Why Texas Insurance Agents Still Need Schema Markup, Even Though Google Says They Don't

Google said the quiet part out loud

Somewhere in the last few months, a version of this made it into a Facebook group for insurance agents, or a text from whoever built your website, or a headline you skimmed and half-remembered: "Google says you don't need schema markup for AI search." It's not wrong. Google's own documentation says it in plain English. What usually doesn't survive the retelling is the one word that matters most in that sentence: Google's.

Insurance is a market where a real share of shopping now starts with a question typed into an assistant instead of a search box, and Texas carries an unusually large piece of that shopping. The state ran up 4.17 million ACA marketplace plan selections during the 2026 open enrollment period, a 5 percent increase over 2025 and the largest number-of-enrollees swing of any state CMS tracked⁶. Every one of those people, plus everyone still shopping Medicare Advantage during this fall's enrollment window, is a candidate to type a question into ChatGPT, Perplexity, or Google's AI Mode before they ever call an agent.

So when an agency owner reads that Google doesn't require schema for its AI features and concludes the whole subject is settled, the math that follows is understandable and wrong at the same time. Understandable, because the sentence is a real, accurately reported statement from Google. Wrong, because it answers a narrower question than the one most agents think they're asking. The question an agent actually cares about is "will AI name my agency," and that question has at least three other engines in it besides Google.

What Google's 2026 guidance actually says

Start with the actual text, because it's worth reading precisely rather than as a paraphrase. Google's AI features documentation states: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add."¹ The same page adds that there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" beyond what classic Search already requires¹: a page has to be indexed and eligible to show a snippet, the same bar every organic result already clears.

On June 15, 2026, Google's Search Central changelog logged a second, related clarification, this one aimed specifically at llms.txt and similar files: a note confirming that Google Search doesn't use them at all, so maintaining one neither helps nor hurts a site's presence in Google Search or its generative features³. Between those two statements, Google has now directly addressed the two things most AEO advice tells agencies to build first: schema markup and llms.txt. Google's answer to both, for Google specifically, is that neither is required.

Read alongside Google's description of how AI Overviews and AI Mode actually pull an answer together, that statement stops sounding like an oversight and starts sounding like an engineering choice. Both features "may use a 'query fan-out' technique, issuing multiple related searches across subtopics and data sources, to develop a response,"¹ which is Google's own language for generating several related sub-questions behind the scenes and pulling supporting material for each one from whatever's already indexed. The mechanism runs on the index, not on a parallel structured-data layer built just for it.

Notice the last two rows. Google didn't say schema markup stopped mattering everywhere. Its general structured data guidelines, last updated July 10, 2026, still require that "structured data must be a true representation of the page content" and that a page missing required properties "is not eligible for rich results"². That's a live rule governing ordinary Google rich results, the review stars and expandable listings that still appear in classic search, sitting right alongside the statement that none of it is required for AI Overviews specifically. Both things are true. Agencies keep hearing only the second one.

Why Google gets to say that about its own answers

Here's the part worth sitting with for a second, because it explains why this isn't a contradiction on Google's part. Google operates a crawler and an index built over more than two decades, dense enough that its own systems can extract meaning, entity relationships, and factual claims directly from rendered HTML most of the time, without a separate machine-readable summary standing in for the page. Its AI features inherit that index rather than building a new one, which is exactly what the eligibility language says: indexed and snippet eligible, full stop, nothing extra¹. Google doesn't need a translation layer for a language it already reads fluently.

That's a genuinely different starting position from a third-party AI product. ChatGPT's browsing tool, Perplexity's search layer, and Claude's web access don't inherit twenty years of Google's crawl history. Each one runs its own retrieval against a page at request time, or against a narrower index it built itself, and a clean, complete piece of structured data removes ambiguity those systems would otherwise have to infer from prose. Google can afford to say "we don't need this." The other three systems your prospects are already using were never part of that sentence, because Google was only ever describing its own product.

What ChatGPT, Perplexity, and Claude still need

None of the three engines an agent's prospects are most likely to ask a question in are Google, and none of them operate under Google's specific "we already understand your page" claim. What that means in practice, on an insurance agency site specifically, comes down to a handful of concrete gaps a missing schema leaves open.

Without InsuranceAgency or LocalBusiness schema, nothing on the page formally states, in a format a machine can parse without guessing, what kind of business this is, what area it serves, or what its verified name, address, and phone number are. A retrieval system reading a paragraph of prose has to infer all of that from context. A retrieval system reading a JSON-LD block just reads it.

Without Service schema on each line of business, an assistant answering "does this agency write Medicare Advantage" or "does this agency handle final expense" has to parse marketing copy to guess at an answer instead of reading a declared list. Without Article schema carrying a named organization as author, a blog post reads to a machine as anonymous text with no stated source, which is precisely the kind of content GEO research finds AI systems are least likely to quote and attribute by name.

None of this is a claim that ChatGPT, Perplexity, or Claude have published a rule requiring schema the way Google publishes its documentation. They mostly haven't said so explicitly, in either direction, which is itself the point: in the absence of a stated rule, giving a smaller, less mature retrieval system a structured, unambiguous answer instead of prose it has to interpret is the safer bet, not the wasted one.

One question, three different engines, three different answers

Abstract talk about crawlers and indexes is easier to follow with an actual query run through it. Picture a prospect in Harris County typing something close to "which Medicare Advantage plans in Houston have no monthly premium" into three different tools on the same afternoon.

In Google's AI Mode, the query fan-out Google describes kicks in: the system generates related sub-questions behind the scenes, covering plan availability, premium structure, and star ratings, and pulls supporting passages from whatever's already indexed and snippet eligible for each piece¹. An agency page that ranks well and answers the question thoroughly in plain text has a real shot here, schema or no schema, because Google's own crawl already understands the page.

Hand that same question to Perplexity or to ChatGPT with browsing turned on, and the retrieval step looks different. Neither system carries Google's twenty-year crawl history behind it. Each pulls a smaller set of pages at or near request time and has to decide, quickly, what a given page is about and whether to trust it enough to quote. A page with Service schema explicitly declaring "Medicare Advantage" as a line of business, and an InsuranceAgency block declaring Harris County as part of the service area, hands that decision to the retrieval system in a form it doesn't have to infer. A page with neither leaves the system guessing from marketing copy, under more time pressure than Google's index ever puts it under.

The practical result: the same agency, with the same underlying expertise, can be a strong candidate in one engine's answer and functionally invisible in another's, not because the content changed, but because one engine can read an unmarked page as confidently as a marked one and the others can't. That gap is exactly what the schema stack in this guide closes, and it's why "Google doesn't need it" was never the same claim as "nobody needs it."

Google removed FAQ rich results. Drop FAQ schema too?

This is the specific change that's driving most of the current confusion, so it's worth walking through exactly what happened and when. On May 8, 2026, Google's Search Central changelog announced that the FAQ rich result feature "will no longer appear in Google Search starting May 7, 2026"³. The visible documentation for the feature was removed from Search Central a little over a month later, on June 15, 2026, with the changelog entry noting simply that the feature was "no longer shown in Google Search results"³. As of this writing, the expandable FAQ snippet that used to appear under a search result is gone, for every site, everywhere.

What often gets skipped is that this feature was already narrow before it disappeared. Since September 14, 2023, Google had restricted the FAQ rich result to "well-known, authoritative government and health websites" only³, which means a typical independent insurance agency site had already been ineligible for the visual snippet for nearly three years before the removal made it official for every remaining site too. If your agency's FAQ section never actually produced that expandable snippet in a Google result, and for almost every agency reading this it never did, then May 2026 changed nothing you were relying on.

Put plainly, FAQPage schema was doing two jobs the whole time. One job was competing for a rich snippet slot in Google's results, and that job is over, and honestly was mostly already over for agency sites specifically. The other job was declaring, in a format ChatGPT, Perplexity, Claude, and Google's own query fan-out can all parse without guessing, exactly which sentence answers exactly which question. That second job never depended on the visual snippet existing, and it's still worth doing on every page that carries real questions and real answers.

What it costs a Texas agent specifically

Zoom out from the schema mechanics for a second to the actual size of who's asking these questions in Texas this year, because the stakes are not abstract. Texas posted 4.17 million ACA marketplace plan selections during the 2026 open enrollment period, a 5 percent increase, roughly 206,000 more people than 2025, and the largest single-state swing by raw enrollee count that KFF's analysis of CMS enrollment data identified⁶. An earlier semifinal CMS snapshot, published January 13, 2026, put the state's total at 4,113,465 selections, itself already about 150,000 more than the prior year, before the final count came in higher still⁷.

Selecting a plan and keeping it active turned out to be two different things for a meaningful share of those Texans. By February 2026, effectuated enrollment, meaning coverage people had actually kept active, sat around 3.28 million, about 146,000 fewer than the same point in 2025, a 4 percent decline even as raw plan selections rose⁶. That works out to roughly 79 percent of people who selected a plan actually carrying active coverage a few months later, one of the largest gaps between selection and activation among the states KFF was able to compare⁶.

The gap between those two KFF-sourced figures, 4.17 million selections against roughly 3.28 million effectuated, works out to about 890,000 Texans who selected a plan and, for one reason or another, didn't keep it active by February, plus everyone in the state currently weighing a Medicare Advantage plan ahead of this fall's enrollment window. Both groups are, by definition, people already in the market, already comparing options, and increasingly likely to open that comparison with a question to an assistant rather than a phone call to an agency. Every one of those questions gets answered by whichever engine the person happens to be using, and whichever source that engine trusts enough to name. An agency invisible to three of the four major engines isn't losing a small edge on that population. It's opting out of being considered by most of it.

The schema stack that still earns its keep

None of this requires exotic markup nobody's heard of. It's a fixed, known list of schema.org types, each doing one specific job, none of them invented for this guide. Here's what an insurance agency site should be carrying, and what each type is actually declaring to whatever's reading it.

One rule cuts across every row in that table, and it's the same one Google's own general structured data guidelines state for classic Search: "don't mark up content that is not visible to readers of the page," and the markup "must be a true representation of the page content"². A FAQPage block listing questions that don't appear anywhere in the visible text, or a Review schema citing a rating count higher than what's actually shown on the page, isn't a shortcut. It's the exact kind of mismatch that gets a listing removed from rich results and gives any retrieval system, human or machine, a real reason not to trust the rest of the page either.

How to check what your own site is shipping

This doesn't require a paid tool, and it takes a few minutes per page. Run through this on your homepage, your top service pages, and one or two blog posts.

1. View the page source and search for it directly. Right-click the page, choose View Page Source, and search for application/ld+json. Every block between those script tags is a piece of schema currently live on that page. If the search comes back empty, that page is carrying none.
2. Run the URL through Google's Rich Results Test. At search.google.com/test/rich-results, paste in a page URL. It still validates the underlying JSON-LD and flags missing required properties, even for schema types, like FAQPage, that no longer produce a visual result in Google itself.
3. Read what each block actually says against what's on the page. Open the FAQPage block specifically and compare its questions to what's visibly written on the page. A mismatch here is the exact violation Google's structured data guidelines call out², and it's common on sites where a template shipped generic placeholder questions nobody swapped out.
4. Check whether your author is a real named person or organization, or just "admin." Look for an author property inside any Article or BlogPosting block. A generic system username there reads as anonymous to anything trying to judge whether your content has a real, trustworthy source behind it.
5. Confirm the schema survives your platform's own rendering. Some page builders and GoHighLevel funnel templates inject JSON-LD through a settings panel that only fires on certain page types, or that a later template swap silently strips. View source on the actual live URL, not a preview or staging link, since a builder can render differently between the two.
6. If you're on a platform where you can't edit schema directly, know that before you build anything else on top of it. Some funnel and page-builder platforms don't expose JSON-LD editing at all, which means the schema question isn't a to-do item, it's a platform limitation worth weighing against a build that does give you that control.

How we build this in from day one

This is the layer Digital Foundation ships automatically rather than leaving to a plugin or a template nobody's audited. Every site we build carries the full stack from the table above on the page types it actually applies to: InsuranceAgency and LocalBusiness on the homepage and location pages, Service on each service page, Article with an Organization author on every blog post, FAQPage on any page with real questions, Dataset on any page citing a government source, BreadcrumbList sitewide, and a real Person bio behind whoever's named as author or reviewer.

Because it's built at the architecture level instead of bolted on page by page, it doesn't drift the way a hand-maintained WordPress or page-builder site does, where a template update or a new plugin can silently overwrite a JSON-LD block or introduce the exact visible-text mismatch Google's own guidelines warn against². We verified every claim in this guide by re-checking the live Digital Foundation pricing page the same day it published, the same standard we hold every number in this article to.

It also means the llms.txt question and the schema question get handled as two separate, correctly scoped decisions instead of one collapsed one. llms.txt ships because ChatGPT, Perplexity, and other agent tools were the actual audience for it, not because Google reads it, and every build's llms.txt and sitemap update themselves automatically the same day a new page publishes. This pairs with what we cover in our guide to the separate technical gate that controls Google AI Overviews eligibility specifically, and with what changed in AI citation more broadly this year: this guide is the schema layer underneath both of those, scoped correctly to the engine that actually reads it.

What actually changes

Not a guaranteed spot in any specific AI answer. No one, including Google, can promise that, and this guide isn't going to pretend otherwise. What changes is more basic than a promise: your site stops being unreadable to three of the four systems your prospects are already asking, for the sole reason that one true, narrow sentence about the fourth system got generalized into advice it was never meant to be.

Google's statement about its own AI Overviews was accurate the day it was published and it's accurate today. It was also never a statement about ChatGPT, Perplexity, Claude, or the roughly 890,000 Texans who selected an ACA plan this year and didn't keep it active by February. Building the schema stack doesn't contradict what Google said. It just answers the three-quarters of the question Google's sentence was never trying to.

Questions Texas agents ask