Every "AI traffic converts 4x better" chart you have seen was made by someone selling AI search services. This one is too. The difference is that we are showing the setup, the bots, and the number we could not measure.
There is a claim going around that visitors arriving from ChatGPT, Claude and Perplexity are worth several times a Google visitor. We would like it to be true, since we sell the work that earns those visitors. So we pulled our own GA4, separated the AI channel properly (the method is in the GA4 tracking guide published earlier in this series), subtracted the bot traffic that was polluting the sitewide numbers, and wrote down what we found and what we could not find. The engagement half of the claim holds on our site. The lead half is unmeasured, here and in most of the charts.
Evolve Media Agency sells AI search visibility services; every number below is about our own site, and a better-looking result helps our business. All figures are from GA4 for evolveamz.com, pulled September 15, 2026: a twelve-month window (September 15, 2025 to September 14, 2026) and a last-30-day window (August 16 to September 14, 2026). Derived figures (engaged-session counts, ratios, the bot-adjusted estimate) are computed from those and labeled. Lead tracking was misconfigured for the whole period, so no lead-conversion figure appears anywhere in this post.
The Question, And What "Convert" Means Here
"Convert" on an ecommerce store means an order. On an agency site it means a booked call or a form. We cannot report the second for this period (section 10), so this post uses the two measures GA4 reports reliably for every channel and that precede any conversion: engagement rate (the share of sessions that lasted over ten seconds, had a key event, or viewed two or more pages) and average engagement time per session.
Engagement rate (GA4). The percentage of sessions that were "engaged": lasted longer than ten seconds, triggered a key event, or included at least two page views. Average engagement time per session is the time the page was in the foreground, averaged across sessions. Neither is a conversion, but both are what a conversion is made of, and both are comparable across channels without any event setup.
So the question this post can answer is: do AI assistant visitors engage more than other visitors, and by how much. The question it cannot answer yet is whether that engagement turns into leads at a different rate. The distinction matters because the second question is the one every vendor chart claims to answer, usually without showing the events behind it.
The Data And Its Limits
Three limits to hold in mind. The AI channel is small, so its percentages move with a handful of sessions. The sitewide figures are polluted by bots, which makes AI look better than it should against "sitewide" and is why section 4 compares against organic search instead. And this is one agency site with a content cluster built specifically to be cited by assistants, which is not a random sample of the web.
Twelve Months: AI vs Sitewide
| 12 Months (Sep 15, 2025 - Sep 14, 2026) | Sessions | Engagement Rate | Avg Engagement Time | Engaged Sessions (Derived) |
|---|---|---|---|---|
| AI Assistant channel (GA4 label) | 409 | 36.7% | 32s | ~150 |
| All AI referrals (source-level, incl. misfiled) | ~670 | n/a (mixed labels) | n/a | n/a |
| Sitewide (bots included) | 32,686 | 24.5% | 8s | ~8,000 |
On the labeled channel, AI visitors engaged at 1.5 times the sitewide rate and stayed four times as long. Two caveats already: the sitewide rate is depressed by bots (section 5), and 409 is the sessions GA4 correctly labeled, with about 260 more AI sessions filed elsewhere because the assistant's referrer did not match GA4's channel rules. The engagement figures for those 260 are mixed into Referral and (not set) and are not separable in the channel report, which is why the GA4 tracking guide builds a source-based channel group; the 30-day figures in the next section were pulled after that fix.
Last 30 Days: AI vs Organic vs Sitewide
| Last 30 Days (Aug 16 - Sep 14, 2026) | Sessions | Engagement Rate | Avg Engagement Time | Engaged Sessions (Derived) |
|---|---|---|---|---|
| AI Assistant | 126 | 47.6% | 48s | ~60 |
| Organic Search | 383 | 43.6% | 48s | ~167 |
| Sitewide (bots included) | 7,144 | 20.6% | 7s | ~1,470 |
This is the comparison that matters, because organic search is human traffic arriving with intent, which is the fair benchmark. AI visitors engaged at 47.6% against organic's 43.6%, a difference of four points, or about 1.09 times. Average engagement time was identical at 48 seconds. Against the bot-polluted sitewide figure, AI looks 2.3 times better; against the honest benchmark, it is a little better and within the noise of a 126-session sample.
That is the finding, stated plainly: on our site, AI assistant visitors behave like good organic search visitors. Slightly more engaged, equally long, and a small fraction of the volume: 126 sessions against 383 organic and 7,144 total, or under 2% of sessions. Anyone reporting AI traffic as several times more engaged than search is either comparing against a bot-inflated baseline, as our twelve-month table would if read carelessly, or has a different site.
Subtracting The Bots
About half of the site's users in the last 30 days were from Singapore, at around one second of engagement: datacenter traffic hitting pages and leaving. They are in the sitewide denominator, which is why the sitewide engagement rate is 20.6% and 7 seconds. A rough correction shows what the sitewide human number probably looks like.
The reason to show this: the bot inflation is not unique to us. Any site behind Cloudflare with a public robots.txt that welcomes crawlers gets this traffic, and any AI-versus-sitewide engagement comparison that does not subtract it is overstating the AI advantage by a factor of about two. The crawler log guide earlier in this series covers finding the bots; the fix on the analytics side is a filter, and it is in section 12.
By Assistant: ChatGPT, Claude, Gemini, Perplexity
| Source (12 Months, Source-Level Count) | Sessions (Approx.) | Share Of AI Total | Note |
|---|---|---|---|
| ChatGPT | ~316 | ~47% | Largest by far; the most consistent referrer month to month |
| Claude | ~173 | ~26% | Disproportionate to its market share; the site's Claude-specific guides are the likely reason |
| Gemini | ~103 | ~15% | Growing; Google's AI surfaces also send traffic labeled as organic, which is not counted here |
| Perplexity | ~51 | ~8% | Small but present every month |
| Doubao, Yuanbao and others | remainder to ~670 | ~4% | Chinese assistants; a reminder that the channel is wider than the four names |
Per-source engagement rates are not reported here because the per-source samples (51 Perplexity sessions in a year) are too small to compare honestly; the differences would be noise dressed as insight. What the source split does show is that ChatGPT is about half of AI referral volume, and that Claude's share is far above its share of assistant usage, which we attribute to having written specifically about Claude for ecommerce operators. Assistants cite content about themselves. The twelve-month citations guide covers how that cluster was built.
Where AI Visitors Land
AI visitors and Google visitors enter through different doors, and the difference explains some of the engagement.
- The top AI entry page is the local business AI search guide (22 AI-assistant entries in the period, the most of any page), which is also a top-five Google entry page. It is a long how-to that answers a question completely; assistants cite it, and the visitor arrives to read the rest.
- The AI crawler list is the number two Google entry page (69 entries in three months) and a steady AI entry, for the same reason: a reference someone came to use.
- Service pages get AI entries too: /pricing/ receives both Google and AI entries at over a minute of engagement. An assistant asked "what does an Amazon agency cost" and sent someone to the page with the numbers.
- The homepage is not a major AI entry. Assistants link to the page that answered the question, and the question is rarely "who is Evolve Media."
The pattern: AI visitors land on the deepest, most specific page for their question, having already read a summary of it, and then either leave (the summary was enough) or read for a minute (it was not). That is a different visit from a Google click on a title tag, and it is the mechanism behind the engagement numbers. Section 9 takes it further.
The Trend: The Only Channel Growing
The growth is the reason the channel is worth measuring carefully now rather than later. At under 2% of sessions it does not move the business. At the current rate, and with the assistants adding shopping and agent features monthly, it is the channel most likely to matter in a year, and the site that has clean measurement in place when that happens is the one that can act on it. The AI search reporting dashboard guide is the Looker Studio build we use to watch it.
One seasonal caution on the trend. August into September is a period when B2B traffic to this site normally rises as brands plan for the fourth quarter, so some of the 15% is calendar rather than channel. The comparison that removes that is AI sessions against organic sessions over the same two periods: AI rose while organic did not, which is the reason we read the direction as real. Three more months of the same shape would make it a trend rather than a good month, and that is the standard the dashboard is set up to check.
Why AI Visitors Engage More
Four mechanisms, from the data and from reading the sessions.
- Pre-qualification. The assistant already gave the visitor a summary and the visitor clicked through anyway. That is a stronger intent signal than a search result title; the ones who would have bounced on the summary never arrived.
- Deep entry. They land on the specific page for their specific question, past the navigation and the hero. There is nothing to bounce from except the answer itself.
- Selection bias in what gets cited. Assistants cite complete, specific, well-structured pages. Those pages engage everyone better. Some of the AI advantage belongs to the pages rather than the visitors, and the organic comparison in section 4 partly controls for it (the same pages rank on Google too).
- Fewer accidental visits. No ads, no rich-result thumbnails, no misclicks. An AI referral is a deliberate act at the end of a conversation.
The honest reading of the four is that AI visitors are engaged because assistants send fewer, better-matched visitors to better pages. That is good for the visitor and good for the site. It is not evidence that AI traffic is magic, and it means the way to earn more of it is to write more pages that answer one question completely, which is the same advice as always. The AI search visibility guide is that advice in full.
What We Cannot Claim, And Why
Our key events (call bookings, form submits) were legacy and misfiring for the entire period. We have no reliable lead count by channel, so we have no lead conversion rate by channel. We are rebuilding the events now; the first honest comparison is a quarter away.
Without lead data there is no value per visit. With lead data, a 126-session month would still be too small to price a visit. Anyone quoting a multiple should show the events and the sample.
Fifty-one Perplexity sessions in a year is not a sample. The source table shows volume only.
One B2B agency site with a content cluster built to be cited. An ecommerce store selling a $30 product will see different assistants, different pages and, once orders are tracked, a real conversion rate. Measure your own.
We are publishing the engagement result because it is real and useful and we can show the work. We are declining the conversion claim because we cannot, and because the industry's habit of making it without showing the events is why brands are being sold AI search work on numbers nobody can check.
What This Changes About What We Build
- Fix the events before anything else. The lead question is the one that matters and it is unanswerable until the key events fire correctly. That is the first project, and the reason this post has a hole in it.
- Filter the bots in GA4. A sitewide baseline that is half datacenter traffic makes every channel comparison wrong. Section 12 has the filter.
- Keep writing the deep page. The top AI entry pages are complete answers to specific questions. The strategy that earns AI visitors is the same one that earns engaged Google visitors, and the numbers say both are the same kind of visitor.
- Put the next step on the pages assistants cite. An AI visitor on the local guide reads for a minute and leaves. The page's job now is to offer the audit, the call, or the next guide, because the visitor has already been pre-qualified.
- Watch the trend rather than the share. Under 2% of sessions is not a business. Fifteen percent a month is a direction. Measure monthly and act when the base is large enough to price.
The Ecom Profit Box
Our library of ecommerce growth guides, including the AI search measurement and content frameworks behind this post.
Get It FreeGet Your AI Channel Measured Honestly
Bring your GA4. We will separate the AI referrals, filter the bots, check whether your conversion events actually fire, and tell you what your data can and cannot say.
Book A CallMeasuring It On Your Own Site
The counterweight, since we sell measurement: none of this needs us. It is a channel group, a segment and a DebugView session, and the tracking tools roundup covers the paid tools that add citation monitoring on top. What we would ask of anyone selling you AI search work, including us, is the same thing this post tried to do: show the events, show the bots, name the sample, and say what the data cannot support.
On engagement, AI visitors look like good organic visitors, a few points better, on a small base that is growing. On leads, we do not know yet. Anyone who says they do should show you the events.
What To Remember
- On evolveamz.com in the last 30 days, AI assistant visitors engaged at 47.6% against organic search's 43.6%, both at 48 seconds; AI visitors behave like good organic visitors, slightly better, on a 126-session base.
- The sitewide baseline (20.6%, 7 seconds) is polluted by bot traffic that made up about half of users; a bot-adjusted human sitewide rate is roughly 41%, and any AI-versus-sitewide multiple that ignores bots overstates the AI advantage by about two times.
- Over twelve months the labeled AI channel had 409 sessions, and about 670 once misfiled AI referrals are counted: ChatGPT roughly half, Claude a quarter, then Gemini and Perplexity.
- The AI channel grew 15% month over month while every other sizable channel was flat or down, from under 2% of sessions.
- AI visitors land on deep, specific pages that answer one question completely, having already read a summary; that pre-qualification is the mechanism behind the engagement.
- No lead-conversion claim is made, because key events were misconfigured for the period; that is the first thing being fixed, and it is the number most vendor charts assert without showing the events.
- Measure your own: source-based AI channel group, bot filter, organic as the benchmark, verified events, and no rate comparisons under about 100 sessions.
Where This Came From
- Google Analytics 4 property for evolveamz.com, pulled September 15, 2026: Traffic acquisition and Engagement reports for the twelve-month window (September 15, 2025 to September 14, 2026) and the last-30-day window (August 16 to September 14, 2026), plus the prior 30 days for the trend; source-level Explore for the AI referral count and landing pages; country-level Explore for the bot identification. First-party.
- GA4's AI Assistant default channel definition (introduced May 2026) and the source-based custom channel group described in this site's GA4 AI tracking guide, for the two AI counts (409 labeled; about 670 by source).
- Inference note: engaged-session counts, ratios, the bot-adjusted sitewide estimate and the share figures are derived from the reported numbers and are labeled as such; the per-assistant split is approximate; the mechanisms in section 9 are our interpretation. No lead or conversion figure is reported because key events were misconfigured for the period.