nobody warns you about

Here is the one that costs real money. A team reads that FAQ rich results are gone, finds QAPage in the gallery of surviving features, and swaps the type. It validates. It will not make the page eligible for the Q&A rich result.

Google's Q&A guidance is explicit that the markup is for a page with one question and answers that users can submit. The documentation says not to use QAPage markup for FAQ pages or for pages with multiple questions, and names as an invalid case an FAQ page written by the site itself with no way for users to submit alternative answers. The required shape backs this up: one Question in mainEntity, an answerCount, and at least one acceptedAnswer or suggestedAnswer.

So a marketing FAQ cannot become a Q&A page by changing a string. If your page has eight questions and you wrote all eight answers, there is nothing left for it in Google's structured data gallery, which we checked on 12 October 2026. That is a statement about Google. Other engines publish no equivalent eligibility list, so we cannot tell you what they do with either type. Accepting that early saves a quarter of pointless work.

Why keeping the markup is still the right call

Removing valid markup has a cost and little upside. Google's own introduction to structured data says it uses structured data found on the web to understand the content of the page, and to gather information about the web and the world in general, and that properties beyond the documented features may be used to enable future Search features. That is the sourced argument for keeping a block. The unsourced one, which you will hear often, is that unused markup is harmless; we could not find a Google statement saying so, so we are not repeating it. Deleting a valid block throws away a description of the page that something else may be reading.

The something else is the part people underrate. Google's search features are not the only consumer of a JSON-LD block. Other crawlers, aggregators, your own site search and any internal retrieval system you build can all read the same graph, and a FAQPage node is a clean, pre-split list of question and answer pairs. That is a convenient artefact to have lying around regardless of what any one engine does with it.

The honest limit: do not assume a language model is reading your JSON-LD as structured data. The assistants we test do not document how they treat a JSON-LD block, and we cannot see inside one, so we work on the assumption that the visible text is the part that gets quoted rather than on any claim about how a given model parses the graph.

Our broader note on structured data for AI search covers which types still earn their place, and the related piece on sameAs and the identity links engines check covers the one property where the graph genuinely does work that prose cannot.

What to do with the effort instead

Google's own page on AI features is unusually blunt about this. It states that you do not need to create new machine readable files or markup to appear in AI Overviews or AI Mode, and that there is no special schema.org structured data you need to add. It then asks for structured data that matches visible text, which is a different instruction from the one most schema advice gives.

Read that as a redirection, not a dismissal. The lever moved to the visible layer:

  • Write the question as a heading, not just as a JSON string. Systems that split a

    • page into passages are commonly described as cutting at structural boundaries. We cannot inspect any commercial engine's chunker, so treat that as the prudent assumption rather than a documented rule. Either way, a question that exists only inside a script tag is not part of the visible text at all.

  • Put the answer in the first two sentences under that heading. Forty words that stand

    • alone beat two hundred that need the paragraph above them.

  • Keep the markup and the text identical. Google's AI features page asks explicitly

    • for structured data that matches the visible text on the page.

  • For procedures, number the steps in the HTML. Google's How-to rich result is gone.

    • An ordered list of short imperative steps is still readable text with clear boundaries, which is what you wanted the markup to buy you in the first place.

What changes in Google AI Mode reinforces the same point: Google describes those features as working from the content of the page rather than from an extra machine readable layer you add for them.

A check you can run in ten minutes

Take one FAQ page and answer four questions about it.

First, does every question in the JSON-LD appear as visible text on the page. Open the rendered HTML, not the source, and search for the question string. If it appears once, it exists only in the script tag.

Second, does the answer in the markup match the answer on the page word for word, or at least in substance. Paste both into a diff.

Third, does the markup still parse. With the feature gone from Google's gallery an eligibility test tells you nothing, so use a plain JSON-LD validator or the schema.org validator and read the output as a syntax check only.

Fourth, ask an assistant the exact question in your FAQ, a few times, and see whether the answer that comes back resembles yours. Run it more than once, because the same prompt can return different sources from one run to the next. If you want a structured version of this pass, the self-check walks through ten of these on your own site.

Honest limits

Four gaps worth stating, and the first three are hedges we put into the post above on purpose. Every eligibility claim here is about Google, because Google is the only engine that publishes a list of what its features read; we do not know what ChatGPT, Perplexity, Gemini, Copilot or Grok do with a FAQPage node, and we have no way to find out from the outside. We have not sourced the common claim that unused markup is harmless to Search, so we have stopped repeating it. Our advice about headings and passage boundaries is an inference from how retrieval is described in public, not something we have verified inside a live engine. And we cannot measure how much weight a FAQPage node carries once the visible text already contains the same question, because the two signals move together on almost every real page, and we have not found a clean way to separate them without breaking the page.

One more thing this post does not establish: the schema.org vocabulary outliving a Google feature tells you nothing about what the next consumer will want. HowTo survived the removal of the rich result. Whether anything new ever reads it is open, and we have no evidence either way.

FAQ

Should I delete my FAQPage markup? No. It is valid and Google's documentation says properties beyond the documented features may be used to enable future Search features. We could not source the usual claim that unused markup is harmless, so that is not the argument. Delete it only if it describes content that is not on the page, which was a problem before the rich result went away.

Can I switch FAQPage to QAPage to get a rich result back? Not for a normal marketing FAQ. Google's Q&A guidance restricts the type to pages with a single question where users can submit their own answers, and names a site-written FAQ page as an invalid use.

Is HowTo deprecated? Not at schema.org. Google's rich result for it was removed in September 2023, but the type itself is current in schema.org version 30.1, dated 16 September 2026, and still carries its step, tool, supply and time properties.

Does FAQ markup help with AI Overviews or AI Mode? Google says no special structured data is needed for those features and asks that markup match visible text. The practical read is that visible question and answer structure does the work, and the markup keeps your description of the page consistent.

Sources

Related reading

Structured data for AI search: what to mark up firstsameAs and the identity links engines checkGoogle AI Mode: what changes for a publisherWhy an AI answer changes between two identical runsJSON-LD mistakes that quietly break parsingKnowledge graphs, Wikidata and the brand facts engines reuse