This question has a well-documented half and a genuinely undocumented half, and the honest answer keeps them clearly separate rather than blending certainty from one into the other.
The documented half: traditional rich result eligibility
Google’s structured data general guidelines explicitly require markup to accurately reflect the visible content of the page. When structured data and the actual page content disagree, a product marked up as in stock while the visible text says “sold out,” a review rating in schema that doesn’t match anything a user can actually find and verify on the page, Google’s documented policy position is that this is a guideline violation, and the typical, well-established consequence is that the rich result eligibility for that markup is lost. In more serious or deliberate cases (markup designed to mislead rather than an honest sync error), this can rise to a manual action under Google’s spam policies rather than simply a withheld rich result. This part of the answer is solidly grounded in Google’s own published structured data guidelines and is not a speculative area.
The undocumented half: how AI-generated systems handle this specific conflict
For AI Overviews and similar generative search features, Google has not published a specific, disclosed mechanism describing how markup-versus-prose contradictions are detected or resolved during answer synthesis. There’s no public statement saying “our AI Overview system specifically checks structured data against visible text and does X when they disagree.” This is a genuine gap in what’s documented, and it should be presented as a gap, not papered over with a confident-sounding but unverified explanation.
The reasonable inference, clearly labeled as inference rather than fact, is that a page carrying an internal markup-versus-content inconsistency likely reads as a lower-quality or lower-trust signal overall to whatever combination of systems the AI Overview draws from, consistent with how such inconsistency already damages trust and eligibility in the traditional, documented context. It would be a defensible extrapolation to expect that a page with this kind of internal contradiction is, at minimum, not a strong candidate for confident citation in a generated answer, but there’s no specific, disclosed resolution mechanism to point to beyond that general expectation.
Why the distinction between these two halves matters
Collapsing the well-documented traditional-search behavior and the undocumented AI-search extension into one confident-sounding paragraph is exactly the kind of fabrication risk this topic invites. A reader could easily walk away thinking Google has published specifics about how its generative systems handle this exact conflict, when in fact the AI-specific half of the answer is reasoned extrapolation from a documented policy that was written for, and demonstrably enforced in, the traditional rich-results context.
A concrete example of how this drift typically happens
The most common real-world path to this conflict isn’t deliberate manipulation, it’s operational drift between systems that update on different schedules. A retailer’s inventory management system updates stock status in real time, while the page’s structured data is generated from a separate content management process, a nightly batch job, a cached template, a manually maintained feed, that updates on a slower or less reliable cadence. A product goes out of stock at 2pm; the visible page text (pulled live from inventory) reflects that immediately, while the JSON-LD block (regenerated only once daily) still asserts “InStock” until the next batch run. For the several hours in between, the page carries exactly the kind of internal contradiction this whole question is about, not because anyone intended to mislead, but because two systems describing the same fact update on different schedules and nobody built a check to catch the gap between them.
Why this is a higher-priority fix than it might initially seem
Because the documented consequence (lost rich result eligibility, potential manual action for sustained or severe cases) applies regardless of whether the AI-search extension of this problem is ever resolved or disclosed, fixing markup-content synchronization has a fully justified business case even without any reference to generative search at all. Treating this as a “nice to have” cleanup item, deprioritized behind other technical work, undervalues a problem that already has a documented, verifiable cost in the traditional rich-results context. Sites experiencing frequent, high-value data changes (pricing, availability, ratings) are the highest-priority candidates for building an automated consistency check between the data feed and the rendered markup, since manual periodic audits are unlikely to catch the kind of intermittent, timing-driven drift described above before it’s already caused a documented eligibility problem.
Where the ambiguity is worth explicitly flagging to stakeholders
When reporting on this issue internally, particularly to stakeholders who may not distinguish between documented Google policy and reasonable inference about AI systems, be explicit about which half of the explanation you’re giving them. Saying “Google’s guidelines require markup to match content, and mismatches risk losing rich result eligibility” is a fully citable, documented claim you can stand behind without qualification. Saying “and this is why the AI Overview won’t cite us” states a specific causal mechanism as fact when it’s actually a reasonable but unconfirmed inference, and stakeholders who hear it without the hedge may walk away believing Google has confirmed something it hasn’t.
The risk profile differs sharply between honest sync-lag and deliberate manipulation
Not all markup-content conflicts carry the same risk, and treating an honest synchronization lag the same as deliberate manipulation both overstates the danger of the former and understates the danger of the latter. A sync-lag conflict, like the inventory-batch-job example above, is characteristically transient, affects a narrow, predictable category of frequently-changing fields (price, stock status, ratings pulled from a separate feed), and resolves itself once the systems re-sync, with no intent to mislead a user or Google’s evaluation systems. Google’s documented guidance and enforcement history treat this category as a quality and eligibility issue: the practical consequence is typically limited to losing rich result eligibility for the affected markup until it’s corrected, without escalating to a manual action, because the pattern is recognizably an operational gap rather than an attempt to game evaluation. Deliberate manipulation looks structurally different: markup asserting a rating, review count, or availability status that has no basis in anything on the page or verifiable elsewhere, maintained persistently rather than resolving on its own, and typically applied in a way clearly intended to make the page look more favorable than the visible content supports. This pattern is squarely what Google’s spam policies on structured data are written to address, and it’s the category where manual action risk becomes genuinely live, not merely theoretical. The practical implication for prioritization is that a business discovering a sync-lag issue should treat it as an urgent but routine technical fix, while a business discovering markup that looks more like the deliberate pattern should treat it as a policy-compliance issue requiring more than a technical patch, since the underlying cause and the appropriate response genuinely differ between the two, even though both technically produce the same surface-level symptom of markup disagreeing with visible content.
Practical implication
Treat markup-content consistency as a hard requirement regardless of which surface you’re optimizing for, since the traditional consequence (lost rich result eligibility, possible manual action for deliberate mismatches) is real, documented, and independently sufficient reason to fix any discrepancy you find. Audit your structured data against your actual visible page content directly, particularly for frequently-changing data like pricing, availability, and ratings, where drift between a data feed and the rendered page is a common, avoidable failure mode. Don’t build a content or schema strategy around a specific assumed AI-conflict-resolution behavior, since no such specific behavior has actually been disclosed, fix the underlying inconsistency because it’s already a documented problem in the context Google has actually explained, and treat any AI-specific benefit from doing so as a reasonable but unconfirmed bonus.
