The number we kept quoting

ChatGPT now sends our store more traffic than Google

In the 28 days to 1 September 2026, ChatGPT sent the store 1,770 visits and Google sent 1,013. That is an outlier: across 13,770 websites, AI assistants send 1.08% of traffic on average (Conductor, 2026 benchmark). Below is 14 months of the measurement, month by month, including the months that undercut it.

Published 24 August 2026 · updated 10 September 2026

50.4%of sessions from AI assistants in August 2026, the full month

Context

Floreva is a skincare brand we own and operate: a live store taking real orders. When we rebuilt it we cut paid advertising by 70% and replaced it with organic and AI-assistant traffic. Ever since, we have quoted a figure, 40% of that brand's traffic now comes from AI assistants, on this site and in every proposal. A number repeated that often should be checkable, so here is the data underneath it.

The constraint

Analytics has no button for this. Google Analytics offers no 'AI assistants' channel, so these visits land in referral traffic or unassigned, and every team measuring this is building the definition themselves. That makes it easy to quote a flattering number and never publish the method. We would rather publish the method and let the number be judged.

What we built

Pulled session data straight from the Google Analytics Data API for the property, by month and traffic source, for the 14 months from July 2025 to 31 August 2026. No sampling, no dashboard screenshots. The pull is re-run when a month closes and the table below is replaced, not appended to.
Counted a session as AI-assistant traffic when its source matched a known assistant: chatgpt.com, claude.ai, perplexity.ai, gemini.google.com, copilot.com and the other named engines. Nothing inferred, nothing estimated.
Left unattributed traffic out of the AI column rather than allocating it. In the last 28 days 11.3% of sessions arrived with no usable source, and assigning any of it to AI would have flattered the result.
Corrected one row. The first version of this page, pulled on 24 August, showed July 2025 as 910 sessions with 1 from an assistant. The re-run on 2 September, same API, same property, returned 1,034 and 10 for that month, and we could not reproduce the earlier figures. The table now carries the later pull for every month, one method throughout, and the difference is noted here rather than quietly overwritten. It moves the first month from 0.1% to 1.0% and changes nothing about the trend.

The data

AI share per month, drawn from the same rows printed below. The dark point is the highest month.

Month	Sessions	From AI	AI share
Jul 2025	1,034	10	1.0%
Aug 2025	2,052	14	0.7%
Sep 2025	695	56	8.1%
Oct 2025	765	152	19.9%
Nov 2025	259	103	39.8%
Dec 2025	1,324	60	4.5%
Jan 2026	663	122	18.4%
Feb 2026	222	77	34.7%
Mar 2026	339	69	20.4%
Apr 2026	2,296	562	24.5%
May 2026	3,162	949	30.0%
Jun 2026	2,364	861	36.4%
Jul 2026	2,140	880	41.1%
Aug 2026	3,688	1,857	50.4%

Sessions by month, and the share arriving from AI assistants. Source: Google Analytics Data API, pulled 2 September 2026. Full months only.

The result

In the 28 days to 1 September 2026, ChatGPT sent the store 1,770 sessions and Google sent 1,013. A single AI assistant now delivers more traffic than the search engine, 51.0% of all sessions against Google's 29.2%. Across the 14 months the share went from 1.0% to 50.4%, and it did so while monthly traffic grew about three and a half times, so this is AI traffic added rather than search traffic relabelled.

What we'd do differently

We would have instrumented this from the first day instead of reconstructing it a year later. The trend was already visible in late 2025 and we only started measuring it deliberately once the number got big enough to argue about. Every client site we build now tags AI referrals from launch.

Why this matters for you: Start with scale. Conductor's 2026 benchmark measured 13,770 websites and found AI assistants send 1.08% of their traffic on average, so read our figure as an outlier, not a forecast. Then two honest caveats before anyone treats it as a law of nature. First, this is effectively one assistant: 3,512 of the 3,524 AI sessions from June to August 2026 came from ChatGPT, so this is concentration, not diversification, and it could move against us as fast as it moved for us. Second, it is one brand in one market, and the monthly line is volatile, December 2025 shows 4.5% because a burst of non-AI traffic diluted the share, not because AI traffic collapsed. What we take from it is narrower than the headline: for a small consumer brand, being readable and quotable by AI assistants stopped being a nice-to-have some time in early 2026. We keep saying 40% on this site rather than 50% because the last three months together come to 43%, and a single month is a peak until the next one confirms it.