Which Schema Markup Tactics to Avoid for AI Search Success

What changed in Schema Markup guidance for AI search?

The change is a correction to the promise, not a reason to delete accurate structured data. Google now frames visibility in AI Overviews and AI Mode around its existing Search ranking, quality, crawling, indexing, and snippet eligibility systems. Its generative AI guidance says a page must be indexed, eligible to appear with a snippet, and included in Search generative AI has in Search Console before it can be eligible for display.

That matters because Schema Markup is often sold as if it creates an AI citation pathway. Google documents structured data as explicit information that can help it understand a page and support rich-result eligibility. It does not document a Schema Markup type, property, validator, or implementation pattern that guarantees a citation in an AI answer.

The before-and-after is simple. Before this clarification, teams could treat markup as an AI-search tactic with a presumed direct payoff. After it, the defensible position is narrower: use structured data to describe the visible page accurately, maintain technical eligibility, and measure actual presence in answer engines rather than crediting markup for citations it did not demonstrably cause.

Klarivo's September 16, 2026 analysis makes the same distinction. It notes that the often-repeated rich-result engagement figures are not evidence of AI citations. A 25 percent click-through-rate comparison for Rotten Tomatoes pages and an 82 percent comparison for Nestlé rich-result pages describe classic Search outcomes, not inclusion in generative answers.

1. Before: Schema Markup was commonly framed as a direct way to get cited by AI systems.
2. After: Schema Markup is a technical description layer with documented Google Search uses, not a documented AI-citation factor.
3. What to do: keep accurate markup, remove deceptive or obsolete implementations, then measure Answer Presence and Citation Share across the engines that matter.

Who does this change affect most?

This change affects the people who own implementation tickets and the people who promise outcomes from them. B2B SaaS growth teams can spend weeks adding FAQPage, HowTo, review, product, and organization markup while leaving the underlying answer thin, stale, inaccessible, or unsupported. Local and multi-location businesses face the same risk when schema work substitutes for pages that actually answer service-and-location questions.

It also affects agencies and vendors that use AI-citation claims to justify a markup project. A statement such as adding FAQPage will get your brand cited by ChatGPT is not backed by published engine documentation. It turns a legitimate technical practice into an unsupported promise.

Editorial teams are affected too. A page can be beautifully marked up and still has nothing distinct to quote. Google's generative AI guidance emphasizes original, people-first content and warns against creating separate pages for every query variation mainly to manipulate rankings or generative responses. Markup cannot make commodity copy become source material.

Technical teams should not hear this as an argument for neglect. Google says structured data problems can lead to loss of rich-result eligibility, and it recommends the Rich Results Test and URL Inspection tool for many technical errors. The real discipline is to separate technical correctness from claims about visibility outcomes.

1. Growth leaders who need a credible AI-search measurement model.
2. Developers maintaining templates, structured-data generators, and CMS plugins.
3. Editorial teams publishing comparison, definition, and FAQ content.
4. Agencies that must distinguish a documented rich-result benefit from an unproven citation claim.

Which Schema Markup tactics should you avoid?

Avoid marking up text that visitors cannot see. Google's structured-data policies say markup must is the main visible content of the page. Adding hidden answers, invented specifications, unshown prices, or unsupported claims creates a compliance risk instead of a citation asset.

Avoid using irrelevant types because a competitor uses them. A comparison article is analysis, not a product listing. A blog post is not a SoftwareApplication. A company about page is not a review page. Select a type from what the page actually is, then ensure its properties describe that page truthfully.

Avoid self-serving review markup designed to manufacture stars. Google restricts review snippets for self-serving reviews of an Organization or LocalBusiness. A third-party badge embedded on a company site does not automatically make that site eligible for review presentation.

Avoid treating FAQPage as a rich-result growth tactic. Google restricted FAQ rich results in 2023 and Klarivo reports that the FAQ rich result stopped showing on May 7, 2026. FAQPage vocabulary may still help label a visible question-and-answer section, but the reason to use it is accurate description, not an expectation of FAQ expansion in Search or a citation promise.

Avoid continuing to build HowTo markup around a rich-result outcome. Google reduced HowTo visibility in 2023 and later removed its supporting documentation after the result stopped appearing. If an existing HowTo implementation accurately describes visible content, assess it as maintenance work. Do not make it the center of a new AI-search plan.

Avoid copying large schema blocks from generators without ownership. A template can leave old dates, placeholder authors, incorrect entities, conflicting canonical URLs, or price fields that no longer match the page. These are data-quality problems. More properties do not fix them.

1. Hidden or misleading markup.
2. Types selected for hoped-for visibility rather than page meaning.
3. Self-serving review markup.
4. FAQPage and HowTo projects sold as citation tactics.
5. Unowned generator output that is never audited.

What should teams keep and validate?

Keep markup that accurately states durable facts a machine would otherwise need to infer. Organization markup can identify a company and its official profiles. Article or BlogPosting markup can describe an editorial page, its author, publication date, modification date, and image. BreadcrumbList can clarify site hierarchy. Product or SoftwareApplication markup can be appropriate where the visible page genuinely supports those claims.

Use JSON-LD where it fits the implementation. Google recommends JSON-LD, while its general guidelines also support Microdata and RDFa. The important test is not the format alone. The markup must be accessible to Googlebot, match visible content, follow policy, and remain current when the page changes.

Validate in two layers. First, use Google's Rich Results Test and URL Inspection tool to catch technical and indexing issues. Second, manually compare the output against the rendered page. A validator can identify syntax and has eligibility, but it cannot fully decide whether a price, claim, rating, date, or entity description is truthful and representative.

Treat manual actions seriously. Google says a structured-data manual action removes rich-result eligibility without changing ordinary web ranking. That distinction is useful: a markup issue can hurt presentation without proving anything about AI citations. Fix the issue because accurate implementation is the standard, not because a repair guarantees citation growth.

1. Match every marked-up claim to visible page content.
2. Use the type that describes the page, not the has you hope to trigger.
3. Run Rich Results Test and URL Inspection after template changes.
4. Review dates, offers, authors, ratings, entities, and canonical URLs on rendered pages.
5. Track actual engine-level outcomes after changes.

How should you respond if your site has legacy markup?

Start with an inventory, not a rewrite. Export the schema types emitted by each template, map them to representative URLs, and identify which implementation owns each block. Separate accurate markup that needs routine validation from markup that is misleading, unsupported, duplicated, or attached to the wrong page type.

Then prioritize by risk. Remove deceptive or invisible markup first. Correct prices, dates, reviews, and entity facts next. Audit legacy FAQPage and HowTo deployments after that, especially where teams are still reporting rich-result or AI-citation expectations that no longer have a documented basis.

Do not use a schema cleanup as a substitute for answer quality. Google describes its generative has as drawing from its Search index through retrieval-augmented generation and query fan-out. That means a page still needs to be crawlable, indexed, eligible for a snippet, easy to understand, current, and distinctive enough to support the user question.

Finally, measure the result where the decision happens. Track whether your brand appears in relevant AI answers and whether it is cited, rather than declaring success because a validator passed. Omnicite calls the percentage of relevant answers in a category that cite a brand Citation Share. It is the outcome metric. Valid markup is supporting infrastructure.

1. Inventory schema by template and page type.
2. Remove misleading, hidden, duplicated, and unsupported claims.
3. Validate surviving markup with Google tools and a visible-content review.
4. Improve the answer, evidence, freshness, and crawlability of priority pages.
5. Measure Citation Share and Answer Presence after deployment.

What does a credible AI-search Schema Markup plan look like?

A credible plan treats Schema Markup as one part of technical clarity. It starts with a page that answers a real question directly, uses clear headings, exposes evidence, and is available to crawl and index. Then it adds structured data that faithfully describes the content. The order matters because no JSON-LD block can rescue a page that does not deserve to be surfaced.

The plan also distinguishes engine behavior. Google publishes structured-data documentation and rich-result testing tools. Other answer engines do not publish an equivalent schema contract that says a particular type earns citation treatment. Do not turn Google's documentation into a blanket claim about ChatGPT, Perplexity, Gemini, Copilot, or other systems.

The better operating model is Citation Engineering: build authoritative coverage, keep it fresh, make its facts clear, and track which answers cite it. Schema Markup has a role in clarity. It is not a shortcut around quality, coverage, or measurement.

That is the practical news reaction. Keep Schema Markup accurate. Stop selling it as an AI citation hack. Put the investment into pages that can survive scrutiny when an answer engine needs a source.

1. Use schema to describe facts, not to simulate authority.
2. Maintain Search eligibility and crawlability.
3. Publish distinct answers with evidence that readers can inspect.
4. Separate rich-result reporting from AI-citation reporting.
5. Use Citation Share to judge whether the work changes visibility.

Sources

Source: Google Search Central

Google's generative AI guidance says eligibility requires indexing, snippet eligibility, and inclusion in Search generative AI has in Search Console. Google Search Central, 2026-09-29

Source: Google Search Central

Google documents structured data as a way to provide explicit clues and support rich-result experiences, including a Nestlé case study reporting an 82 percent higher click-through rate for pages shown as rich results. Google Search Central, 2026-09-29

Source: Google Search Central

Google's structured-data policies require visible, relevant, accurate markup and state that structured-data manual actions remove rich-result eligibility. Google Search Central, 2026-09-29

Source: Klarivo

Klarivo distinguishes documented rich-result effects from unproven AI-citation claims and reports the 2026 FAQ rich-result retirement context. Klarivo, 2026-09-16

Source: Google Search Central

Google's documentation update log records changes to Search documentation, including 2026 structured-data updates and the AI optimization guide clarification. Google Search Central, 2026-09-15