The SEO industry, has a pattern: new tactic -> sickening, massive adoption. Everyone jumps on new methods that are working, or at least what’s said to be working by the general SEO community like moth to a flame.
With SEOs being the early owners of AI Search, similar bandwagon patterns have started to show up. Schema has been the biggest hot button of these. Some SEOs call it the foundation of how AI understands your brand. Others say LLMs strip it out and ignore it. And files like llms.txt and markdown versions of pages have only added to the noise.
So at BrightonSEO in San Diego, we asked four experts the same question: how effective are schema, llms.txt, markdown files and other machine-readable formats for AI search, really?
We got four different answers, ranging from "zero impact" to "this is your knowledge graph AI uses to understand your brand." The disagreement is actually surprisingly useful when we break it down. Each expert is looking at a different part of the problem, and together they point to a clear strategy, rather than the confusion it may seem on the surface.
The four positions at a glance
| Expert | Position | In one line |
| Thomas Peham, CEO & Co-Founder, OtterlyAI | Doesn't work for AI citations | In OtterlyAI's tests, schema had zero impact on citations in ChatGPT and Gemini. |
| Chris Long, Co-Founder, Nectiv | Not game-breaking, but do it | Schema's value for LLMs is unproven, but it's cheap, and early adopters win if agents adopt llms.txt. |
| Chima Mmeje, Senior Content Marketing Manager, Moz | Schema yes, llms.txt no | Schema is important; llms.txt won't make the difference people claim. |
| Martha van Berkel, CEO & Co-Founder, Schema App | Essential, if you rethink it | Schema isn't a rich-result checklist. It's the knowledge graph AI uses to understand your brand. |
Thomas Peham: "It doesn't work right now for search"
Thomas Peham, CEO of AI visibility platform OtterlyAI, gave the most stark answer (watch Thomas's clip). He separates Google from AI chatbots:
"Schema markup's super important for Google, right? … Not so much on ChatGPT."
OtterlyAI tested it. In Thomas's words, schema markup had "zero, literally zero impact" on citation visibility in ChatGPT, Gemini and other AI search engines in their experiments.
Thomas also predicts that within a year, AI Mode will be the default interface on Google.com and SEO and GEO will merge. If that's right, schema's Google value and its AI value may stop being separate questions.
Chris Long: "Not game-breaking" but worth doing
Chris Long, co-founder of Nectiv, lands in the middle (watch Chris's clip):
"I don't think any of them are game-breaking, honestly."
On schema, he notes the research is split. Some studies show it helps and some don't. His lean is that it probably doesn't help much with LLMs, because they tend to strip markup out when they read a page.
But his advice is still to implement it with how much of a low lift it is. Schema, llms.txt and markdown files "take no time to implement," and schema is good technical practice anyway. If AI agents ever adopt llms.txt, early adopters will have one less thing to build. In other words, it's low cost with real upside.
Chima Mmeje: Schema yes, llms.txt no
Chima Mmeje, senior content marketing manager at Moz, splits the question in two (watch Chima's clip). She cited a talk from Moz's event in New York, where a speaker from Meta, quoting someone at Google, said Google isn't using llms.txt when it evaluates website content.
Her verdict:
"Schema is important, don't get me wrong, but llms.txt is not."
She isn't against llms.txt. "Do it, cool," she says. Just don't expect it to deliver the difference people have been promising. There's a lot more you can do to optimize content for AI search that has nothing to do with an llms.txt file.
This matches what our team found in Why llms.txt isn't a magic bullet for AI (but you should probably have one anyway).
Martha van Berkel: Schema is your knowledge graph
Martha van Berkel, CEO of Schema App, does schema markup "all day every day," so her answer isn't a surprise. Her reasoning is the most useful part (watch Martha's clip).
She argues most SEOs still think of schema as a checklist item to win rich results. That misses the point. Schema is how you are
"connecting the dots across every single thing on your website."
Call it entities, business concepts, a semantic data layer or a context layer: it's the data that helps AI make the right inferences about your brand. Schema App works with more than 100 enterprises, including Wells Fargo, and Martha says they're seeing results in both visibility and conversion rates.
Her challenge to SEOs is to "think like a data architect." Adding markup to a page is easy. Building a real ontology or enterprise knowledge graph is harder and takes more time and effort, but it's where the value is. She points to research from data.world finding LLMs are far more accurate when they work from a knowledge graph. Schema App's free "Schema Markup 101" training is a good place to start getting familiarized with how a knowledge graph might work best for your site!
She also looks further ahead, predicting AI agents will start acting on sites, booking appointments and filling out forms, and agents will need data they can trust (watch).
We went deeper on this in Alyssa Felix's Entity mapping and knowledge graphs: the context layer we all need, which features Martha's thinking.
What we recommend
The four answers disagree less than they seem to. Thomas and Chris are measuring whether schema directly raises citations in chatbots today. Martha is describing how structured data helps AI understand who you are. Both can be true. Here's how we advise clients at 97th Floor:
- Implement schema, but don't expect a lift in individual LLM citations. It's low effort, it still matters for Google, and Google's AI features are becoming the default. Just don't expect a ChatGPT boom from markup alone.
- Build for entities, not rich results. Use schema to connect your organization, products, people and locations: sameAs, parent and sub-organizations, breadcrumbs. Consistent, connected data is what keeps AI from confusing your brand with someone else's.
- Add llms.txt if it's easy, then move on. It's another low-cost readiness step, especially for sites with docs or APIs, but it's not a priority for most brands.
- Treat markup as infrastructure. As our team put it in 97F Reacts: what markup, markdown and schema are really doing for AI search, clean HTML and structured data give content meaning for machines. Whether markdown beats schema is still an open question, so test it, and never hide content from users that you show to bots.
- Spend most of your effort where citations come from. Machine-readable files help AI read your site. Getting cited still depends on content worth citing, a site structure AI can follow, and off-site mentions in the sources LLMs trust. Those are the three pillars of our GEO/AEO approach.
Add the schema, then invest in what actually gets you cited
Ask four AI search experts whether schema matters and you'll get four answers. That isn't confusion, though. It shows that schema does a different job than many SEOs expect. It probably won't push you into a ChatGPT answer on its own. It will help machines understand your brand clearly and consistently, and that matters more as AI starts acting on websites, not just reading them.
So implement schema, skip the silver-bullet thinking, and put your real energy into the content, structure and off-site signals AI actually cites.
Want more takeaways from the conference? Read Mike Witham's 45/40/15: What BrightonSEO San Diego taught me about where SEO is headed.
Not sure how AI sees your brand? Let's talk.
