Does Google still use JSON-LD schema in 2026?
Yes, but not the way most people think. JSON-LD is a Rich Results eligibility gate and an entity-clarity signal, not a ranking lever. Schema tells search engines what a page is. It does not tell them the page deserves to rank. Google has been consistent on this point since 2015.
What changed in 2026: AI engines (ChatGPT, Perplexity, Claude, Google AIO) parse a smaller subset of Schema.org types for entity resolution. AI engines parse several schema types, with Article, FAQPage, HowTo, Organization, and Person showing the strongest signal in 2026 studies. JSON-LD is one of five GEO levers (entity, content structure, citation density, freshness, schema) per the GEO playbook. Schema does the labeling work that helps an AI engine identify your brand as an entity. Brand mentions and editorial links do the heavier work of earning citation odds.
Which JSON-LD schema types should I add to my site first?
Order matters. Ship in this priority:
- Organization on the homepage, with a sameAs array pointing to LinkedIn, Crunchbase, Wikipedia, and X. This is the entity-grounding schema.
- Article on every blog post, with a real author, dateModified, and a 1200x630 image (not an SVG).
- BreadcrumbList site-wide on any nested page (blog posts, category pages, product pages).
- FAQPage on cornerstone Q&A pages where the questions are real, not fabricated to game Rich Results.
- Person on author bio pages with knowsAbout and a sameAs array.
The Organization schema is the most leveraged of the five because it sets the entity that every other piece of schema references. Pair it with off-site signal: a sameAs array of LinkedIn + Crunchbase + Wikipedia is only useful if your brand actually appears on those properties. That requires placement work, not schema work.Off-site brand mentions remain the strongest AI-citation lever regardless of how clean your JSON-LD graph is.
How is JSON-LD different from llms.txt and robots.txt?
Three layers, three jobs:
- robots.txt is the permission layer. It tells crawlers (including GPTBot, ClaudeBot, PerplexityBot) whether they are allowed to fetch a path.
- llms.txt is the curation layer. It tells LLMs which canonical URLs to prioritize, once they are allowed to fetch.
- JSON-LD is the entity layer. It tells engines what each page is, in machine-readable form, once they have fetched it.
You need all three. Schema does not bypass robots.txt; if you block GPTBot, no amount of Organization markup will make ChatGPT cite your site. A clean JSON-LD graph also benefits from a matching llms.txt: pair this with an llms.txt file so LLMs know which canonical URLs to surface once your schema has told them what each URL is.
What JSON-LD mistakes will hurt my rankings?
Fabricated aggregateRating and reviewCount are the biggest spam-action risk in 2026. Google's 2024 and 2025 spam policy updates explicitly target review schema abuse. Manual actions for fake ratings show up in Search Console under "Spammy structured markup". Four high-risk patterns to avoid:
- Fake aggregateRating or reviewCount. Never emit these unless you have a working review system feeding real data into the markup.
- Product schema on non-product pages. Service pages, marketplaces, and SaaS landing pages should use Service, not Product. Mismatched types are treated as spam.
- author.url pointing to a non-existent route. If your author bio page does not exist, leave author.url blank. Googlebot does 404 the URL.
- SVG image on Article schema. Googlebot does not accept SVG for image fields. Use PNG or JPG at 1200x630.
Beyond the spam-action risk: schema does not earn links by itself. The real lift on AI citation rate and SERP rankings comes from third-party publishers naming your brand.Earn schema-rich backlinks from real publishers to populate the entity graph that your Organization schema points at.