Is AI Visibility Tracking Worth It? What Three New Traffic Studies Say for SaaS

Three new studies say AI assistants send very little traffic compared with Google. One found ChatGPT sessions equal to 0.62% of Google organic volume across 100 sites. Two others show AI visibility numbers are noisy. For most B2B SaaS teams, AI visibility is worth tracking carefully, but not worth a large budget shift yet.

On October 22, 2026, Search Engine Roundtable summarized three studies under the headline that AI visibility may not be worth investing in. This article walks through what each study measured, what the two measurement studies say about tracking tools, and how a SaaS marketing team can decide how much to spend. It is written for marketers and founders who run their own SEO. Cekpr covers how to get found by search and by AI, and that means being honest about what the numbers can and cannot show.

What Do the Three New Studies Say About AI Traffic?

All three studies point the same way: AI assistants send a small share of traffic and conversions. They use different data, so their numbers are not directly comparable. The table sets them side by side, based on the Search Engine Roundtable summary.

The pattern is clear, but the caveats matter. The studies cover limited samples, and one does not state its dataset size. Read them as strong hints, not final proof.

Does AI Traffic Convert Better Than Google Traffic?

In one study it did, and it still did not matter much. Marie Haynes reported that ChatGPT visitors converted at 13.8% against 3.02% for Google organic, more than 4.5 times the rate. Her analysis, "GEO is a distraction," covered more than 145 million sessions.

The catch is volume. Because ChatGPT sent so few visitors, Google still drove 96.5% of conversions across the channels studied, and 27 times more than all other large language models combined. A high rate on a tiny base is still a tiny result.

The SEO Works benchmark found a different pattern for conversion. It reported that AI referral traffic converts at 0.67 times the rate of paid search and 2.48 times the rate of direct traffic. For every 100 conversions from organic search, AI referrals delivered 3.2. The details are in the SEO Works AI referral traffic benchmark.

Two studies, two conversion stories. That disagreement is a useful reminder. Conversion rates depend on the business, the offer and how each analytics setup attributes traffic.

Is AI Traffic Hiding in Your Direct Traffic?

Kevin Indig tested that idea and could not find it. In his Growth Memo analysis of AI and direct traffic, he wrote that he could not yet detect hidden AI traffic without self-reported attribution. Direct sessions with identifiable AI traces made up about 0.003% of total direct traffic on average.

He also found that a lot of direct growth was bot traffic. One site recorded 409,000 direct sessions, of which only 198 were engaged. If you see a jump in direct visits, check engagement before you credit AI or celebrate.

The study is careful about limits. Across four brands, he ran 632 correlation tests between AI visibility and later direct or branded search traffic. Some strong correlations were likely chance, and mismatched timelines often matched the real data. One company, Ramp, estimated that 30% to 40% of its direct traffic came from answer engines, based on self-reported attribution. So the effect can exist, but it is hard to isolate.

Why Are AI Visibility Rankings So Unstable?

AI answers change every time you ask. Rand Fishkin of SparkToro and Patrick O'Donnell of Gumshoe.ai had about 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI features, for 2,961 combined runs. They found less than a 1-in-100 chance that ChatGPT or Google's AI returns the same brand list in any two of 100 runs.

Their report, AI recommendations are highly inconsistent, says ranking position is not a reliable metric. In one example, a brand appeared in 69 of 71 ChatGPT answers, a 97% visibility rate, but was the top mention only 25 times. The study was not peer-reviewed.

A separate preprint reached a similar view. Ron Sielinski, whose company sells AI visibility software, wrote that a ranking can be trusted only when the order stops changing as answers accumulate and the gap between sites exceeds the margin of error. Search Engine Journal covered it in AI Visibility Rankings Aren't Stable.

In that research, meeting the rule took between 33 and 94 answers with citations across 30 tests. Three tests never got there, even after 125 questions. For top-10 sites, the typical margin of error was about five positions. If a tool shows you a rank from one run, you are mostly looking at noise.

What Should a SaaS Team Measure Instead?

Measure visibility rate across many runs, referral sessions that convert and what customers tell you. Visibility rate is how often your brand appears across a large set of prompts. SparkToro suggests running prompts at least 60 to 100 times and averaging the results, rather than trusting one answer.

A practical scorecard has three parts, matching what Indig recommends:

The second item is the cheapest and most overlooked. Add a "How did you hear about us?" field to your demo form and trial signup, and include AI tools as options. Self-reported answers reveal influence that analytics cannot see.

For the technical side, our guides on how to measure AI search visibility in Search Console and how to read Bing's AI performance report show what first-party data is available today. Both use data from the search engines themselves, which beats a third-party score.

Should B2B SaaS Teams Stop Working on AI Visibility?

No. The studies argue against overspending, not against the work. Content that is clear, sourced and useful helps both search and AI answers, and little of it is wasted if AI traffic stays small. The mistake is to cut a proven channel to fund one that is still unproven.

B2B software is also not the same as the average site in these samples. Buyers research with long, specific questions, and an AI answer may shape a shortlist without a click. That influence is hard to see in analytics, which is the reason to ask customers directly.

Our look at where AI answers get their sources for B2B software found that only a small share points to your own site. That suggests earning mentions on other sites matters. It is a different job from chasing a tool's score.

How Much Should You Spend on AI Visibility Tracking?

Spend little until a tool shows its method. SparkToro advises requiring AI tracking vendors to provide transparent, publicly reviewable statistics before you buy. A tool that reports one-run rankings, or hides how many times it asked each question, does not meet that bar.

A simple budget rule works for most small teams. Put the bulk of your effort into search content and conversion work that you can measure. Set aside a small, fixed share for AI visibility checks, such as a monthly review of a set of prompts. Increase it only if self-reported attribution or revenue shows a real effect.

Check also that any AI-recommended brand details are accurate. SparkToro noted that some recommendations pointed to inactive or defunct brands. If an answer gets your own product wrong, fix your documentation and pricing pages first.

One more habit helps. Write down your assumptions before you read any report, such as "AI will send 5% of our trials by next year." Then compare the data with that guess each quarter. This keeps a team honest, because it shows whether you are reacting to evidence or to headlines. It also gives your leadership a clear reason to raise or lower the budget.

What Is the Bottom Line for 2026?

The bottom line is balance. AI assistants are real and growing, but today they send a small share of visits. Google still drives most conversions for the sites studied. Treat AI visibility as a signal to watch, not a number to chase.

At Cekpr, we would put it this way: measure what you can verify, ask your customers what you cannot, and avoid any score that changes every time you refresh it. The research is early, the samples are small and the platforms keep changing, so revisit the question each quarter.