Module experimental 4 min
The Standards Race
The contest to standardize machine-content delivery got real entrants in 2025–26: llms.txt from the grassroots, OKF from Google Cloud, NLWeb from Microsoft.
- The contenders: llms.txt, OKF, NLWeb
- Attribution debates
- Reading adoption signals
Why it mattersWhoever shapes the standard shapes the next decade of GEO.
Definition & Foundation
Whoever shapes the standard shapes the decade
For two years "standards for AI" meant one grassroots proposal (llms.txt) that no engine ever confirmed reading. 2025–26 changed the shape of the race: Google Cloud published OKF (the Open Knowledge Format) as an open, agent-readable knowledge spec; Microsoft's NLWeb matured into deployable endpoints with CDN-managed hosting; and the agentic-commerce protocols (UCP, ACP, AP2) proved platforms will standardize the action layer even while the content layer stays contested. The contest to standardize how machines read and act on your site finally has real entrants with real backing.
For a practitioner the trap is chasing every announcement. The discipline is to watch adoption signals, not launch headlines: a CMS plugin that emits a format by default, CDN and bot-management support, and (the real prize) an engine confirming it reads any of them at retrieval time. Until an engine says "we read this," a standard is a proposal, not a ranking factor. This node is a watch-and-position topic, so it carries no impact score: your job is to keep the underlying structured data clean (which every contender builds on) and to know which way the race is trending before you commit.
The Three Live Contenders
Grassroots, Google, and Microsoft
llms.txt
A plain-text file curating your key content for LLMs. Widely adopted, easy to publish, but unconfirmed by engines, the low-cost hedge whose payoff depends entirely on future adoption. Deep dive: llms.txt.
OKF
The Open Knowledge Format, a vendor-backed, agent-readable knowledge specification. Carries the weight of a major cloud behind it, which is exactly the kind of backing llms.txt lacked. Deep dive: OKF.
NLWeb
Reframes the contest from "publish a file" to "expose a queryable endpoint" agents can call. Its MCP-server property makes it the most action-oriented contender. Deep dive: NLWeb & MCP.
The Contenders, in Order
How the standards race actually developed
llms.txt is proposed
A grassroots convention (a plain-text file listing your most important content for LLMs) gains traction among practitioners. It's widely adopted by publishers, but no major engine confirms reading it. See llms.txt.
Microsoft ships NLWeb
NLWeb turns a site into a conversational /ask + /mcp endpoint built from existing schema and feeds, moving the contest from "a file to read" toward "an endpoint to query." See NLWeb & MCP.
The action layer standardizes first
ACP (OpenAI + Stripe), then Google's AP2 and UCP, standardize agent commerce. Platforms prove they'll agree on how agents act even while how they read content stays open, a telling asymmetry. See Agentic Commerce.
Google Cloud publishes OKF
The Open Knowledge Format arrives as a vendor-backed, agent-readable knowledge spec, the first standard with a major cloud behind it, landing alongside llms.txt and NLWeb. See OKF.
Will an engine confirm reading one?
The race isn't won by announcements but by retrieval-time adoption. The milestone to watch for is any engine publicly confirming it reads a format when it builds an answer, the moment a proposal becomes a ranking input.
Putting It to Work
Position without betting the farm
You don't pick a winner in an unsettled race; you keep the shared foundation clean, take the cheap hedges, and watch the signals that separate a real standard from a proposal.
The standards-watch playbook
Keep the shared foundation clean
Every contender builds on accurate structured data and feeds. Invest there first; it's the one thing that pays off no matter which standard wins, and it improves your citations today regardless of the race's outcome.
Take the cheap hedges
Publishing an llms.txt file costs little; standing up an NLWeb endpoint is realistic if your schema is clean. Take the low-cost positions, but size the effort to the uncertainty; don't rebuild your site around an unconfirmed spec.
Watch adoption signals, not announcements
Track the real tells: CMS plugins emitting a format by default, CDN and bot-management support, and any engine confirming retrieval-time reading. A launch post is noise; default emission and confirmed reading are signal.
Assign a standards watcher
One person tracks the race and reports material moves to the team, the same way you'd watch any emerging surface. Centralized watching beats everyone forwarding hype threads.
Commit when a standard crosses the line
The trigger to invest properly is an engine confirming it reads a format at retrieval time, or your platform emitting it by default. Until then, hold your position: clean data, cheap hedges, and attention.
Update Cadence & Dependencies
Keeping it alive
Works together with
Go deeper
Always current
These links resolve live: what you get is generated or filtered the moment you click.
Fresh from the field feed refreshed July 8, 2026
- Jul 7 AI Search: Is Your Content Strategy Accidentally Recommending Your Competitors? Search Engine Journal
- Jul 6 AI Content Didn't Stop Working, Your Metrics Did via @sejournal, @MattGSouthern Search Engine Journal
- Jul 6 Self-Promotional Content Works - Until It Backfires (AI SEO Experiment) Ahrefs Blog
- Jun 25 Keeping Data-Driven Content Fresh Was a Monthly Slog. So We Taught an Agent to Do It. Ahrefs Blog