Most schema markup guides stop at “add LocalBusiness and call it done.” That advice was fine for 2022 SEO. It is not enough for 2026 AEO. AI engines read structured data differently than Google's ranking algorithm does, and the schemas that move citation visibility are not always the ones with rich-result icons attached.

In May 2026, Google removed FAQ and HowTo rich results from search results entirely (Google Search Central). A lot of SEO teams interpreted that as a signal to strip the schemas. They were wrong. The rich result is gone. The machine-readable signal is not. AI engines still parse FAQPage and HowTo markup, and our scans show they cite content marked up with them at a meaningfully higher rate than unmarked equivalents.

This is the deeper guide. Which schemas to prioritize, the specific properties that matter, and the implementation mistakes that get businesses downranked or, in one case, exposed to federal penalties.

The hierarchy of AEO schema impact

Not every schema is equal. Based on what our AEO Grader sees across thousands of professional services sites, schemas fall into three tiers by AI-visibility impact:

  1. Tier 1 (citation drivers): FAQPage, HowTo, Article. These are the schemas AI engines reach for when constructing answers. Q&A format and step-by-step format both mirror the shape of AI conversation.
  2. Tier 2 (entity authority): LocalBusiness, Service, Person, Organization. These tell AI what you are, where you are, and who is behind the work. Required for recognition; not sufficient for citation.
  3. Tier 3 (trust signals): AggregateRating, Review, WebApplication. Adds quality and credibility signal. One of these (AggregateRating) is a legal minefield if implemented carelessly.

Most sites we scan get Tier 2 partially right and skip Tier 1 entirely. That is the gap.

FAQPage: the highest-leverage AEO schema

FAQPage is the single highest-leverage schema for AEO citation, full stop. The format is identical to how users ask AI questions, and AI engines extract the answers directly into their responses.

Tactical guidance for FAQs that get cited:

  • Write the question exactly as a buyer would type it into ChatGPT. Not “Our Implant Process,” but “How long does a dental implant take?”
  • Keep answers 2-4 sentences. Long enough to be substantive, short enough to lift cleanly.
  • Include a specific number, range, or price in the answer when possible. AI engines preferentially cite answers with quantified specifics.
  • Match the FAQ schema content word-for-word to visible page content. Schema-content mismatch is a downranking signal in both Google and our AEO scans.
  • Limit to 5-8 FAQs per page. Bloated FAQ schemas (30+ questions) get partially ignored.

HowTo: the cost-calculator and service-guide schema

HowTo schema is for any page that walks a reader through a process. Cost breakdowns, service explanations, “how to choose a…” guides. AI engines pull HowTo content when users ask process questions, which they do constantly.

The properties that matter most are step (each step as a HowToStep object), totalTime when applicable, and estimatedCost for any cost-related guide. The cost property in particular is gold for AEO: questions like “how much does X cost in Y” are among the most frequently asked AI queries, and HowTo with an estimatedCost is the cleanest structured answer available.

LocalBusiness: the properties AI actually parses

Adding LocalBusiness with just name, address, and phone gets you to baseline. Citation-grade LocalBusiness uses five additional properties:

  • geo with explicit latitude and longitude coordinates, not just a postal address.
  • areaServed as an array of every city or neighborhood you serve. AI engines match this directly when users include a location in their query.
  • hasOfferCatalog linking to your Service entries. This is how AI builds the “what does this business actually do” picture.
  • knowsAbout as an array of expertise topics. Underused; we see strong correlation with citation rate.
  • makesOffer for specific paid services with prices when public.

Use the most specific LocalBusiness subtype available. Dentist, LegalService, MedicalBusiness, FinancialService all exist on schema.org. The subtype tells AI engines what category to recommend you in, which matters more than the parent type.

Service schema for multi-service practices

If you offer five or more distinct services, each should have its own Service schema with serviceType, provider (pointing back to your LocalBusiness), areaServed, and an offers block with price when applicable.

Nest Service entries inside the LocalBusiness hasOfferCatalog, do not leave them orphaned on individual service pages with no parent reference. AI engines build entity graphs; disconnected schemas read as disconnected entities.

Person schema for individual practitioners

For attorneys, doctors, financial advisors, and any business where the practitioner is the product, Person schema is an E-E-A-T signal AI engines read directly. The properties that matter:

  • jobTitle with the credential spelled out: “Board-Certified Estate Planning Attorney,” not just “Attorney.”
  • alumniOf for credentialed schools (law school, medical school).
  • memberOf for professional associations (state bar, ABA, ADA).
  • sameAs linking to LinkedIn, state bar profile, professional association directory.
  • worksFor connecting back to the LocalBusiness entity.

Person schema with these properties materially affects whether AI cites the practitioner by name in a recommendation versus just the firm.

AggregateRating: the FTC liability schema

AggregateRating is the schema that displays your review count and average. It is also the schema most commonly fabricated, and that has changed from an ethics problem into a federal compliance problem.

The FTC's Consumer Reviews and Testimonials Rule went into effect October 21, 2024. It carries civil penalties of up to $51,744 per violation for, among other things, “fake indicators of social media influence” and review counts that misrepresent actual customer reviews. Hardcoding reviewCount: 247 into your schema when you have 12 actual Google reviews is now a federal liability, not a growth hack.

The defensive pattern is to gate AggregateRating behind a feature flag tied to real-review verification. If your actual review count is below your honesty threshold, the schema builder returns null and emits no markup. Never ship default review counts. Never use placeholder values. AI engines also cross-check AggregateRating against public review sources and downrank mismatches.

Article schema for blog posts

Article schema (or its subtypes BlogPosting, NewsArticle) is how AI engines identify and weight your content. The properties that affect citation are author (pointing to a Person entity, not just a string), datePublished, dateModified, and about for topic specificity.

Your blog is what AI quotes when answering substantive questions. Unmarked blog posts get treated as undifferentiated text. Properly marked posts get cited with attribution.

WebApplication and SoftwareApplication for tools

If your site has a calculator, configurator, or interactive tool, mark it up as WebApplication or SoftwareApplication. AI engines specifically surface these when users ask “is there a tool that…” questions. Required properties are name, applicationCategory, and operatingSystem (use “Web” for browser tools).

Common implementation mistakes

  • Schema-content mismatch. The schema says one thing, the visible page says another. AI engines catch this and treat it as a deception signal.
  • Orphan entities. Service schemas with no provider reference, Person schemas with no worksFor, FAQs with no isPartOf. Entities need to connect.
  • Missing required fields. Each schema type has required properties per the schema.org spec. Missing required fields can invalidate the entire block.
  • Multiple LocalBusiness blocks competing. If you have one location, ship one LocalBusiness schema. Two competing blocks confuse AI parsing.
  • JSON-LD inside a script tag with the wrong type. It must be <script type="application/ld+json">. Anything else and parsers skip it.

How to validate

Two tools, two purposes:

  1. Google Rich Results Test checks whether your schema is eligible for Google rich results. Use this for Google-specific validation.
  2. Schema.org Validator checks whether your schema conforms to the schema.org spec, independent of Google. Use this for general structural validation, especially for schemas (FAQPage, HowTo) that no longer trigger rich results but still feed AI engines.

Pass both. Then run a free AEO scan to see whether your schema is actually translating into AI citations. Validation confirms the markup is correct. The scan tells you whether it is working.

Where to start

For a typical professional services site, the install order that produces the fastest citation lift is: LocalBusiness with the five properties above, then a FAQPage on every service page answering the top five buyer questions, then Person schema for each practitioner, then Service schema for each offering. Article schema gets added to blog posts as they publish. AggregateRating gets added only when real review counts justify it.

If you do not know what your schema currently looks like to AI engines, that is the right place to start. Our AEO Grader runs your site through 60 real prompts across ChatGPT, Claude, Gemini, and Perplexity and shows which schema gaps are hurting your citation rate. Free, 60 seconds, no account.

Vertical implementation guides: see AEO for dentists for Dentist + MedicalBusiness schema patterns, and AEO for law firms for Attorney + LegalService schema patterns.