Schema markup is the most over-discussed and under-implemented topic in SEO. Everyone agrees you need it. Almost nobody actually has it set up correctly. In the AI search era, some old schema types matter more than ever, while others have quietly become noise. Here's what to use in 2026 and what to remove.
The 2026 schema landscape
Schema.org has hundreds of types. Most of them were defined for Google rich results in 2015-2020, and many were never widely adopted. Of the ones that did get adopted, only a small subset matters for AI search now.
What changed: AI models read schema differently from Google. Google uses schema to decide whether to show a rich result snippet — a star rating, an FAQ accordion, a product card. AI models use schema to decide what your content actually means. The structured data tells them "this is a product, this is a price, this is a review, this is a step in a process" with machine certainty, instead of inferring from layout.
That shift makes a small set of schema types disproportionately valuable, and a large set of others irrelevant.
The 5 schema types worth implementing in 2026
In priority order. If you only do five things in the schema column of your audit, do these five.
1. Organization
On every page, in your site's root layout or template. Tells AI models what your company is, what its name is, what its logo looks like, where it's located, how to contact it, and what other profiles (LinkedIn, Twitter, GitHub) belong to the same entity.
This is the single most important schema type for AI visibility. When a model is composing an answer that mentions a company, the Organization node is what it pulls from to describe that company accurately. Missing this means the model has to guess.
Minimum fields: name, url, logo, description, contactPoint. Optional but high-value: sameAs (your social profiles), foundingDate, address.
2. Product (if you sell products) OR Service (if you sell services)
On each product or service page. Tells AI models what the offering is, how it's priced, what category it belongs to, and what reviews exist for it.
For B2B SaaS, this is usually SoftwareApplication with nested Offer for pricing. For e-commerce, it's Product with Offer. For service businesses, it's Service with serviceType and areaServed.
The pricing field is the one most brands skip and it's the highest-value one. When ChatGPT is asked "what does X cost?", the Offer.price schema is the most reliable way to make sure it gives the right answer.
3. FAQPage
On any page with genuine question-and-answer content. Tells AI models "these are exact questions, these are the exact answers."
FAQPage is the cheat code of AI extractability. The Q-and-A pair structure is exactly what AI models look for when composing answers. A page with FAQPage schema is dramatically more likely to be cited than the same content in prose paragraphs.
Use it on: product pages with FAQ sections, pricing pages, comparison pages, any page where you anticipate buyer questions. Don't over-use it — putting FAQPage on a page that doesn't actually have a question-and-answer structure will get flagged as spam by both Google and AI parsers.
4. HowTo (or Article with step properties)
On tutorial, walkthrough, and procedural content. Tells AI models "this is a step-by-step process with N discrete steps."
If your content teaches the reader how to do something, HowTo schema makes the steps machine-readable. AI models extract individual steps and surface them in answers to "how do I X?" queries. Without the schema, the AI has to infer step boundaries from heading structure, which it often gets wrong.
Most SMB sites have at least 2-3 pieces of content that should have HowTo schema but don't. Adding it is usually a one-hour project.
5. Article (or BlogPosting) on every blog post
On every blog and editorial page. Tells AI models the publication date, the author, the headline, and the topic.
The datePublished field is the one that matters most for AI search. Models weight recency strongly when composing answers about evolving topics, and an article without a clear date may be treated as "unknown age" and weighted down. The author field also matters: AI models building trust signals look at author identity, and unsigned content scores worse than authored content.
If you have a blog without BlogPosting schema, fixing that is a 30-minute job that pays back for years.
The 3 schema types that have become noise
These were valuable in 2018-2022 and are largely obsolete or low-value in 2026.
1. BreadcrumbList
Google still uses it for some search result formatting. AI models effectively ignore it. The breadcrumbs are inferable from URL structure and navigation anyway. Not harmful, but if you're prioritizing what to implement first, this is at the bottom.
2. Review (standalone, not nested in Product)
When Review was first introduced, standalone review schema on any page could populate star ratings in search results. Google has tightened the eligibility rules to the point where standalone Review schema rarely produces a rich result anymore, and AI models penalize "self-reviewed" content as a manipulation signal. Reviews should live on third-party platforms (G2, Capterra, Trustpilot), not as self-published Review schema on your own site.
3. WebPage / WebSite / SiteNavigationElement
These were over-implemented in the early 2020s when "more schema is better" was the prevailing advice. AI models extract very little from them — they're structural metadata that's already obvious from HTML. Not harmful, but they bloat your JSON-LD and make legitimate signals harder to parse. If you're adding schema today, skip these.
How AI models actually use schema
It helps to know what's happening under the hood. Three things, in order of impact:
First, schema acts as a confidence multiplier. When a model reads "this page says X about topic Y" from prose, that's one signal. When the page also has schema confirming "type: Product, name: X, category: Y," the confidence in the inference jumps significantly. The schema doesn't replace the content — it validates it.
Second, schema enables direct extraction. Fields like Offer.price, Product.brand, Article.datePublished, FAQPage.question can be lifted verbatim into AI answers because they're explicit, machine-readable values. Prose content gets paraphrased; schema fields get quoted.
Third, schema reduces hallucination on your brand. When a model is asked about your company and the only Organization schema it finds has the wrong founding date, the model may emit the wrong date. When the schema is correct, the answer is correct. Most AI hallucinations about brands trace back to missing or wrong structured data.
The 8 AI Invisibility Traps we covered earlier include the most common schema-related mistakes: missing Organization, conflicting JSON-LD blocks, schema that doesn't match the actual page content. Fixing those alone moves audit scores measurably.
The implementation order for a typical SMB site
If you're starting from zero, do them in this order. Each step takes 30 minutes to 2 hours.
- Organization in the root layout. Highest ROI per minute spent.
- Product or Service on your main offering pages. Second-highest.
- FAQPage on your top-traffic page that has Q-and-A content. Often the homepage or pricing page.
- BlogPosting on your blog template. Fix all existing posts in one pass.
- HowTo on your top 2-3 tutorial pieces. Then add it to new tutorials as you publish.
That's it. Five schema types, one site, one afternoon of focused work. The rest is maintenance.
Common mistakes
A few that we see in audits repeatedly:
Over-stuffing. Putting schema on every page for every conceivable type. AI parsers (and Google's validators) flag this as suspicious. Less is more. The five types above, applied correctly, beat fifteen types applied carelessly.
Type mismatches. Using Article schema on a product page. Using Product schema on an editorial post. The schema type should match the actual purpose of the page or the AI parser will discount it.
Validating once and never again. Schema breaks. A CMS template change, a plugin update, a refactor — and suddenly your structured data has a missing required field. Validate every quarter at minimum. Google's Rich Results Test, Schema.org's validator, and most AI visibility tools (including ours) flag schema issues automatically.
Conflicting JSON-LD blocks. Two scripts on the same page describing the same entity differently. The classic version is a layout template emitting an Organization with one set of details and a page-specific block emitting the same Organization with different details. AI parsers can't tell which is authoritative and may discount both.
Schema is unglamorous. It doesn't show up on your homepage. It doesn't impress your customers. But it's one of the two or three highest-leverage technical signals AI models read, and the SMBs that get it right in 2026 will be the ones quietly cited more often than the ones that didn't. Five schema types, done correctly, beat fifty schema types done sloppily. Start with the five.
If you want to know which of these five are missing on your site today and how to fix each, run a free audit here. About 8 minutes, no credit card.