ChatGPT shoppers convert better than Google searchers, or do they?

Search Engine Land ran a piece in February on a Visibility Labs study that has been doing the rounds ever since. The finding was that ChatGPT traffic converts at 1.81% across 94 ecommerce sites, compared to 1.39% for non-branded organic search. That's a 31% lift, and it held up across 10 out of 12 months in 2025. 

First instinct, looking at a number like that, is to start reshuffling priorities. If LLMs convert better, the logic goes, we should be optimising for them. Job done. But the conversion rate is one aspect of the data, and the rest of it tells a much more measured story.

This article unpacks the data, the user behaviour behind it, and what it means for your website and marketing strategy.

Search behaviour is shifting, including here in Australia

What is genuinely new is how people are using ChatGPT before they buy. Instead of typing "best running shoes for flat feet" into Google and wading through ten product roundups, users are having a back-and-forth conversation. They tell it they overpronate, want something for half marathons, prefer American brands, and have a budget. 

They refine the brief in real time. By the time they get a recommendation, they're not browsing anymore. They're checking something specific.

The shift is showing up clearly in Australian data. From :

• 48% of Australians have already used AI assistants for online shopping searches
• 66% of under-45s are doing the same
• 78% expect AI assistants to become a mainstream part of online shopping
• 53% plan to use them in the next 12 months

A separate Cint/Omnisend survey put current usage at 55%, with 40% preferring ChatGPT over traditional search for product research. The behaviour change is real, and it is happening on local screens.

Visibility Labs called this "intent compression". The buyer has done most of their thinking inside the chat window, so by the time they click through to your product page, they're already past awareness and consideration. That is why they convert at a higher rate.

What most write-ups leave out is what users do after they get a recommendation. A lot of them do not click the link in ChatGPT. They open Google and search the brand or product directly to read reviews, check delivery, and verify the price. The conversion then shows up in Google Analytics as branded organic search, even though the journey started in an LLM. 

“Your branded traffic might be quietly absorbing a chunk of ChatGPT-influenced revenue, and your reports will not show it.”

The data is real. The volume is not.

Here is the part that tends to get glossed over. ChatGPT might convert 31% higher per session in the Visibility Labs study, but in the same dataset, non-branded organic traffic was 70 times the size of ChatGPT overall, narrowing to about 47x by Q4. ChatGPT contributed 1.48% of organic revenue across the 94 sites, lifting to 2.2% in the second half of the year.

Independent academic research from October 2025 paints an even more sober picture. Maximilian Kaiser and Christian Schulze at the Frankfurt School analysed 973 ecommerce sites and over 50,000 ChatGPT transactions against 164 million transactions from traditional channels, in a working paper on SSRN covered by Digiday and Modern Retail. 

They found ChatGPT accounted for less than 0.2% of total ecommerce traffic and that, on most metrics, it underperformed every traditional channel except paid social. They projected that LLMs would not reach parity with organic search within the next year.

Two studies, two slightly different framings, both useful:

So the channel is converting well, growing fast (ChatGPT visits in the Visibility Labs dataset grew over 1,000% across 2025), and is still small. All three are true at the same time.

What we're seeing across our client base

The data tracks with what we are seeing in our own work. LLM traffic is accounting for less than 5% of total visits across most of our clients. Some conversions can be attributed directly to those sessions through the custom Google Analytics channels we have set up. Plenty more cannot, because users routinely hop from ChatGPT to organic search mid-journey and the attribution chain breaks down right where it matters.

There is also a brand size effect that the headline studies tend to underplay. The Visibility Labs and Frankfurt School samples are both weighted toward seven and eight-figure ecommerce brands, and that matches what we see locally. The clients showing up inside ChatGPT recommendations tend to be the ones already established in their category, with consistent marketing spend, recognisable share of voice, and a track record of being mentioned online. 

Smaller ecommerce businesses are not being surfaced at the same rate, because there is less for an LLM to draw on when forming a recommendation. AI referral is not a great equaliser at this stage. If anything, it rewards the brands already winning attention through other channels.

Shopping inside ChatGPT keeps shifting

OpenAI launched Instant Checkout in September 2025, letting shoppers buy products discovered in ChatGPT without leaving the chat. In March 2026, the company quietly scaled it back, routing purchases to retailer websites and apps instead, .

The retreat does not mean the commerce play is dead. ChatGPT is still becoming a discovery and shortlisting layer above the open web. It does mean retailer sites are back at the centre of the actual conversion event.

The broader momentum has not slowed. Adobe reported that ecommerce traffic from generative AI platforms grew 693% across November and December 2025 compared with the same period the year before, . The catch is trust. 

PayPal's research found 92% of Australians have at least one concern about using AI when shopping online, with privacy and data security topping the list at 64%. People are getting comfortable using AI to research. They are still working out whether they trust it enough to hand over their card.

What this means in practice

For most Australian ecommerce businesses in 2026, ChatGPT is sending you well under 5% of your traffic and contributing to even less of your direct revenue. The number you see in Google Analytics is also probably understating its real influence, because so many of those journeys end on a branded Google search rather than a direct ChatGPT click.

A few practical things worth doing now if you have not already:

1. Set up a custom channel grouping in Google Analytics for AI referral sources. That includes chatgpt.com, openai.com, perplexity.ai, claude.ai, and gemini.google.com. Default GA lumps these in with general referral traffic and you will not see them clearly until you separate them out.
   
   
2. Add a post-purchase survey asking customers how they found you. It is the only reliable way to catch the ChatGPT-then-Google journeys that GA misses by design.
   
   
3. Make sure your product information is clean, accurate, and crawlable. If ChatGPT is going to recommend you, it needs current prices, correct specs, and live availability. Stores that get high bounce rates from AI traffic usually have a mismatch between what AI says about them and what shoppers actually find on the page.
   
   
4. Watch the channel quarterly, not weekly. Growth is real but volatile. The 1,000%+ traffic spikes everyone is quoting come off tiny baselines, and individual months can swing wildly until volume builds.
   
   
5. Be cautious of third-party tools claiming to show you LLM search data. We have evaluated a number of these, and the reality is that LLMs do not publicly share the queries users are running. The "data" on offer is reverse-engineered and aggregated rather than direct, useful as a directional signal, not as a planning input. There is no equivalent of Google Search Console for ChatGPT yet.
   

One thing we have learned doing this for our own marketing as well as for clients is do not lean on quantitative data alone. Even with full HubSpot attribution on our agency leads, we still ask every prospect how they heard about us. 

The qualitative answer regularly fills in gaps that no analytics dashboard captures, especially for the LLM-influenced journeys that end somewhere else before converting. PayPal is starting to bake similar capture into its checkout flow, which is a good signal of where this is heading.

The behaviour change is the headline, not the conversion rate.

Australians are using LLMs as a research layer above traditional search, and that layer is going to keep growing whether or not the in-chat checkout features stick this time around. 

The brands that come out of the next two years in good shape will be the ones with proper measurement set up early, who treat AI referral as a genuine channel worth understanding, not the ones who tear up their SEO strategy on the back of a single 31% conversion stat from one study.

For our part, we have built AEO and GEO audits into our standard SEO work for clients alongside the traditional ranking factors, and we are tracking AI referral as a separate channel across the accounts where we manage analytics. That is how we are thinking about it at the moment, and we will keep watching the data as it develops.