Ryan ShelleyCEO, SMA Marketing An SEO agency checked GA4's "Direct" bucket against raw AI agent logs on its client sites. Pages optimized for AI agents drew 16x to 35x more AI visits than the pages next to them.
At a glance
More AI-attributed visits on optimized pages than on non-optimized pages of the same site. Four sites, three client relationships.
Share of "impossible to type" Direct traffic that landed on pages AI agents were confirmed to have visited.
On one site, the optimized pages were 20.7% of tracked URLs and took 88.0% of AI-attributed traffic.
On one site, non-optimized pages converted at 2.19x the rate of optimized ones. The finding that cuts the other way, kept in on purpose.
Analysis by SMA Marketing, based on a trailing 90-day traffic window in 2026, using GA4 and Salespeak page-level AI visit logs.
Background
SMA Marketing is an SEO agency led by Ryan Shelley. Over the last two years Ryan changed what the agency sells.
"Over the last 24 months we've done very intentional dual optimization for AI and regular search. Our clients' conversion rates have gone through the roof. Even as traffic decreased, we've been driving better-targeted visitors. So we shifted our whole approach: not growing your traffic, but growing your pipeline through organic channels."
Ryan ShelleyCEO, SMA MarketingSMA runs Salespeak on part of its client portfolio. The sites in this analysis belong to three clients, among them a carbon fiber materials manufacturer and a financial compliance technology company. One of the three operates two branded sites.
The problem
An agency that sells pipeline has to show where the pipeline came from. For AI search, the standard tools can't.
"So much of today's traffic isn't coming from people anymore — it's bots, moving at machine speed — and GA4 can't tell the difference as accurately as it claims to."
Ryan ShelleyCEO, SMA MarketingThe question got specific when Ryan watched Google Analytics record a real-time AI Mode visit as "Direct." The visitor had come from an AI answer. The referrer was stripped on the way, and GA4 filed the session under the one label that means "we don't know."
Direct traffic was rising across SMA's client sites. If one AI visit could hide there, how many others were hiding with it?
The method: URL plausibility
Ryan's test starts from a plain observation. A person cannot type
/careers/careers-post/4913293101 from memory. A Direct session that lands on that
page followed a link, and the referrer was lost.
The method has three steps:
- Pull every Direct landing page from GA4.
- Classify each one. Plausible: a URL a person could type or bookmark. Implausible: a long slug or an ID nobody has memorized.
- Check the implausible pages against Salespeak's page-level AI visit log, which records each AI agent visit by URL.
"It gets us data we honestly just don't have at the agent level: how they're actually using the site."
Ryan ShelleyCEO, SMA MarketingThe method needs no trust in SMA or in Salespeak. Anyone with GA4 and an agent-level log can rerun it on their own site.
What SMA found
Most untypeable Direct traffic lines up with AI activity. On every site tested, between 81.2% and 100% of implausible Direct traffic landed on pages that AI agents were confirmed to have visited. SMA estimates that this puts a quarter to a half of all Direct traffic on those sites in the AI column. That second figure is Ryan's estimate, not a measurement.
Optimized pages get the visits. Pages that Salespeak had optimized for AI agents received 16.3x to 34.7x more AI-attributed visits than non-optimized pages on the same site. The gap showed up on all four sites.
"We saw anywhere from 16 to 35 times more AI visits when we used Salespeak on a specific page."
Ryan ShelleyCEO, SMA MarketingAgent attention concentrates. On one site, the optimized pages were 20.7% of tracked URLs and captured 88.0% of all AI-attributed traffic.
A note on scale: this is four sites across three client relationships, not a broad sample. What makes it worth reading is the consistency. The lowest multiplier SMA saw was 16.3x.
The same pattern in Salespeak's own data
After Ryan shared his analysis, Salespeak ran the same comparison across its full customer base: more than 10 million AI agent visits logged between August 2025 and September 2026. Across the 27 customer sites with optimization live, optimized pages got 26x more AI visits than non-optimized pages on the same domain. That sits inside SMA's 16x to 35x range.
Both analyses read the same kind of log, so these are not two independent datasets. What is independent is the analyst. SMA ran its numbers first, for its own clients, without being asked.
A gap between two groups of pages does not prove optimization caused it. Companies optimize their important pages first, and important pages draw more visits anyway. Salespeak tested that objection with a before-and-after comparison: 208 batches of pages on 22 sites, each measured 30 days before and 30 days after optimization, against untouched pages on the same site.
Growth over 30 days, 208 page batches on 22 sites. The difference is an 82% lift in the first month.
So the 16x to 35x gap is an observed difference, and part of it is caused by the optimization.
What optimization does not do
SMA's analysis includes a result that a vendor would normally leave out. On one client site, non-optimized pages converted their visitors at 2.19x the rate of optimized pages.
Ryan reads it as two separate levers. AI visibility decides whether an agent finds and reads the page. Conversion readiness decides what a human does after arriving. Optimizing for agents moves the first lever. It does not move the second, and on that one site the pages that agents favored were not the pages that converted best.
Salespeak's position is the same. Getting read by the agent and getting chosen by the buyer are different jobs, and a page that wins the first can still lose the second.
What's next
SMA now uses the optimized-page multiplier in its client reporting, and is extending Salespeak from a subset of its portfolio to more of its client base. The next question on SMA's list is conversion readiness: making sure the pages AI agents favor are also the pages built to convert the humans those agents send.
Run the check yourself
- Open GA4 and list your top 100 Direct landing pages.
- Flag every URL nobody would type or bookmark.
- Add up the sessions on the flagged pages. That is the size of your mislabeled bucket.
- Compare the flagged pages with an AI agent visit log. Salespeak's agent traffic analytics are free.
See which AI agents visit your site, or book a demo to see page optimization on your own traffic.
About SMA Marketing
SMA Marketing is an AI-first search marketing agency founded in 2009 and based in Central Florida, offering SEO, Generative Engine Optimization (GEO), PPC, and content marketing. Founder and CEO Ryan Shelley has applied AI and machine learning to marketing for more than 15 years. Learn more at smamarketing.com.
More customer stories on Success Stories.