How Jerry became the source PerplexityChatGPT trusts - growing AI traffic +60% and more than doubling conversions from AI in eleven weeks

With Sitefire’s agents, Jerry’s team ran a diagnosis across its site: which questions real buyers ask AI engines, which pages the engines actually read, and where competitors get cited instead. The output was a prioritized page-level brief for each of the pages Jerry went on to optimize.

The first improved pages went live on April 23 and 24, with a second set through June.

For each page, Jerry’s team used Sitefire to identify the structural elements that the most-cited content in its category shares - the elements answer engines consistently reward - and paired them with Jerry’s own brand positioning.

To measure the effect of the work itself, we compared the pages Jerry improved with Sitefire against the rest of jerry.ai over the same weeks. Both groups sit on the same site and face the same conditions - the only difference is the improvements. The chart runs February 9 to June 29; the dashed line marks April 23-24, when the first improved pages went live.

Each line shows weekly AI traffic relative to that group’s own level before March. Both groups dropped when GPT-5.4 launched on March 5, and both recovered - but not equally. The pages improved with Sitefire averaged 112% of their old level and ended above where they started, while the untouched pages averaged only 72%. The gray line is what doing nothing looked like: the median untouched page lost 38% of its AI traffic across the period, and the share of pages that grew was twice as high among the improved pages.

The pattern behind these numbers - where AI traffic actually lands - is worth understanding for any team planning content for AI search.

Jerry already had what most brands want: a deep, well-ranked set of car-insurance content pages, built over years for how Google reads the web. Then a new discovery layer arrived on top of search, with different selection criteria. Through the spring, AI traffic to long-standing content declined across the ecosystem, while a small set of pages captured a growing share of it.

Worse, the layer is nearly invisible. When people research insurance inside ChatGPT or Google’s AI Mode, the engine reads specific pages to form its answer, then usually shows the brand’s homepage as the clickable link. Using Sitefire, Jerry’s team spotted the pattern in their own data before most of the industry had a name for it.

The number above each bar is the share landing on the homepage and signup pages.

AI engines spread their reading across the whole site: the homepage stays between 2% and 4% of what they read. Human visits go the other way: from May, about two of every three AI-referred sessions land on the homepage or signup pages - pages the engines barely read.

In the answers themselves, Jerry’s homepage rarely appears without one of Jerry’s content pages cited alongside it. When both appear together, the homepage usually carries the link a reader can click.

One fitted line explains 85% of the week-to-week variation (R²): roughly every two to three answers that cite a Jerry content page pull the homepage into one more answer. With no content-page citations, homepage appearances drop to almost zero.

Using Sitefire, Jerry made its existing content pages the ones AI answers rely on. AI traffic grew 60%, visits from AI answers grew 37%, and monthly conversions more than doubled - all measured in Jerry’s own logs and analytics. The content pages earn the citation. The homepage collects the click.

If you publish content pages, this split is already happening in your data - your analytics just credits the wrong pages. Sitefire shows you which pages AI actually reads, and what each one is missing.

Jerry and Sitefire are now building the missing measurement to drive further growth. It connects the content pages cited alongside the homepage to the signups they produce, so the next round of improvements goes to the pages that grow the business.