HowTo Schema: Why It Still Works for AI Despite Google's Deprecation

HowTo schema still works for AI, even though Google scrapped the accompanying rich result. The markup no longer produces a visual search result, but it does give language models like ChatGPT, Perplexity and Google AI Overviews explicit, machine-readable structure about the steps in your procedure. In this article you will read what exactly changed at Google, why HowTo markup still holds value for generative engine optimization, and how to use it wisely without wasting your time on decorative markup.

This is part of the complete GEO guide, in which we explain how to get found in AI answers instead of only in classic search results.

What is HowTo schema?

HowTo schema is structured data that describes a step-by-step procedure in a format machines can read directly. It is a type from the Schema.org vocabulary that breaks a task down into separate steps, often with supplies, tools and an estimated duration. You usually add it as a JSON-LD block in the source code of a page.

Where plain text forces an AI to work out for itself what a step is and where it begins and ends, HowTo markup makes that explicit. Each step gets its own field, its own text and possibly its own image or anchor link. It is essentially an invisible layer that says: this is an instruction, these are the steps, in this order.

Important to know: schema markup changes nothing about what your visitor sees on the page. It is meant for machines that interpret your page, not for the reader’s eye.

Did Google really remove HowTo rich results?

Yes, since September 2023 Google no longer shows a HowTo rich result in search results. In August 2023, Google announced that the HowTo rich result would disappear, and from 13 September 2023 it no longer appeared on desktop. On mobile, its display had already been sharply scaled back earlier. The result type is thereby fully deprecated on both devices.

Google went further than just stopping the visual display. The company also removed the HowTo report in Search Console and support for HowTo in the Rich Results Test, and pulled the HowTo documentation from the Search Central pages. Google’s message was clear: the feature was used too little and added too little value for users.

An important nuance from Google itself: you do not need to proactively remove existing HowTo markup. Unused structured data causes no problems for Search, but simply no longer has a visible effect in the classic search results. This deprecation is part of a broader cleanup in which Google scrapped several rarely used structured-data types at once.

The logical question then is: if Google does nothing with it anymore, why should you still care?

So why does HowTo schema still work for AI?

HowTo schema remains valuable because AI models process your content in a fundamentally different way than a classic search engine like Google that shows a rich result. A language model that summarizes or cites a procedure benefits from every explicit structure that helps it recognize the steps correctly and in the right order.

A few mechanisms sit behind this. First, many AI systems read the raw HTML and structured data of a page, even when they see no visual formatting. For a machine, JSON-LD is a clean, unambiguous source: a step is a step, not a paragraph that might be a step. That reduces the chance the model guesses and renders the instruction incorrectly or incompletely.

Second, explicit structure helps with grounding, the process by which AI determines which information is correct and which source it comes from. We explain this further in grounding: how AI determines what is true. The clearer and more consistently you supply a procedure, the more reliably a model can adopt your version instead of a competitor’s.

Be honest, though, about the limit of what we know for sure. AI providers rarely publish exactly which signals their models weigh, and HowTo markup is not a guaranteed citation boost. What we can say with confidence: explicit structure can never confuse an AI, and can only make extraction easier. It is a small, low-risk addition, not a magic bullet.

What does the markup deliver versus the underlying structure?

The biggest gain is not in the JSON-LD block, but in the step-by-step text structure that lies beneath it. A page with a crystal-clear procedure in the visible text, with numbered steps, descriptive subheadings and self-contained instructions, almost always performs better with AI than a page with perfect markup on top of an unreadable mush.

That is because most AI systems primarily process the text itself. The markup is a useful extra signal, but it is the content on the page that gets cited. So if you have to choose where to invest your time, choose the text first: short, standalone steps, each starting with the core action, in a logical order.

A handy working order:

• Write the procedure first as clear, numbered steps in plain text.
• Give each step its own line or short paragraph that still holds up when cut loose.
• Only then add HowTo markup that mirrors exactly those same steps.
• Always keep markup and visible text in sync, so they confirm each other.

This approach fits the broader principle of content architecture for AI extraction: build pages that a machine can navigate effortlessly and cite in pieces.

When should you use HowTo schema and when not?

Use HowTo schema only for pages that describe a real, linear procedure. It is meant for instructions with clear steps you carry out in order: an installation, a configuration, an application process. On that kind of content, the markup reinforces what the text already does.

Do not use it as decorative markup on pages that are not step plans. Forcing a product page, a general explanation or an opinion piece into a HowTo structure delivers nothing and makes your markup unconvincing. For question-and-answer content, a different approach fits, which we discuss in what is AEO.

A few practical guidelines:

• One procedure per page works better than ten half-finished step plans mixed together.
• Describe concrete actions, not vague advice, because an AI prefers to cite something testable.
• Mention realistic preconditions, such as supplies or prior knowledge, instead of suggesting everything works in three clicks.

That sober attitude fits how we look at GEO: no promises about guaranteed citations or positions, but deliberate choices that increase the chance you get found and cited correctly. For B2B companies that publish instructional and knowledge content, that is often an underrated lever. Want to know how this fits into a broader approach? Then read our service page on generative engine optimization (GEO).

The short summary

Google pulled the HowTo rich result from search results in September 2023 and wound down its support, so for classic SEO the type has played out. For AI the story is different: HowTo schema and especially the underlying step-by-step structure help language models extract and cite your procedure correctly. Treat it as a small, low-risk addition on a page whose content already holds up, use it only for real step plans, and keep markup and text in sync. That way you get value from a feature Google left behind.

Want to know which structured data and content actually move you forward in AI answers? Book your free intake and we will look together at where the gain lies for your site.