The last couple of weeks have been loud. Google delivered something called the Open Up Understanding Style Then Google Developers introduced the Agentic Source Discovery (ARD) spec

Meanwhile, every search engine optimization LinkedIn feed is lit up with someone either declaring markdown the future of the internet or describing why you ought to neglect all of it.

The fact, as per usual, rests someplace more intriguing than either camp.

The web is creating an identical machine-readable facilities (MCP/WebMCP, OKF, ARD, LLMs.txt …) and Search engine optimizations who recognize what each layer in fact does, rather than treating all of it as “AI SEO” or a silver bullet, will make much better decisions concerning where to spend their time.

First: The Layer Cake

There go to the very least 6 distinct things being reviewed under the umbrella of “making your site AI-ready.” They rest at different layers and serve various objectives:

• Crawlable HTML Pages: Still the structure. Nothing has actually transformed here. Whatever else remains on top.
• Schema.org/ Structured Data: Semantic tips baked into HTML that inform machines explicitly what a page is about. It is, basically, a vocabulary.
• LLMs.txt: Basically a navigating data. Its purpose is to basically inform an AI agent that’s already on your website which web pages matter. However as John Mueller places it on the Search Off the Record podcast :

“If somebody is currently on your website, possibly some type of computerized system is practical. Where if it goes, I intend to most likely to Martin’s Splitt and get a picture, after that the LLM system can most likely to your internet site and can browse, like, just how do you purchase a photograph? Maybe he has some standards for me as a representative for acquiring pictures. That type of makes sense.”

• MCP / WebMCP: Before ARD entered into play, we were presented with an additional solution for the challenge of interoperability. An MCP, in its simplest explanation, is a standard method for an AI to attach to your solutions to draw out expertise or do something about it. WebMCP, as the name itself suggests, gives internet sites a method to involve with representatives straight. WebMCP is for real-time web browser interactions on a webpage; MCP is for devices and services beyond the page.
• Open Expertise Layout (OKF): A package of markdown files with YAML frontmatter.
• Agentic Resource Exploration (ARD): A new open specification for exactly how agents locate and confirm tools, abilities, and other agents across the web. Below, the focus is not your web content; it’s your abilities.

For ecommerce, there’s one more layer worth naming independently– the product feed — quite potentially the future of retail exploration

Each layer does something different.

I could maintain contributing to this checklist; there’s a brand-new layer appearing every 5 minutes. I’m stopping right here. It’s ballooning.

What OKF In Fact Is (And Isn’t)

Google published the OKF spec quietly, bolted to a rebrand of Dataplex into Understanding Brochure.

The style itself is almost disarmingly basic: a directory of markdown files, each with a small YAML header declaring a type, title, description, resource, and some tags. The documents connect to each various other like any kind of markdown paper would certainly. That’s it.

As Google’s own blog site places it, OKF is “simply markdown, just files, just YAML frontmatter.”

SEARCH ENGINE OPTIMIZATION Suganthan Mohanadasan has a clear breakdown of this He explains OKF as one flooring in a stack that now includes sitemap.xml (which Links exist), LLMs.txt (which pages you most desire read), and OKF (the collection itself). They pile rather than compete.

The complication sets in not when you take a look at what OKF is, but what it does and in which layer of the agentic and search mayhem it sits.

In my mind, OKF is not a retrieval system. It doesn’t replace crawling. And, personally, I do not see a future where AI systems no longer consume huge quantities of HTML or where search and dustcloth are not a multistep complicated pipeline that contains self-reported and “unbased” signals.

Any type of self-reported system can and will be gamed. So thinking you can simply slam a number of markdown files on your site and be THE recommended selection in retrieval and exploration is far-fetched.

OKF is a higher-signal source among many. It might minimize parsing cost and enhance signal top quality, yet it does not replace existing pipes.

It’s additionally worth being honest below: OKF was constructed for information groups, not marketing sites.

It showed up as a way to share interior knowledge, i.e., table schemas, runbooks, statistics definitions, in between AI agents inside organizations. Aiming it at a public website to me seems a bit like we are yet once again repurposing.

Francois Vanderseypen makes one of the most specific factor regarding what OKF actually is and isn’t : a routed chart of markdown data is an internet of papers, not an expertise chart (at the very least not in its purest feeling). A real KG has explicit, queryable, keyed in relationships. OKF leaves what a web link implies completely as much as the producer, and an LLM still has to presume the semantics every single time it reviews it.

For me, this indicate the crux of just how I recognize the web and what we do as SEOs. OKF does not change the pile. It adds another input right into it.

It’s not a shortcut. There are no shortcuts.

The Schema.org Parallel, And Why It Matters

Among the patterns to comprehend below is the one Schema.org currently experienced.

Structured data complied with a foreseeable arc:

Adoption– ranking increase– prevalent use (and gaming)– platform knowing– lowered dependence as a ranking signal.

FAQ schema had a minute in SERPs, after that Google stopped the frequently asked question rich result The systems learn from the signals, fold the lessons right into the formula, and the specific markup comes to be less needed.

OKF and LLMs.txt might adhere to the exact same path. They’re most useful early, as clear signals in a globe where AI systems are still learning to parse the internet.

Over time, if the formats work, the systems find out. Explicit markup becomes redundant or remains a verification layer. For instance, in ecommerce, particularly, schema and feed alignment has ended up being more and more essential. An additional notch in the call for co-ownership of the item feed in between SEO and paid groups!

There’s additionally a subtler factor worth making below regarding the partnership in between schema.org and exploration. Jarno van Driel’s deep dive on item versions in Internet search engine Journal shows this well: For years, Google Look and Google Merchant Center had clashing organized data requirements, requiring publishers to replicate markup. Schema.org advances to close gaps, but it’s slow, it’s complicated, and execution is still often a mess.

Structured information has actually never ever been a plug-and-play ranking bar. OKF will not be either.

Should You Convert Your Site To Markdown?

It’s a huge fat no from me. That doesn’t indicate I won’t check it and use thoroughly!

And John Mueller stated it on the Search Off the Record podcast :

“When it pertains to things like an internet search engine or most likely likewise in generic LLM system, having an internet site that utilizes regular HTML for the pages is critical. Due to the fact that an internet search engine or spider can simply go to that web page. It can identify every one of the other links that are within the website.”

The structural information in HTML– nav web links, footers, header power structures, interior web links– is just how crawlers comprehend your website’s shape. Markdown data strip all of that out. You would certainly be breaking exploration in order to partially improve maker readability of specific pages.

Lately, on LinkedIn, I also saw a piece of research showing how “Your navigation may be consuming your LLM (it’s ChatGPT Deep Research study actually) reading budget.” Interesting findings, however please do not remove your navigation to “save some symbols”!

Jono Alderson makes this factor remarkably : “A page is not simply a container for words. It’s a content artifact.” Power structure, emphasis, positioning, what precedes, what projects, what’s tucked in an explanation … these aren’t rather decorations for humans. “They are signals concerning definition.”

“When you squash a page into markdown, you don’t just get rid of mess. You get rid of judgment, and you get rid of context.” And the minute you release a machine-only depiction, you’ve developed a 2nd prospect variation of reality.

The dull fix still works: Semantic HTML, clear structure, sensible hierarchy, content that exists when the web page tons.

John Mueller covers the markdown dispute extensively in the podcast: The parallel versions problem, the dynamic making lessons we already learned by hand, and why preserving a shadow version of your site for AI increases your maintenance burden and develops a debugging problem no one will inform you around.

The one exemption Mueller takes is designer documentation:

“If you have something like developer paperwork, where, again, if the agent or the LLM system already finds out about your site and the customer claims, just how do I usethis API? After that if you provide the LLM system a Markdown data, it’s a lot much easier for it to understand.”

Now, I can definitely see a straightforward usage instance there.

What ARD Is Actually Doing

The Agentic Resource Exploration spec , announced by Google on June 17, 2026, is a different beast completely. It got here only a number of days behind OKF, not a coincidence, and is already making huge waves.

The trouble ARD solves is a control one. Right now, an agent has to be wired to each tool, MCP server, or API it uses before it can do anything with it.

That works when you’re attaching a handful of known solutions. It quits scaling the moment the number of readily available capabilities expands beyond what any kind of group can pre-configure by hand.

ARD relocates that exploration out of setup and right into runtime. The representative finds what it needs when it requires it, rather than only knowing what it was told about in advance.

It’s improved 2 primitives:

• Directories: An ai-catalog. json file held on your domain, defining your available capabilities (MCP servers, A 2 A representatives, OpenAPI tools). Ownership of the domain serves as the cryptographic foundation for identity and trust fund.
• Pc registries: Search engines for the agentic web. They creep magazines, index them, and return matching capacities with the metadata needed to validate the author prior to connecting.

If OKF has to do with product packaging knowledge for consumption, ARD has to do with advertising capabilities for link.

These are parallel efforts at various layers of the arising agentic pile. Both delivered within inches of each other and currently embraced with the rate of light by some very big gamers in the game, i.e., Embracing Face and their Discover Tool.

It’s potentially a much more practical wager than the official logic layer that came before it and never got to internet scale. Time will inform.

A Void Worth Seeing

Within days of both specifications shipping, a factor opened friend problems on the ARD and OKF repos pointing out something fundamental was missing: There’s no agreed media type for an OKF bundle , so a directory can list one yet can not really recognize it as OKF without sniffing the contents.

In the meantime, publishers are currently promoting bundles in manufacturing using their very own interim types, which, as the concern itself keeps in mind, will not agree with each various other.

Presumably, this appears like a tiny ask, just an ask for a shared label.

After a little a dive into this specific rabbit hole, it ends up that’s rather regular practice. Waiting for complete arrangement prior to anybody ships anything is specifically how a specification passes away in committee, and shipping quick and patching as actual adoption surfaces is an olden strategy.

Application/json itself had not been formally registered till 2006 , about 5 years after JSON was already in vast, casual usage. Nobody stressed over that, because the price of the label being unsettled was low: A parser could deny something or drop back ungracefully.

But OKF is different, due to the fact that what occurs after the bring is various. The artefact behind the tag is a package a self-governing agent is indicated to consume, validate, and possibly act upon, inside a discovery system developed especially for agent-to-agent and agent-to-tool link. Obtain the kind incorrect right here, or leave an agent to infer it, and the threat isn’t a parse mistake; it’s a system acting on something it shouldn’t have relied on, without any one examining the result initially.

I question the risk associated with settling this later instead of faster in this instance. I presume it depends on how rapid it obtains dealt with relative to just how fast fostering runs in advance of it.

What This Means If You’re A SEO

A couple of truthful verdicts and my existing reasoning:

For a lot of advertising and marketing and material sites, very little has changed. HTML, well-structured for human beings, is still the appropriate structure. A contact-us type and a clean website design will certainly serve you better than any OKF package ever before will. Exploration still depends on links, authority, individual signals … and indexing.

LLMs.txt is a signpost, not a search engine optimization tool. It serves for aiding a representative browse within your site once it’s currently there. It most likely does not make a huge difference in how representatives find you in the first place. And, possibly never will.

MCP/WebMCP. Neither is urgent for most advertising websites today, however if you’re developing anything with programmatic user interfaces or ecommerce streams you want representatives to browse, this is the instructions the framework is heading.

OKF makes a lot of sense if you’re remaining on organized interior understanding , i.e., documentation, API references, item specs … and you want to make it easier for representatives to eat. The free OKF generator Suganthan built will produce a bundle and provide you a chart sight of your inner web link framework as a side advantage. The structural audit alone seems worth it. Yet I will be running it on my web site initially, not on my customer’s internet site.

ARD is worth seeing if you’re constructing services with programmatic interfaces. If you have tools, representatives, or APIs you desire discoverable by other representatives, ARD is the emerging standard for how that gets done. Just recognize the identification layer below it, what a representative is in fact looking at when it finds your magazine access, is still being worked out in actual time, so I ‘d treat this as facilities to view very closely rather than develop important courses on right now.

The schema adoption cycle could duplicate. These layouts are most important now, as early signals. Implement them if you can do it inexpensively. Don’t build your method around them holding worth permanently and don’t bank on them as a silver bullet.

Ultimately, recognize the shiny points — if your company has larger fish to fry, i.e., a horrible site, a brand no one knows or takes care of, an audience you don’t understand … after that manage this initial before you get captured up in any one of these brand-new glossy points.

The Hidden Shift

What every one of this indicate is a web that’s genuinely expanding a second layer or a third head, one written for makers together with the one composed for browsers and humans.

Sitemap.xml informed spiders which URLs existed. Robots.txt informed them where not to go. LLMs.txt, OKF, and ARD are comparable framework for agentic systems: navigating tips, material packaging, and ability discovery.

None of it is obligatory today. None of it changes strong HTML, authoritative web content, sensible framework, or the important things that really rests beneath all of it: a brand worth searching for.

But the SEOs who comprehend what each layer in fact does, instead of treating it as a solitary undifferentiated “AI SEO” category, will certainly make much better bets on where to invest their time.

My money is on the second layer, an identical facilities written for equipments, not a replacement wherefore currently exists.

The third head circumstance, where agentic systems fully deviate from the human internet, would call for a various collection of bets than any of us are presently making.

Huge many thanks to Jarno van Driel, Jono Alderson, Chris Eco-friendly, Suganthan Mohanadasan, Kristine Schachinger, Gianluca Fiorelli, Victor Frying Pan, Renee Bigelow (and anyone else I’ve missed out on) for some dazzling conversations on this topic over the last few weeks.

Much more Resources:

Featured Image: Collagery/Shutterstock