Sauce · AI search measurement
AI traffic looks flat. The customers inside it may not be.
Quick answer
If you run a local service business, count the leads AI may send you, not only the clicks GA4 records. A study of 73 online stores found AI visits flat for a year while orders from those visits nearly tripled. That is store data, not service-business data. GA4 can miss AI visits that arrive with no referrer, and it files Google’s own AI answers under Organic Search. Ask every lead how they found you, then compare their answers with GA4.
Want your AI customers counted, not guessed?The free scan checks the facts every platform reads.
Get the free scanFor a service business, the question is whether AI is sending calls, forms and bookings. Shero’s study of 73 online stores shows a pattern worth testing, and this post turns it into a check you can run on your own numbers.
What the study found
Shero published the study on September 28, 2026. It used Google Analytics 4 data from 73 online stores that the agency manages or supports. Together, the stores logged 96.4 million sessions and 976,668 orders in 2026. Shero counted visits from ChatGPT, Gemini, Perplexity, Copilot, Claude or a similar assistant as AI visits, and counted a GA4 purchase as an order. INDEPENDENT RESEARCH
Visits barely moved. Orders from those visits rose from 5.9 a day to 16.1, about 2.7 times as many. ChatGPT’s conversion rate rose from 0.69% to 1.77%, while the other assistants combined rose from 0.34% to 1.38%. Visitors did not browse more: the engagement rate stayed around 60%. They bought more often. INDEPENDENT RESEARCH
Shero also compared AI visitors with Google organic visitors. On 41 of the 61 stores with enough AI traffic, AI visitors converted better. For the median store, the conversion rate was 25% higher and revenue per visit was 23% higher, though orders were 15% smaller. Shero notes that its organic figures include branded searches, which usually convert better, making this a tougher comparison for AI. INDEPENDENT RESEARCH
What the study does not show
Flat visits can hide a growing channel. They can also hide a very small one.
Why clicks are the wrong thing to count
A store counts orders. A service business needs to count leads: a call, a form, a text, a booking or a request for directions. Neither Shero’s study nor IDHL’s counts phone calls. DHM INTERPRETATION
Three things can make a lead hard to see in a click count. First, a visit may arrive with no referrer. Shero says taps on links in the ChatGPT mobile app often arrive that way and are counted as Direct. INDEPENDENT RESEARCH Second, a customer may read an answer, then search your business name. GA4 records that as Organic Search. INDEPENDENT RESEARCH Third, the AI answer may come from Google itself. GA4 puts visits from Google’s AI Overviews and AI Mode in Organic Search, not in its AI Assistant channel. GOOGLE DOCUMENTED
Even a complete click count has limits. IDHL says a visit counts as AI only if the person clicked a link inside the answer and accepted cookies on the site. INDEPENDENT RESEARCH Pew Research found that people who saw a Google AI summary clicked a result link in 8% of visits, compared with 15% for searches without one, in March 2025 browsing data. INDEPENDENT RESEARCH When fewer people click through from an answer, more customers remain uncounted as website visits.
| What the customer does | How GA4 records it | Source |
|---|---|---|
| Clicks a link in an assistant such as ChatGPT or Gemini, and the link passes a referrer | AI Assistant channel, new on May 13, 2026 | Google release notes |
| Taps a link in the ChatGPT mobile app | Often Direct, because no referrer arrives | Shero’s stores |
| Reads a Google AI Overview or AI Mode answer, then clicks | Organic Search, not AI Assistant | Google channel definitions |
| Reads a recommendation, then searches your business name | Organic Search | Shero’s stores |
| Reads an answer, then phones | Not a website visit. Only your call log or call tracking sees it | DHM interpretation |
| Taps Call or Directions on your Business Profile | Business Profile metrics, which you can link into GA4 | Google help pages |
Check your own gap
This quick check uses your own numbers. Pull your leads and website sessions for the last 90 days. Enter how many leads you asked “How did you hear about us?”, how many said an AI assistant, your total GA4 sessions and the sessions GA4 labelled AI Assistant.
Your AI gap check. The numbers are examples. Replace them with yours.
- Customers say AI sent10.0%6 of 60 leads
- GA4 labels as AI0.5%9 of 1,800 sessions
- Customer share against GA4 share20xCustomers credit AI far more often than GA4 does. Some AI customers reach you by routes GA4 cannot label.
The starting numbers are made-up examples, not data from any business. Replace all four.
In Shero’s stores, AI’s share of orders (0.26%) was about 1.2 times its share of sessions (0.21%). INDEPENDENT RESEARCH These are store figures, so use them as a reference, not a target. If customers credit AI at more than about twice the rate GA4 labels it, some AI customers may be reaching you by routes GA4 cannot label. The twofold line is our rule of thumb, not a published threshold. Small counts can shift the result, so use 90 days or more. DHM INTERPRETATION
The page the click lands on
The studies disagree about where an AI click lands. In Shero’s stores, 51.0% of AI visits started on a product page and 22.8% on the homepage. INDEPENDENT RESEARCH SE Ranking found 60% of AI-referred traffic landing on homepages, compared with 17% for organic search. INDEPENDENT RESEARCH Similarweb reports that ChatGPT’s homepage share rose from roughly 26 to 29% to about 62 to 63% after the May change that made brand names clickable. INDEPENDENT RESEARCH
| Where AI visits start in Shero’s stores | Share of AI visits | Converts at |
|---|---|---|
| Product page | 51.0% | 1.77% |
| Homepage | 22.8% | 0.57% |
| Blog or guide | 5.1% | 0.29% |
| Collection page | 4.2% | 1.78% |
The studies use different samples and months, and their findings differ. For a service business, an AI click may land on your homepage or a service page. Either page should answer the customer’s question. DHM INTERPRETATION
Shero advises stores to answer the shopper’s question in plain text on the page: size, price, stock, delivery and returns. A service page needs its own details: what the job is, where you work, what affects the price, when you answer the phone, what you stand behind, and how customers can reach you. Tick each detail your main service page already includes.
Does your main service page answer what a customer asked?
What it says up front
The facts a buyer asks about
What happens next
Tick every line your main service page already says
A DHM checklist based on Google’s guidance and Shero’s advice for product pages. It is not a Google score. Nine or ten ticks looks strong, six to eight means fixable gaps, and fewer means the page is thin.
Do this week: count it in six steps
- Check the AI Assistant channel. In GA4, open Reports, then Acquisition, then Traffic acquisition, and set the primary dimension to Session default channel group. Google added the AI Assistant channel on May 13, 2026. It covers sources such as ChatGPT, Gemini, Deepseek, Copilot and Grok, and it excludes Google’s AI Overviews and AI Mode. GOOGLE DOCUMENTED
- Build your own list if you need one. Google’s help page explains how to make a custom channel group with a “matches regex” condition. It says custom channel groups can be applied to past data. GOOGLE DOCUMENTED GA4 matches the whole value by default, so put
.*before and after the names, for example.*(chatgpt|openai|perplexity|gemini|copilot|claude|grok|deepseek).*. Move the channel above Referral, because GA4 assigns each session to the first channel it matches. GOOGLE DOCUMENTED This pattern is our example, not an official list. - Keep a year of data. In GA4, open Admin, then Data Settings, then Data Retention, and set event data retention to 14 months. This affects Explorations and funnel reports, not the standard reports. GOOGLE DOCUMENTED Shero recommends the same setting so a year-on-year comparison is possible.
- Link your Business Profile. In Admin, open Product links, then Google Business Profile links. This brings calls, directions, website clicks, messages and bookings into GA4 for the last six months, as totals. GOOGLE DOCUMENTED
- Ask every lead. Add one required question to every form and call script: “How did you hear about us?” Include “AI assistant” as its own option, alongside Google search, Google Maps, referral, social and other. Record the answer in your CRM or a sheet, and run the gap check above every quarter.
- Write the dates down. Note the May 2026 ChatGPT link change and the GA4 channel launch in your reports, so a jump or a drop is not mistaken for a trend. Our note on ChatGPT referral tagging covers the October reports.
What our own numbers show
We read our own GA4 on October 20, 2026. Between July 22 and October 19 the site had 137 sessions: Direct 96, Google organic 30, referral 8 and organic social 3. No session carried an AI Assistant label or an AI source such as chatgpt.com. GA4 also recorded no key events. DHM DATA
With so few sessions, we cannot tell whether AI sends us customers, and a gap check on our own numbers would be noise. That is why we ask: the answer has to come from the people who book, not from GA4. DHM DATA
What not to do
- Do not treat Direct as AI. Direct includes typed addresses, bookmarks, links in apps and anything that arrives with no referrer. A rise in Direct is a reason to ask customers, not evidence of AI visits.
- Do not buy an AI visibility score before you can count leads. Shero’s test is that if a score rises while AI orders do not, the score is not measuring anything that makes money. INDEPENDENT RESEARCH Google also warns that no third-party tool has access to its internal ranking or AI systems. GOOGLE DOCUMENTED
- Do not write a page for every prompt. Google says creating separate content for every variation of a query, mainly to influence rankings or AI responses, violates its scaled content abuse policy. GOOGLE DOCUMENTED
- Do not copy a store’s conversion rate as your target. 1.77% is an online-store figure for September 2026. INDEPENDENT RESEARCH Your lead numbers will differ.
- Do not promise yourself a mention. No one can guarantee that an assistant names a business, and we do not.
None of this changes how local SEO works. A complete, accurate Business Profile, real reviews and clear pages still come first. What changes is how you read the numbers. This week’s Update adds the wider data from IDHL, SE Ranking and Similarweb, and where local AI answers differ from the Map Pack. For tracking calls, see our call tracking guide.
Sources
- Shero Commerce (Gentian Shero), “What 96 Million Sessions Show About AI Referral Traffic and Conversions”, September 28, 2026: the 73-store sample, AI visits and orders, conversion, landing pages, the top-10 concentration and Shero’s own limits.
- SE Ranking (Yulia Deda), “Referral Traffic from ChatGPT Hit an All-Time High in May 2026”, July 9, 2026: the 101,574-site sample, the May step change and where AI-referred visits land.
- Similarweb, “AI Search Stats in 2026”, July 29, 2026: the May 2026 brand-link change and the homepage share of ChatGPT referrals.
- IDHL, “The State of AI Search 2026”, September 2026: the definition of an AI referral session, including the cookie condition.
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, July 22, 2025: click rates with and without a Google AI summary, March 2025 browsing data.
- Google Analytics Help, “Analytics release notes”, entries dated May 13 and June 8, 2026, accessed October 20, 2026: the AI Assistant channel and the Business Profile link.
- Google Analytics Help, “Default channel group”, accessed October 20, 2026: the AI Assistant, Organic Search and Direct definitions.
- Google Analytics Help, “Custom channel groups”, accessed October 20, 2026: the AI assistants example, channel order and past-data rule.
- Google Analytics Help, “About regular expressions (regex)”, accessed October 20, 2026: GA4 matches the whole value unless the pattern uses wildcards.
- Google Analytics Help, “Data retention”, accessed October 20, 2026: the 2-month and 14-month event data options and the path to the setting.
- Google Analytics Help, “Connect Google Business Profile (GBP) to Google Analytics”, accessed October 20, 2026: the link steps, metrics and limits.
- Google Search Central, “Optimizing your website for generative AI features on Google Search”, last updated July 10, 2026: the scaled content abuse policy and the warning about third-party tools.
- OpenAI Help Center, via the Internet Archive, “Publishers and Developers FAQ (archived copy)”, captured October 4, 2026: the statement that ChatGPT adds utm_source=chatgpt.com to referral links. The live address now redirects to another article.
- Pexels, photo by Pixabay, “Woman in gray vest with yellow hard hat inside room”, accessed October 20, 2026: hero image.
- Pexels, photo by kaboompics.com, “Overhead shot of a person having a phone call”, accessed October 20, 2026: inline image.
- Pexels, photo by Rene Terp, “Smiling man in workwear holding tools”, accessed October 20, 2026: inline image.
Shero’s study covers online stores, and applying it to service businesses is our inference. Shero, IDHL, SE Ranking and Similarweb are third-party analyses, and the agencies’ samples are their own clients. Our own figures are Google Analytics 4 (July 22 to October 19, 2026), read on October 20, 2026. The two calculators use only the numbers you type in, and their example values are made up. The regex is our example, not an official list. Search Engine Land’s October 15 report was not re-opened today and is covered in our October 16 note.
