A service page names one author, while its code names another. A product price has changed, but the markup still carries the old amount. The website looks orderly to visitors while its machine-readable information disagrees with the content. Another schema plugin will not solve an unclear data source.
In brief: Schema describes objects and relationships; JSON-LD is a format for expressing them. Start with accurate content and consistent data. Valid markup does not establish an author’s qualifications or guarantee an AI citation.
Canem Errant compares visible information with structured data: provider, author, modification date and current terms. That produces specific corrections. Changes in mentions, clicks or enquiries require separate observation afterwards.
What do structured data tell a search engine?
Text explains the business and offer. Schema identifies objects and properties such as an organisation, article author, price or business address. Stable @id values can connect related objects instead of several templates describing the same company as different entities.
JSON-LD separates those details from the visual layout. Google also supports Microdata and RDFa, while generally recommending JSON-LD for easier implementation and maintenance. That does not establish that every AI parser reads or prefers it. Google’s structured data introduction.
A RAG system retrieves relevant documents as context for an answer. Author, organisation and period help explain the material. A qualification entered by the publisher remains a publisher claim. Adding a university and job title to Person does not itself establish E-E-A-T.
Generative AI features in Google Search require no special schema markup. Google’s guidance does not automatically describe Perplexity, ChatGPT or every other system. Google’s AI optimisation guide. Our GEO guide covers the broader process.
Eight schema groups, chosen by content
Choose the type for the actual object. Every page does not need every group.
Organization: who provides the service
Check the name, website, logo and supported company information. A template must not invent a legal form. Reused information needs a maintained source.
sameAs refers to a page that unambiguously identifies the same entity, such as an official profile. A link to an unrelated page on a well-known platform is not identity verification. Schema.org on sameAs.
Person: who takes responsibility for the material
Names, jobTitle, hasCredential, alumniOf and professional profiles must reflect actual facts. Do not invent education or certification for apparent authority. Verifiable responsibility is particularly important for medical, legal and financial content.
If an author has no public description, explain their real work first. JSON-LD does not replace sources or editorial accountability.
Article, BlogPosting and NewsArticle: what was published
These types connect the headline, author, publisher and dates. datePublished records publication; dateModified records a real revision. Updating the date daily without editing the article creates a misleading impression of freshness.
BlogPosting fits a blog article; NewsArticle fits an appropriate news item. Not every promotional page is news. Check that the visible headline and author match the markup.
FAQPage: visible questions and answers
FAQs should address genuine questions about work, data requirements and limits. Write useful answers first, then check their markup. Hidden promises do not belong in JSON-LD.
Google stopped displaying FAQ rich results on 7 May 2026. FAQPage remains in the schema vocabulary, but is not evidence of a current Google result feature or preferential use by Perplexity and ChatGPT. Google Search FAQ change.
HowTo: a real sequence of actions
HowTo describes a process and steps. A HowToStep must match a visible action; images should help someone carry it out. A benefits list does not become an instruction merely by receiving this type.
Semantic description and support for a particular search appearance are separate questions. An image at each step does not establish priority in AI answers. Schema.org HowTo.
Product and Offer: the item and its terms
Check the product, brand, offer, price, currency and availability. Variants and markets may have different conditions. Visible product information and markup should use an agreed source.
aggregateRating requires suitable real ratings and compliance with the relevant search feature’s rules. Inventing an average is not optimisation. Having Product markup does not establish correct offers or stars in search results.
LocalBusiness: a real local operation
Name, address, phone and hours must describe the actual business. Compare the website, markup and Google Business Profile. First establish whether a difference reflects a move or distinct branches.
An extra location in the code does not create an office. Language and location should describe real business information rather than a list of desired markets.
Dataset: traceable structured information
Dataset describes a structured collection of information. Research needs a method, period, authorship and, where appropriate, a publishable data file. A single percentage in a sales paragraph is not a documented dataset. Schema.org Dataset.
Check quality and publication rights beforehand. Markup does not anonymise customer data. Dataset cannot validate unsupported statistics or guarantee a credit when someone uses a number.
Which three errors should you fix first?
Inconsistent business identity. One template names a brand while another adds an incorrect legal form. Verify names and @id, then correct the data source. Copying a global snippet may simply distribute the error to every page.
Unsuitable sameAs links. Remove unrelated or invalid profiles. Each link must identify that company or author. There is no universal requirement for at least three links.
Invented author credentials. Qualifications in a demonstration are not facts about your writer. Remove unsupported properties and explain the author’s real background.
Google requires current information that matches visible content. Correct syntax does not justify a false price or rating. Structured data policies.
How should you implement it in your system?
Astro and other HTML-generating systems
JSON-LD can be generated from article or catalogue data. Inspect the resulting HTML, not just the source object. A build check can detect syntax problems but cannot establish a qualification or determine whether a price is current.
Maintain organisation and author data, with article-specific information for each page. Fewer manual copies make upkeep easier, but editorial review is still needed after changes.
WordPress
First identify what the theme and existing plugins such as Yoast or Rank Math already output. Another plugin, Schema Pro or custom code should address a specific gap. Yoast’s schema implementation guide.
Custom code is not automatically better. A second generator may duplicate correct information. Changes to functions.php need a suitable approach to theme updates and rollback. Consider maintenance and cost before changing the setup.
Shopify
Compare theme, app and Liquid output in the rendered product page: variants, currency, stock and real reviews. An app should close an identified gap rather than add more markup for its own sake.
After a product change, review visible content and structured data. Costs include maintenance and consistency across sources, rather than only an app subscription. Shopify’s SEO guidance.
How can you assess the result?
Record the URLs, questions, objects and contradictions before editing. Repeat the same review after a bounded change. Distinguish a brand mention, source link, website visit and qualified enquiry.
A web server log records a URL request, not proof of JSON-LD parsing, trust or citation. A user agent can be spoofed. A server-side tag manager is also different from a web server access log.
Server-side tracking may support event measurement, but it does not automatically recover missing referral information or prove schema ROI. If text, prices and links change together, the outcome cannot confidently be assigned to markup alone.
Summaries, tables and blockquotes may help readers. A supposed “Fact Density Score” is not an established public scoring rule for Perplexity or Google AI Overviews. Bold percentages do not replace a method and source.
Two practical review scenarios
Consultancy: Compare organisation, author and FAQs against visible information. Then observe source links for the same questions. This is a review scenario, not a verified client success within three weeks.
Furniture store: Product and Offer must agree with the current price, stock and real ratings. Then observe search presentation, clicks and orders. Consistent data guarantees neither AI Overview stars nor a CTR increase.
What can you do this week?
- Check output: use Schema Validator for vocabulary and Rich Results Test for supported Google features. A passing result does not prove facts or AI inclusion. Start with the homepage, important articles and service pages.
- Provide real answers: collect, for example, ten customer questions for three key service pages. This is a working scope, not a ranking formula. Questions and answers must be visible and consistent with the markup.
- Connect identities: review Organization, Person,
@idand genuine sameAs profiles.llms.txtis a voluntary specification for suitable consumers, not a Google requirement or a promise that every AI system uses it. The llms.txt proposal.
Acceptance should identify the specific information corrected. It cannot establish a fixed percentage increase in AI citations.
Frequently Asked Questions
Does schema replace on-page SEO?
No. Accessible, useful text, sources and internal links remain important. Schema describes the content rather than replacing it. Google requires no special markup for its generative AI features.
Can I use Microdata or RDFa instead of JSON-LD?
Google supports all three. JSON-LD is generally recommended for easier maintenance. Assess correct output and upkeep, rather than an assumed universal speed ranking of AI parsers.
Which types matter for B2B?
Organization, Person and appropriate article data are often useful. Visible FAQs can address customer questions. Choose by content; no combination guarantees more citations.
How quickly will AI systems see an update?
There is no universal period. Retrieval, search updates and selection for an answer are different processes. Observe access and specific answers without promising hours or weeks.
Is a WordPress plugin enough?
It may be enough if the output is complete, accurate and consistent. Custom code should address a demonstrated gap and adds maintenance work.
Why does Perplexity not cite my FAQ?
A missing citation does not identify the cause. Review access, relevance, sources and technical errors. An audit can establish concrete problems, but cannot prove the system’s undisclosed selection decision.
Can citation coincide with fewer clicks?
Yes, someone may get an answer without visiting the site. Compare visibility, visits and enquiries; timing alone does not prove a cause. Do not hide material information to force a click. Explain the value of reading the complete document.
Discuss a review of your content and structured data with us. An introductory conversation clarifies the task; the GEO diagnostic is agreed with specific URLs, questions and verifiable outputs.
