ChatGPT sent theoryroad.com 77 sessions in the last 90 days, and 4 of them filled out our contact form. Google organic sent 208 sessions and 5 leads. That is a 5.2% lead rate from ChatGPT against 2.4% from Google. The sample is tiny, the window is one quarter, and the site is our own, so treat this as a field note rather than a study. But the pattern shows up again in the landing pages, and it has already changed what we build.
This is our agency site, not client data. Every number below comes from Google Analytics 4, Google Search Console and Ahrefs Brand Radar for theoryroad.com, July 18 to October 16, 2026. We are publishing it because almost nobody shares real AI search numbers with conversions attached, and because the fix for both channels turns out to be the same work. If you run a site and want to know whether AI search is worth your time yet, this is the most honest answer we can give you.
The numbers, channel by channel.
GA4 recorded 214 Organic Search sessions, 78 AI Assistant sessions, 29 Referral, 7 Email and 2 Paid Search sessions over the 90 days. Direct showed 563 and we threw it out, for reasons covered in the next section. Of the 78 AI Assistant sessions, 77 were ChatGPT: 73 tagged chatgpt.com, 3 tagged chatgpt.com/organic and 1 tagged chatgpt.com/referral. Gemini was a trickle, 4 sessions from gemini.google.com plus 3 more under a different Gemini tag. Perplexity, Copilot and Claude recorded zero referrals in the period.
Leads are our generate_lead event, which fires when a prospect submits the contact form. Google organic produced 5 leads from 208 google/organic sessions, a 2.4% rate. ChatGPT produced 4 leads from 77 sessions, 5.2%. Paid search produced 1 lead from 2 sessions, which is a rounding error, not a result. Direct produced 2, and Meta and the site's own URL-box flow produced 1 each.
| Channel | Sessions | Form starts | Leads | Lead rate |
|---|---|---|---|---|
| ChatGPT (AI Assistant) | 77 | 5 | 4 | 5.2% |
| Google organic | 208 | 8 | 5 | 2.4% |
| Paid search | 2 | n/a | 1 | too small to rate |
| Direct (bot inflated, excluded) | 563 | 11 | 2 | 0.4%, not meaningful |
| Referral | 29 | n/a | 0 | 0% |
| 7 | n/a | 0 | 0% |
Form starts tell the same story one step earlier. 5 of 78 AI Assistant visitors moved from step one to step two of the form, 6.4%. 8 of 208 Google organic visitors did, 3.8%. Direct logged 11 starts, which looks impressive until you remember most of that channel is not human. If you only track completions today, add a form start event this week. It lets you see intent before it converts, and it gives small channels enough signal to read.
Why Direct is out of the analysis.
Direct was our biggest channel at 563 sessions and our least useful. When we opened it up, a large share came from headless browsers and datacenter IP ranges in Singapore and India with zero engagement: no scroll, no second page, no measurable time on site. That is scrapers, uptime monitors and crawler farms, not prospects typing our URL into a browser.
GA4 does not filter this for you. A bot that executes JavaScript shows up as a session, and because it carries no referrer, it lands in Direct. Leaving 563 inflated sessions in a channel comparison would make every human channel look better than it really is, so Direct sits in the table for completeness and stays out of every rate we quote.
If your own Direct number looks suspiciously large, open the Tech and Geo reports in GA4. Zero engagement time plus a datacenter country you do not sell into is the tell. Build a segment that excludes that traffic before you compare channels, or every conclusion you draw about conversion rate will be wrong by the size of the bot share.
Where ChatGPT visitors land.
Only 15 of the 77 ChatGPT sessions landed on the homepage. The rest went deep. About 20 landed on our who we serve pages, the /for/ business-type pages, spread across 13 different pages: supplement brands 3, art galleries 2, commercial cleaning 2, dental practices 2, gyms 2, product brands 2, and single visits to franchise, home and kitchen, logistics, pet products, print shops and youth sports.
Another 4 landed on /about, 4 on our US market entry page, and a few on service pages: fractional CMO 2, AI consulting 1, paid media 1. Platform pages took 1 visit each (GA4 and Tag Manager, GoHighLevel, Google Shopping and Performance Max, Looker Studio dashboards). Around 12 long-form articles took 1 to 3 visits apiece, led by a guide on how to get junk removal leads with 3, then single visits to pieces like an interior design marketing plan and a conversion tracking audit checklist.
Read that list again. Someone asked ChatGPT something like who does marketing for supplement brands, or how do I get more junk removal leads, and ChatGPT pointed them at the one page on our site that answers it. They did not search our name. They did not browse from the homepage. They arrived on the page that matched the question they had already asked. That is why the lead rate is higher. The question was a buying question before they ever reached us.
The action here is simple. Open your AI Assistant landing pages in GA4 and look at the distribution. If assistants only ever send people to your homepage, write pages they can send people to instead. Our who we serve section is one page per business type for exactly this reason, and it is where a quarter of our ChatGPT traffic arrived.
Where Google visitors land.
Google organic behaved like Google organic. Only 33 of 208 sessions landed on the homepage. Most landed on long-form articles and comparison pieces: software comparisons such as Jobber vs Housecall Pro, lists of CallRail alternatives, and Amazon advertising explainers. These are informational queries from people doing research. Some of them run businesses we could help. Most are just reading.
That is not a knock on Google. We published roughly 150 long-form articles in September on purpose, and the site now carries more than 190 of them. Informational content is how a young domain earns impressions, indexing and trust, and we would do it again. But it means the two channels are not sending the same kind of person, and part of the lead rate gap is simply that difference in intent.
So tag your landing pages by intent (informational, comparison, service, business type) and report lead rate by intent as well as by channel. Channel alone will mislead you. A Google visitor on a service page may convert just as well as a ChatGPT visitor; Google just sends far fewer of them to that page right now.
What Search Console adds.
Search Console over the same 90 days shows the Google side ramping. Impressions ran around 300 per week in August and reached about 12,000 per week by late September, right after the September publishing push. Clicks went from 2 to 4 per week to 24 to 27 per week. 372 pages recorded at least one impression. For a domain this young, that curve is the whole point of the article program.
The commercial terms are not there yet. Queries like amazon ads agency, ai marketing consultant and franchise marketing agency sit at average positions between 20 and 60. Google is sending us informational clicks because that is where we rank. The service and business-type pages that actually convert are still climbing, and on Google that climb takes months of links and age.
Here is the interesting part. ChatGPT is already citing those same service and business-type pages while Google has them on page three. Assistants appear to weigh domain authority less and page specificity more: does this page clearly and directly answer the prompt. So do not wait for Google rankings to validate a specific page. If an assistant is citing it and sending people who convert, it is doing its job already.
AI citations, per Ahrefs Brand Radar.
Ahrefs Brand Radar on October 16, 2026 showed 22 AI citations of theoryroad.com across platforms. ChatGPT accounted for 15 citations across 11 pages. Gemini showed 3 citations across 3 pages. Google AI Overviews, Perplexity and Copilot each showed 0. That is a thin footprint, and we are not pretending otherwise.
The citation counts line up with the GA4 referral data, which is reassuring on both tools. ChatGPT is the only assistant sending meaningful traffic, Gemini is a trickle, and the others have not picked us up at all. Eleven cited pages producing 77 sessions and 4 leads in a quarter is a small but real return on pages that each took an afternoon to write.
Pull your own cited pages list from Brand Radar or a similar tool and lay it next to your GA4 landing page report. The overlap tells you which pages the assistants trust and send people to. Write more pages shaped like those, and update the ones already cited so they stay the best answer.
Caveats, stated plainly.
- 4 leads versus 5 leads is a tiny sample. One more or one fewer lead on either side moves the rates a lot. Do not quote 5.2% as a benchmark.
- One quarter, one site. Our site is a marketing agency with more than 190 articles, which skews Google traffic informational while ChatGPT sends people who already asked a buying question.
- GA4 attributes chatgpt.com by referrer. Visits from ChatGPT apps or surfaces that strip the referrer land in Direct, so AI traffic is undercounted, not overcounted.
- GA4's AI Assistant channel grouping is new, and the source tags do not sum perfectly to the channel total (chatgpt.com, chatgpt.com/organic and chatgpt.com/referral are counted separately).
- Direct is excluded as bot inflated, but we did not hand-audit every one of the 563 sessions. Some real humans are in there and some bots may have leaked into other channels.
None of this means ChatGPT beats Google. It means that on one small site, in one quarter, AI assistant visitors behaved like buyers at roughly twice the rate of Google organic visitors, and the reason is visible in where they landed. Treat it as a hypothesis to test on your own data. Then run the same pull on your site and see if it holds.
What wins AI citations.
The pages ChatGPT cited and sent traffic to share a shape. Each one is about one thing: one business type, one service, one platform, one question. Each one says who it is for in the first paragraph, explains how we work in plain language, and does not pad. Several carry FAQ blocks and schema. None of them is the homepage, and none of them is a slogan with a contact button underneath.
That is the same list that wins Google. A clear entity (who Theory Road is, where it is, who runs it), one specific page per business type and per service, FAQ and structured data, fast indexing so new pages get crawled quickly, and real author pages like Travis Moher so a model can tell a person with a track record wrote it. We lay out the full checklist in how to get found by AI search.
Two things are AI specific. First, robots.txt has to let the AI crawlers in. If GPTBot and OAI-SearchBot, Google-Extended and PerplexityBot are blocked, you will not be cited, full stop, no matter how good the page is. Second, publish an llms.txt at your root that lists your key pages and what each is for. It takes ten minutes and it is exactly the kind of structured hint assistants respond to.
“An assistant does not reward the biggest site. It rewards the page that answers the question with the least noise.”
Start with two actions. Run a crawler check on your robots.txt today, then write one business-type page for your best customer segment this week. That is the whole starting playbook. It is also the core of our SEO and AI search service when a team would rather hand the work off than learn it.
How we measure it.
Nothing here needs a paid analytics stack beyond the Ahrefs seat, and even that is optional. Here is the exact setup so you can replicate the pull on your own site in under an hour.
If your form events are not firing cleanly, none of this works, and most sites we open have at least one broken or duplicated conversion. Our GA4 and Tag Manager page covers how we set the events up so channel reports can be trusted.
What we are doing next.
The data is small, so the plan is to make the pages that work more numerous and the measurement better, then rerun this in 90 days. Here is what is on the list.
- More business-type pages. The /for/ section pulled about 20 of 77 ChatGPT sessions from 13 pages. We are adding pages for every segment we actually serve.
- FAQ blocks with schema on every service, platform and business-type page, written as the questions a prospect would type into an assistant.
- Visible updated dates on every page, kept honest. Assistants and Google both favor pages that are demonstrably maintained.
- More data pieces like this one. Original numbers get cited; opinions get summarized without a link.
- Author bylines on every article pointing to a real author page, so the entity behind the content is unambiguous to a model.
We will republish this with the next 90 days of data whether the gap holds or collapses. If you want the same pages, events and measurement built for your business, Hire Theory Road.
How do I see ChatGPT traffic in GA4?
Open Reports, then Traffic acquisition, and look for the AI Assistant channel in the default channel grouping. If it is missing, create a custom channel group with a rule matching session source to chatgpt.com, gemini.google.com, perplexity.ai, copilot.microsoft.com and claude.ai. Then build an Explore table with session source and landing page so you can see which pages the assistants actually send people to.
Does blocking GPTBot in robots.txt matter?
Yes. If GPTBot and OAI-SearchBot are blocked, ChatGPT cannot read or cite your pages, and the referral traffic described here will not exist. The same goes for Google-Extended and PerplexityBot on their platforms. Check your robots.txt and any firewall or CDN rules that block crawlers by user agent, because a hosting provider may block them by default without telling you.
How many visits from ChatGPT is normal?
For a small business site, very few. Our agency site saw 77 ChatGPT sessions in 90 days, roughly six a week, against 208 from Google organic. Treat AI assistant traffic as a small, high-intent channel rather than a volume channel. The useful question is not how many visits you get but which pages they land on and whether those visitors start your form at a higher rate than other channels.
How do I get cited by ChatGPT?
Publish specific pages that each answer one question completely: one page per business type you serve, one per service, one per platform. State who the page is for in the first paragraph, add FAQ blocks and schema, show a real author, and keep updated dates honest. Allow AI crawlers in robots.txt and publish an llms.txt. Then watch Ahrefs Brand Radar or your GA4 landing pages to see which pages get picked up.
Is AI search traffic worth optimizing for yet?
Yes, because the work is the same work that wins Google. Specific pages, clear entity, FAQ and schema, author pages and fast indexing help both channels, so there is no separate budget to justify. Our data shows AI visitors converting at about twice the Google organic rate on a tiny sample. The volume is small today. The pages you build for it will keep paying in Google while the AI channel grows.