We added Entity Maps, llms.txt, and OKF across 121 websites to find out if any of them lift AI traffic. The short version: none of them moved the needle. Schema still pulls more weight than all three combined. Here is what the data showed.
A new wave of machine-readable formats promises to make your site easier for AI to read. Entity Maps, llms.txt, and OKF all live in a second layer of the web, one written for machines instead of people. We ran a controlled test across 121 websites to see whether any of them change how often AI tools find and cite a page. They did not. The pages that grew were the ones with age, links, and authority, not the ones carrying the newest file format.
The 121 test sites were not all built the same way, which matters because it shows the result is not a WordPress-only quirk. WordPress ran the clear majority at 95 sites, but the study also covered 13 JavaScript-based sites, 7 custom content management systems, 3 sites with no CMS at all, 2 Django sites, and 1 we could not confirm. The new formats did nothing across every one of these platforms, so the flat result is about the formats themselves, not about the stack a site happens to run on.
What Are Entity Maps, llms.txt, and OKF?
Entity Maps, llms.txt, and OKF are three separate machine-readable formats, and each one does a different job. They are easy to confuse because they all use plain text and all claim to help AI. The differences matter, so here is each one on its own.
Entity Maps declare which entities a site owns and how they connect. Think of them as a “who is who” for your brand: the company, the people, the topics you want your name attached to, and the links between them. They are a community spec, not an official Google product.
llms.txt is a single plain text file placed at the root of your domain that points AI agents at the pages you most want them to read. It faces outward, toward visiting crawlers and agents. Google added a Markdown-style check for it inside its developer tools in May 2026, which we cover below.
OKF, the Open Knowledge Format, is a folder of Markdown files that Google Cloud published on 12 June 2026 as version 0.1. Each file describes one concept, and the files link to each other like a small map. OKF was built for the AI agents a company runs on its own internal data, not for public websites trying to rank.
Do Entity Maps, llms.txt, or OKF Improve AI Traffic?
None of these three formats improved AI traffic in our test. Across 121 websites, adding Entity Maps or llms.txt produced no measurable lift in traffic from AI sources. OKF showed a small improvement, but only on the handful of sites where it actually belongs. The honest takeaway is that right now, none of these new formats move the needle, not for traditional SEO, which they were never meant to touch, and not for AI either.
How Did the Flamine 121-Website Experiment Work?
The test ran on 121 websites split into two groups, established sites and clean new sites. Of those, 96 were established, between 3 and 20 years old, and all of them already pulled traffic from AI sources. We added OKF to 12 of them, since those were the only .org and .edu sites in the group and the only ones that should ever use it. The other 25 sites were brand-new, registered just for this test, with no history and no backlinks. Every site in the study had a complete Schema setup with no errors and no warnings.
The clean sites were set up to isolate cause and effect, so the order of changes was the real experiment. We launched 10 new sites and gave them two weeks with no markup at all, then added llms.txt, then added Entity Maps a week later. At the same time we ran another 10 new sites in the reverse order, Entity Maps first and llms.txt later. The last 5 new sites got nothing for the full 60-day window, as a true control. OKF sites ran for a shorter stretch, a little under 30 days.
The 121 test sites were not all built the same way, which matters because it shows the result is not a WordPress-only quirk. WordPress ran the clear majority at 95 sites, with 13 JavaScript-based sites, 7 custom content management systems, 3 sites with no CMS at all, 2 Django sites, and 1 we could not confirm.
The 25 new clean sites were the most controlled of the group: every one ran WordPress on the default Twenty Twenty-Five theme, each carrying 20 posts created with claude.ai.
We split them across 5 servers so hosting could not skew the result, using SiteGround (two separate accounts), DigitalOcean (one instance in the United States and one in Europe), and Vultr (Singapore), with no two in the same city. Each server held two sites that added llms.txt first, two that added Entity Maps first, and one with no markup, and the new formats still did nothing on any platform or in any location.
How Did We Measure AI Traffic Across Every Site?
We measured every site with one fine-tuned report built from Google Analytics 4 and Microsoft Clarity, and nothing else. The setup was identical across all 121 sites for two reasons. First, it kept third-party citation tools out of the picture, since most of those only guess at AI traffic instead of measuring it. Second, one shared report template meant every site was scored the same way, which made the sites simple to compare side by side. A controlled test is only as trustworthy as the consistency of its measurement, so we kept the measurement identical everywhere.
What Happened When We Added Entity Maps?
Entity Maps created problems in Google Search Console without lifting AI traffic. After we installed it, Search Console flagged every page on the site as duplicated and non-canonical, which is the kind of warning that makes you nervous even when nothing is broken. On the other side, it added a brand-new enhancement report called datasets. That is the full picture so far: one new report, one batch of canonical warnings, and zero measurable change in how often AI tools cited the pages.
Does llms.txt Do Anything in 2026?
llms.txt now shows up in PageSpeed Insights, but it still does nothing for AI traffic or rankings. In May 2026, Google added an Agentic Browsing category to Lighthouse and PageSpeed Insights, and one of its checks is whether you have a valid llms.txt at your domain root. So keeping the file is fine, and it ticks a box in that report. The catch is that Google Search itself ignores it. Google has stated plainly in its Search Central guidance that you do not need llms.txt or any special markup to appear in its generative AI search, and that the file neither helps nor harms your visibility. Our data matched that exactly: no lift.
Should You Add OKF to Your Website?
OKF belongs on .org and .edu sites at most, and it is being misused everywhere else. OKF was designed for the AI agents a company runs against its own internal data, like metric definitions and database notes, not as a public file you drop on a marketing site to win citations. We are seeing it bolted onto sites it was never meant for, which is the same hype cycle that follows every new Google acronym. In our test, OKF produced a marginal improvement, and only on the .org and .edu sites where the format has a real reason to exist. On a normal business site, it is a file you have to keep in sync for no clear payoff.
The only sites that showed visible growth were the ones with no markup at all. It is tempting to read that as proof that the formats hurt you, but the established sites that grew were also older, had backlinks, and had real authority, so growth was likely coming from those. The one clean comparison was the 5 brand-new control sites. They did not get much AI traffic, but they did slightly better than the matching new sites that carried the new formats.
Why Did the Sites With No Markup Grow the Most?
The no-markup sites that grew were also the oldest and most linked, so age and authority are the simplest explanation. Most of those sites had years of history and real backlinks behind them, and that kind of authority drives AI citations far more than any file format. The interesting exception was the new clean sites, where age and links were taken out of the equation. There, the control group with nothing added still edged out the sites carrying Entity Maps and llms.txt.
One possibility is that these new formats are feeding language models contradictory signals. If your Schema says one thing and a separate Markdown layer says something slightly different, an AI system has to pick, and confusion is not what you want when you are trying to get cited. We are still researching that, so treat it as a hypothesis and not a conclusion. What we can say is that across 121 sites, Schema carried more weight than any of the three new formats, which lines up with the way Google keeps pointing people back to structured data and solid fundamentals.
| Format | What it actually does | Effect we measured |
|---|---|---|
| Entity Maps | Declares which entities a site owns and how they connect. A community spec, not a Google product. | No lift, side effects |
| llms.txt | A plain text file at your domain root that points AI agents at your key pages. Now a check inside PageSpeed Insights. | No lift |
| OKF | A folder of Markdown concept files built for a company’s own internal AI agents, not for public ranking. | Marginal on .org and .edu |
| Schema | Structured data that labels what each page is about, in a vocabulary search engines and AI systems already use. | Strongest of the four |
Right now, none of these new formats move the needle, not for SEO, which they were never meant to touch, and not for AI either. Schema still does more than all three combined.
What Should You Actually Do Right Now?
Spend your effort on Schema, authority, and content structure, not on chasing the newest file format. The formats that promise AI visibility through a quick file upload are the ones that did nothing in our test, while the boring fundamentals kept doing the work. If you want a starting point, get your structured data complete and error free, build real links, and write pages that answer one question each. Our Answer Engine Optimization services and GEO services are built around exactly those fundamentals, and if you want the writing side, our guide on content that AI summaries pull from covers how to structure pages so AI systems can actually retrieve them.
Not in our test. Across 121 websites, Entity Maps produced no measurable lift in AI citations or traffic. They also came with side effects in Google Search Console, where every page was flagged as duplicated and non-canonical, while a new datasets enhancement report appeared. The format is a community spec, not an official Google product, so treat any AI visibility claim about it as unproven for now.
llms.txt now appears as a check inside the PageSpeed Insights Agentic Browsing category, added in May 2026, so a valid file ticks that box. It does nothing for Google Search visibility or rankings, because Google Search ignores the file. In our experiment it produced no AI traffic lift. Keeping it is harmless, but it is not a growth tactic.
Probably not, unless you run a .org or .edu site with a real reason to use it. OKF was built for the AI agents a company runs against its own internal data, not as a public ranking file. We are seeing it added to sites it was never designed for. In our test it gave a marginal improvement only on the .org and .edu sites where it actually fits.
Yes. Across our 121-site study, Schema carried more weight for AI visibility than any of the three new formats combined. Structured data uses a vocabulary that search engines and AI systems already understand, so it labels your pages without guesswork. If you only fix one technical thing, make your Schema complete and free of errors and warnings before touching any new Markdown format.
Possibly, but it is not proven. In the clean new sites, the control group with no markup slightly outperformed the sites carrying Entity Maps and llms.txt. One theory is that a separate Markdown layer can contradict your Schema and confuse language models, which have to pick one version. We are still researching this, so treat it as a hypothesis rather than a settled fact.