Guides

We opened the primary study behind 20 AI search and GEO statistics. See which hold up, which are misquoted, and which have no source.

Guide 16 · Agents WebMCP on WordPress: a working example, and what Lighthouse agentic browsing checks Mostly yes, if your forms have real labels, your buttons are real buttons and nothing jumps while the page loads: that is what Lighthouse can score today. AI agents are starting to use websites for people: they fill in…

Search Console now reports Google AI Overview and AI Mode impressions. It has no clicks, no queries and no way to split the two. Here is what it measures, where the clicks already sit, and what our own report showed.

Brand mentions correlate with AI visibility more than backlinks do, but Google says seeking inauthentic mentions is not as helpful as it looks. What our citation data shows about where engines actually read.

Google says there is no special way to write for its AI features. Our data shows why matching one phrasing fails: a one-character change to a question changes a fifth of the sources. What does get cited.

Google says there is no requirement to break content into small pieces for AI. In our 45-answer run, the most-cited vendor publishes long comparison articles. Why chunking a site can backfire, and how to structure pages instead.

Google’s official guide to AI Overviews and AI Mode lists five things you don’t need: llms.txt, schema, chunking, AI-style writing and bought mentions. Our data agrees on all five, and shows what the guide cannot tell you about the other engines.

Fixed prompt wording is the published best practice across this category — and an identical query string returned an identical source list from Perplexity, twice. Which metrics that degrades, which it does not, and the five questions to put to a vendor.

Path depth correlates with citation at r = +0.002 — indistinguishable from zero. Query strings are associated with 24% fewer citations, so the rule is a clean path rather than a shallow one. What that means for a WordPress site: almost nothing.

Comparisons are 2.2% of citations against 21.88% for listicles, so a page per competitor pairing aims at a fiftieth of the opportunity. But the few worth building do a specific job — and a plugin with five million installs shows exactly what that job is.

The one controlled test found no lift. Why the argument for schema sounds airtight and where it breaks, what schema is genuinely worth keeping for, and what would change the verdict.

Backlinks correlate at 0.22, unlinked brand mentions at 0.66, YouTube presence at 0.74. Why the mechanism is different from classic search, what ten measured plugin categories show about where answers actually come from, and how much of this to believe.

Generally not — they fetch your HTML and parse it. Googlebot renders JavaScript, which is why a site can rank in classic search and be invisible to the engines that answer questions. How to check in two minutes, and which WordPress setups are actually at risk.

One term comes from a KDD 2024 paper with a published benchmark; the other is an industry coinage. In practice they ask you to do the same things. What is measurably different is retrieval versus attribution, and one engine versus another.

Three different reasons, identical from the outside, opposite fixes. A fifteen-minute diagnostic built on 52 tracked plugins across ten categories and two engines — including one with five million installs that AI answers never name.

Ahrefs checked 137,000 sites with an llms.txt file: 97% had never been fetched by any crawler. What the file is meant to do, why no engine reads it, how to check your own logs, and what to fix instead.

llms.txt, schema markup, backlinks, shallow URLs, mass comparison pages, and the assumption that AI crawlers run JavaScript — what the studies actually show, and what to do instead.