Lost in Translation: What Happens When AI Can’t Read Your Website

Every day, AI engines like Google AI Overviews, ChatGPT, Claude, Perplexity, and Gemini synthesize answers, surface sources, and build shortlists buyers use before contacting a vendor. They read dozens of websites, decide which ones they understand and trust, and respond. For most B2B company websites, this process happens without a clear signal from the website. The AI infers meaning from raw HTML, guessing context, who wrote it, and whether it is credible.

Schema markup changes this. It's a layer of structured data that tells an AI system exactly what your content is, who you are, and why it should trust you, instead of leaving it to draw its own conclusions. Think of it as handing the AI a clear brief rather than asking it to read the room. Getting that interpretation right matters more today than it ever has.

Buyers stopped clicking and started asking

For years, the game was ranking in search results. Schema markup existed then too, but its primary job was cosmetic: earning rich results like star ratings, FAQ dropdowns, and recipe cards that made a listing stand out.

That game has not disappeared, but a new one runs alongside it. Buyers increasingly ask questions and receive synthesized answers rather than ten blue links. The AI engine reads dozens of sources, decides which it understands and trusts, and surfaces a response.

Being cited in that response is becoming as commercially meaningful as ranking on page one. Optimizing for this outcome—Generative Engine Optimization, or GEO—asks a different question than traditional SEO: instead of “can this page rank?”, it asks “can an AI system understand this page well enough to use it accurately?” Schema markup is one of the most direct levers to answer yes.

What schema actually does (and what it doesn’t)

Schema markup is structured data, usually in JSON-LD format, placed in a page’s code that explicitly labels what content is rather than leaving a machine to infer it from raw HTML. Instead of a crawler guessing whether a page is a company profile, product listing, or opinion piece, schema states it directly.

One thing worth being clear about: schema is necessary but not sufficient. It tells an AI system what something is. The content still has to earn the citation. A page with weak or inaccurate content remains unusable to an AI engine even with a perfect schema implementation. The interpreter can only work with what you give to translate.

The most valuable schema types are the unglamorous ones

Here, most companies get it backward. The schema types that have attracted the most attention produce visual features in search results, such as star ratings, FAQ dropdowns, and recipe cards. They’re easy to see and measure. They also tend to be deprecated, restricted, or abused into irrelevance. Google restricted FAQ-rich results for standard commercial sites back in August 2023, limiting them almost exclusively to authoritative government and health sites). Several other display-triggering schema types faced restrictions after widespread misuse.

For GEO, the value lies in schema types that produce no visual features. Organization, Person, and sameAs are the most important. They establish entity trust: telling an AI system not just what your page says, but who you are and whether you’re the same company mentioned in other credible places across the web.

Think about what an AI engine must do before deciding to cite you. It needs to resolve your identity. Is this the same organization from that industry report? Is this author the same person mentioned in that peer-reviewed paper? Without clear entity signals, the AI guesses. Organization and sameAs schema link your identity across Wikidata, Crunchbase, LinkedIn, and other authoritative sources, giving the system the disambiguation it needs to trust you enough to cite you.

What to prioritize right now

When implementing or auditing a schema, the practical priority order is:

Start here:

• Organization / Person / sameAs: Establishes who you are and links your identity across authoritative sources like Wikidata, Crunchbase, and LinkedIn. This foundation of entity trust is the schema type most directly correlated with AI citation.

Add next: 

• Article / BlogPosting with accurate datePublished and dateModified: Signals content freshness and authorship, both important to AI systems evaluating source credibility.

• Product / LocalBusiness: Here, relevant for transactional or location-specific queries.

Use with intention:

• FAQPage / HowTo: Keep these where the content genuinely is question-and-answer or instructional. Google confirmed it still uses FAQPage markup to understand pages even after deprecating the visual rich-result feature in May 2026. The signal remains. Stop adding it just to chase a badge that no longer shows up.

Give the interpreter something to work with

The room is still making decisions about your company. AI engines synthesize answers and surface sources for buyers who will never see a traditional search results page. Whether your brand appears in those answers accurately, partially, or not at all depends on whether the machines in the room can read you clearly.

Schema markup is how you hand the interpreter a clear brief rather than leaving them to guess from context. Because JSON-LD, the standard format, is what all major AI engines rely on, implementing it once serves both traditional search and every AI answer engine. The interpreter only needs to be briefed once.

Strong content without schema is a well-prepared guest who never got introduced. In a room that moves this fast, you cannot afford to rely on a good impression.

Strong content without schema is a well-prepared guest who never got introduced. In a room that moves this fast, you cannot afford to rely on a good impression. If you're not sure whether your website is speaking a language AI can understand, find out where your digital strategy stands.