GEO strategyBy 6 min read

AI recommends your brand but the visit comes from Google: how to measure it

Similarweb: after a ChatGPT recommendation, visits to a brand rise 2.5x, yet 56% arrive through a Google search. What Analytics misses and how to measure it.

Cover: AI recommends your brand but the visit comes from Google, how to measure the downstream effect

In short

According to Similarweb, people who get a brand recommendation from ChatGPT are 2.5 times more likely to visit its website within seven days, but 55.9% of those visits arrive through a Google search and only 8.8% through the AI channel. Google Analytics logs them as organic or direct traffic and the recommendation disappears from your reports. To measure the effect, place your share of presence in AI answers next to branded searches in Search Console and new-user direct visits, with a one- or two-week lag.

  • Similarweb study (US desktop panel, July-December 2025, finance, travel and beauty): after a ChatGPT recommendation the probability of a visit within 7 days is 2.5 times higher than for the competitor that was not recommended.
  • 55.9% of AI-influenced visits arrive through search (against 40.4% of the others) and only 8.8% through the AI channel: people search for the brand name on Google and Analytics credits Google.
  • These visits are worth more: 12 pages and 11.8 minutes on average against 6.5 pages and 5.6 minutes.
  • To measure the downstream effect: branded searches in Search Console, a fixed list of monitored prompts, new-user direct visits, a 'how did you hear about us' question, a 7-14 day lagged comparison and a control term.
  • The limits: correlation not proof, large US brands, alternative causes of branded searches. The method works for everyone; the numbers must be measured on your own site.

Written by the Geosnap AI agent, reviewed and approved by Rinald Sefa.

Translated from the Italian original. Read the original

Your brand shows up in ChatGPT's answers, yet in Google Analytics the "AI" channel is still a tiny line. That is where many teams start to doubt: is AI visibility worth anything if nobody clicks? Similarweb's study The Downstream Impact of AI Visibility answers with real browsing data: the visits do come, but through a different door. Here is what it says, where it needs caution, and how to measure the effect on your own site.

What happens in the seven days after an AI recommendation?

People who get a brand recommendation from ChatGPT are 2.5 times more likely to visit that brand's website in the following seven days than people who do not. That is the headline finding of the Similarweb study published on 23 June 2026, based on a panel of US desktop users between July and December 2025 in three industries: finance, travel and beauty.

The method is simple. Similarweb took pairs of competing brands, isolated the people who asked ChatGPT an industry question and received a recommendation for one of the two, excluded anyone who had named the brand in the prompt or visited it in the previous four weeks, and tracked where they went over the next seven days. As Search Engine Land reports, in finance 14.2% visited the recommended brand's site versus 3.8% for the competitor; in travel 12% versus 3.4%; in beauty 7.9% versus 3.3%.

Two details matter. First, even in the best case most people visited neither site within a week, so the recommendation shifts the odds rather than guaranteeing a visit. Second, the study covers large consumer brands whose audience already knows them through other channels.

Where do these visits come from, if not from ChatGPT?

Mostly from search: 55.9% of AI-influenced visits arrive through a search engine, against 40.4% of ordinary visits. In most cases the person closes ChatGPT, opens Google, types the brand name and clicks the first result. In Analytics that visit lands under "Organic Search" or, if you run a campaign on your own name, under "Paid Search". The AI channel accounts for 8.8% of influenced visits, versus 5% of the others: growing, but still the tail of the phenomenon.

Arrival channelAI-influenced visitsNon-influenced visits
Search (organic and paid)55.9%40.4%
Direct19.9%38.8%
Referrals from other sites13.5%10.5%
AI assistants8.8%5.0%

Source: Similarweb, US desktop panel, July-December 2025, as reported by PPC Land. These visits are also worth more: 12 pages viewed and 11.8 minutes on site on average, against 6.5 pages and 5.6 minutes for non-influenced visits.

This fits another Similarweb figure: between mid-2025 and January 2026, visits to AI assistants climbed toward 1.5 billion a month while referrals to websites stayed flat between 240 and 280 million, with an outbound click rate of 1.6-2.5% in Google AI Mode against 17-19% for classic search (AI Search Trends, Similarweb). The AI answer closes the conversation; the visit, when it happens, is born later and elsewhere.

Why your analytics cannot see it

Because last-click attribution credits the channel that delivered the click, not the one that created the interest. If someone discovers your brand in a Gemini answer and searches for it on Google the next day, Google Analytics 4 records "google / organic" and the AI recommendation vanishes from the report. NP Digital puts it plainly: most AI-influenced visits do not arrive with a recognizable AI referral, so anyone looking only at the AI channel underestimates the effect.

Add the labelling problems already known: some ChatGPT redirects strip parameters and the traffic ends up as direct visits (we covered it in this article). And the picture keeps moving: in mid-November 2026 Search Engine Roundtable reported ChatGPT tests with layouts made of web results only, or with the answer collapsed behind a "See more" button, and Similarweb notes that tests with more prominent outbound links roughly tripled outbound traffic, from small volumes. The AI channel in Analytics will grow, but it will never be the complete measure.

How to measure the downstream effect on your site

You need two data series side by side: how often your brand appears in AI answers, and how searches for your name move. Here is a procedure that works for any site, even without paid tools.

  1. Isolate branded searches in Search Console. In the Performance report, filter queries containing your brand name (and its common misspellings) and log weekly impressions and clicks. It is the most direct downstream signal: Similarweb and NP Digital both name it as the first thing to track alongside AI visibility.
  2. Keep a fixed list of prompts and record your share of presence. Pick 20-50 questions a customer would ask without knowing you ("what is the best software for...", "who should I turn to for...") and check every week on ChatGPT, Gemini and Google AI Mode whether you appear and in what position. Geosnap automates this step and places your mention share next to Search Console data; done by hand, a spreadsheet with date, engine, prompt and answer is enough.
  3. Watch direct visits from new users on the home page. In GA4 build a segment "new users, Direct channel, landing page home": that is how people arrive when they have memorized the name and type the address. A rise with no offline campaign running is a clue.
  4. Add one question to your form. In the contact form or after checkout ask "how did you hear about us?" with the option "from an AI assistant (ChatGPT, Gemini, others)". It is the least precise measure but the only one the person states themselves.
  5. Compare the two curves with a one- or two-week lag. If your share of presence in answers rises and 7-14 days later branded searches and new-user direct visits rise too, you have a correlation you can defend in a meeting. If only the first one rises, check that the answers actually name your brand and not just your site: ghost citations do not generate branded searches.

Which mistakes to avoid when reading these numbers

The most common mistake is turning a correlation into proof. Rand Fishkin, who collaborated on the study, reminds readers on SparkToro that the data is aggregate, covers big brands and cannot say whether those people would have found the brand anyway; a B2B company or a local brand has to verify the effect in its own data. Here are the other points to check.

  • Branded searches have many causes. Advertising, press, seasonality, a viral post. Use a control term: category searches without your name, or searches for a competitor that does not appear in AI answers.
  • A US desktop panel is not your audience. The study's percentages describe three American industries. Take the mechanism (AI recommendation, then branded search), not the numbers, and measure your own.
  • The AI channel in GA4 is not the measure of AI visibility. It measures the clicks assistants let through, which is a small fraction. For the channel's real weight against Google and its conversion rate, read the 2026 data on AI traffic.
  • Defend the results page for your own name. If more than half of AI-influenced visits pass through a branded search, that Google page is where the recommendation turns into a visit: make sure the first result is yours, the brand panel is correct and no competitor is buying your name with an ad.

The goal is a dashboard with three rows side by side, week by week: presence in AI answers, branded searches, new-user direct visits. It is not perfect attribution, but it is enough to decide whether AI visibility is creating demand for you or for the competitor the AI recommends in your place.

Sources

Frequently asked questions

If traffic from ChatGPT is small, is AI visibility still worth it?

Yes, because the measurable traffic from the AI channel is only part of the effect. In the Similarweb study 8.8% of AI-influenced visits arrive with an AI referral, while 55.9% pass through a search for the brand name. The AI channel in Analytics measures the clicks assistants let through, not how many people learned about the brand from an answer.

How do I tell branded searches generated by AI from those generated by advertising?

You cannot with certainty, but you can narrow the doubt. Compare branded searches with a control term (category searches without your name, or the name of a competitor the AI does not cite), log the dates of campaigns and press coverage, and check whether the changes follow shifts in your share of presence in AI answers by 7-14 days.

Which tools do I need for this measurement?

Google Search Console for branded searches, Google Analytics 4 for new-user direct visits, and a spreadsheet to record every week whether you appear in the answers to a fixed list of prompts. A monitoring tool such as Geosnap automates the prompt checks on ChatGPT, Gemini and Google AI Mode and places them next to Search Console data.

Do the study's results apply to a B2B or local brand?

It is not proven. The study covers large US consumer brands in three industries, and Rand Fishkin, who collaborated on it, warns against extending the numbers to other contexts. The mechanism (AI recommendation, then a search for the name) is plausible for any company, but the size of the effect has to be measured in your own data.

What if a competitor buys my brand name on Google Ads?

They intercept part of the visits the AI recommendation generated for you, because more than half of those visits pass through the results page for your name. Check that page regularly, make sure the first result is yours, and consider a brand protection campaign if you see other people's ads.