We have just finished building DevelopedDomains.com, a reference site on domain strategy. Fifty-two pages, five browser-based calculators, Lighthouse 100 across Performance, Accessibility and SEO on every template.
The site itself is about domains. The research behind it produced three findings about AI search that apply to anyone doing SEO work in 2026, and all three cut against things that get repeated confidently at conferences.
1. Adding schema markup did not lift AI citations
Ahrefs tracked 1,885 pages that added JSON-LD structured data between August 2025 and March 2026, measured against 4,000 matched control pages that did not.
The effect on AI Mode and ChatGPT citations was statistically indistinguishable from noise. Google AI Overview citations showed a small but statistically significant decline of 4.6%.
It is worth being precise about what that does and does not say, because the overstatement is already circulating. The study measured pages adding schema, not pages starting from none. “Schema markup does not matter” is not the finding. “Adding it to pages that already rank is not the unlock” is.
We shipped full JSON-LD on every page of the site anyway — Article, FAQPage, ItemList, BreadcrumbList, Organization. It is correct, it costs almost nothing when a generator emits it, and it remains the right thing to do for traditional rich results. What we will not do is attach a traffic promise to it.
2. Google says it does not read llms.txt
This one is settled by primary source rather than by study. From Google Search Central’s guidance on optimising for generative AI features, verbatim:
“You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn’t use them.”
Note the scope. Google is describing its own Search, including AI Overviews and AI Mode. Other services may use such files, and shipping one is neither help nor harm. But if the reason on the roadmap is “so we show up in AI Overviews,” that reason is not supported.
We ship an EntityMap on the site for the same reason we ship schema: it is cheap, it is correct, and standards worth supporting have to start somewhere. It is described on the site as standards-ready hygiene, not as a traffic mechanism.
3. The AI citation surface is more open than the classic SERP
This is the finding that should change how you plan work, and it is the most encouraging of the three.
Only 38% of AI Overview citations now come from pages ranking in the top ten — down from 76% a year earlier. Roughly 31% come from pages ranking beyond position 100.
On content type, BuzzStream’s analysis of four million AI citations across 3,600 prompts, published in March 2026, found original editorial content accounting for 81% of news citations, while syndicated press releases accounted for 0.04% of the total dataset. That was a one-week collection window, which is worth stating alongside the number.
Zyppy’s meta-analysis of 54 experiments, patents and case studies scored URL accessibility as the single highest-ranked citation factor at 9.5 out of 10 — ahead of every content signal. Crawlable, fast, accessible pages, in other words. The unglamorous half of technical SEO.
Taken together: a well-built page on a smaller site has a materially better chance of being cited by an AI system than it has of ranking in a classic top ten. That is not a claim I would have made a year ago.
A note on exact match domains, since we had to settle it
The site includes a cluster on exact match domains, which meant resolving whether they still carry any benefit.
The 2012 EMD update is well documented — Matt Cutts announced on 28 September 2012 that it affected 0.6% of English-US queries to a noticeable degree. What matters more is Google’s current position. John Mueller, in December 2023: “A keyword domain name is not going to give you any recognizable SEO advantage on Google.” On keyword gTLDs in November 2025, he was blunter still, saying there is “absolutely no SEO advantage” from one.
The most useful data point we found was from local search. Whitespark’s 2026 local ranking factors survey of 47 experts ranked keywords in the domain name 71st for local pack influence. Keywords in the Google Business Profile business title ranked 3rd. The gap between those two numbers is the practical answer to the question.
Every source claiming EMDs still work that we examined was published by someone selling domains, expired domain tools, or SEO services. That is not disqualifying on its own, but it is worth noticing when the evidence runs one way and the enthusiasm runs the other.
What transfers
The build standards are the part most likely to be useful regardless of subject: static HTML with no CMS, self-hosted subset fonts, separate cacheable CSS and JS, WCAG AA verified rather than assumed, and a verification script that fails the build on broken links, thin pages, duplicate titles, heading-order skips, contrast failures and missing alt text.
Every figure published on the site is attributed and dated. Where the evidence does not exist, the page says so rather than hedging into a claim. That turned out to be the most useful editorial rule of the whole project, and the one that made the research worth doing.
DevelopedDomains.com