By JaySeptember 2026
The Standards Aren't Written Yet
Our first post set a checklist: named bot rules, a sitemap, Content Signals, a Markdown twin for every page. Every post since has added to it. Somewhere in that process, people started calling this whole space "AIEO" or "GEO," like it already has the kind of settled technical spec that SEO eventually got. It doesn't. Some of it isn't even close.
Here's what's actually standardized right now, what's a draft still being argued over, and what's a content-strategy buzzword with no protocol behind it at all.
One real working group, still pre-RFC
The closest thing to an actual standard is the IETF's AI Preferences (aipref) working group: two draft specifications, one defining a shared vocabulary for expressing AI-related content preferences, the other defining how to attach that vocabulary to content via well-known URIs, robots.txt, or HTTP headers. That's the standards-track version of what Cloudflare's Content Signals already tries to do on its own.
Both drafts are still Internet-Drafts, not RFCs. As of this research, the vocabulary draft's latest revision was dated 19 August 2026, expiring 20 February 2027, and carried the standard disclaimer that its contents "do not reflect consensus of the working group either in whole or part." This is a live conversation, not a finished spec.
The header we already ship, doing nothing yet
Our own robots.txt already sends Cloudflare's Content Signals field, the same mechanism the IETF draft above is trying to standardize properly:
Content-Signal: search=yes, ai-input=yes, ai-train=yesRight now, that line does nothing. Google's John Mueller has stated directly that Content Signals has "no effects whatsoever for any crawler or LLM" and, in his words, mostly "adds bloat & future maintenance to your robots.txt file."
We shipped it anyway, and we'd do it again. It's five lines. It costs nothing to maintain. And if the IETF draft above lands anywhere close to Cloudflare's implementation, a decision Cloudflare's own engineers are involved in, we're already positioned for it rather than retrofitting it later. That's a bet on optionality, not on the field doing anything today.
llms.txt: real growth, no proof it matters
llms.txt adoption is genuinely accelerating: rankability.com's tracking puts it at 8.7% of the top 1,000 sites as of June 2026, up from 0.3% a year earlier, with technology companies leading at 36.4% and Cloudflare, Azure, GitHub, Fastly, WordPress, Adobe, and Samsung all publishing one.
Growing from 0.3% to 8.7% of the top 1,000 sites in a single year.
The format itself is genuinely simple, which is part of why adoption is cheap: a title, a one-line summary, then plain Markdown links grouped under a couple of headings. Here's the shape of one, if we published it today:
# addAI.dev
> AI consultancy for small businesses: one task automated, no dev team required.
## Docs
- [Blog](https://addai.dev/blog.html): case studies on AI agent visibility
- [Contact](https://addai.dev/contact.html): get in touch about a projectWhat that data doesn't show is a crawler actually reading the file and doing something different because of it. Independent testing concludes that despite the growing discussion, no AI crawler currently looks for it in practice, and Mueller's own comment above extends to it directly: Google doesn't use llms.txt or its llms-author.txt variant either. Publishing one is nearly free, so plenty of sites do it as a hedge. We're holding off, for the same reason we didn't add it to any of our other posts, until there's a real, tested claim behind it rather than a growth chart.
GEO isn't a protocol, it's a label
"Generative engine optimization" has no technical spec, no working group, no version number. It's a content-strategy label sitting on top of the actual technical layers above: write in a way that's easy to cite, use tables and clear structure, and yes, do the robots.txt and sitemap groundwork our first post already covers. Useful advice, but it isn't a protocol you implement. Treating it like one is how a business ends up chasing a checklist instead of a real mechanism.
The next concrete layer worth adding to this site isn't a new protocol either. It's structured data, schema.org markup describing who we are and what we do in a form a machine can lift directly, which is exactly what the next post in this series covers.
This is the same standard we've held every post on this site to: test the actual claim before building around it, not the marketing version of it. That's the part we bring to your site too.