AI Search Sends Less Traffic Than You Think. Fix Your Catalogue Anyway.

The Number Nobody in This Category Leads With

If you sell software to Shopify merchants in 2026, the easiest pitch in the world is fear: your competitors are being recommended by ChatGPT and you are not. We build a tool in this category, so believe us when we say we know how tempting that pitch is. Here is why we do not make it.

Measured AI referral traffic is small. Similarweb-based tracking puts AI referral traffic at roughly 0.15-0.25% of total internet traffic, against organic search at around 48.5%. Conductor's panel measures it nearer 1% of website traffic. The honest summary of the measurement landscape is: somewhere between a sixth of a percent and one percent, depending on whose panel and whose denominator - and either way, a rounding error next to organic search.

Any strategy pitched to you on the premise that AI search is already where your customers are is selling you a premise the measurement does not support. That includes strategies sold by apps that look like ours.

Three Things That Make the Small Number Matter

The share is small. Three properties of that share deserve your attention anyway.

It is growing fast from its small base. Previsible's study of 6.77 million LLM-driven sessions across 166 sites found monthly AI referral sessions grew 9.9x between November 2024 and May 2026. Growth curves from small bases are routinely over-extrapolated - but this one is real, sustained, and measured across a stable panel.

It converts unusually well. A full-year 2025 analysis of 94 ecommerce sites found ChatGPT-referred traffic converting at 1.81% against 1.39% for non-branded organic search - visitors arrive pre-qualified, because an AI already matched them to the product before they clicked. Low volume, high intent.

It is concentrated. In Previsible's data, 92.4% of trackable LLM referral traffic comes from ChatGPT alone. Whatever the AI shopping channel becomes, right now it is substantially one platform - which makes the next section the most practically useful research published on this topic.

Where ChatGPT Actually Gets Its Products

In March 2026, researchers at Peec AI published a study on Search Engine Land that quietly answered the question every merchant has been asking: how do I get my products into ChatGPT's shopping carousel?

They analysed roughly 5,000 ChatGPT carousels containing 43,000 products across ten retail categories, decoded parameters embedded in ChatGPT's own page source, and compared the carousel products against 200,000+ organic shopping results. The finding: 83% of ChatGPT's carousel products were strong matches to Google's top 40 organic shopping results. The equivalent figure for Bing was 11% - and most of those overlapped with Google anyway. Follow-up analysis found around 60% of matches came from Google Shopping positions 1-10. Paid ads and sponsored products were excluded from the analysis: the pipeline reads organic results.

The mechanism, per the study: when you ask ChatGPT a shopping question, it generates short, targeted "shopping fan-out" queries - averaging about 7 words - retrieves organic Google Shopping results for them, and builds the carousel from what comes back.

Sit with what that means. The most advanced AI shopping surface in existence is not a new discovery channel with new rules. It is, overwhelmingly, a re-ranking of organic Google Shopping - which is itself a function of your product feed and your product data.

Which Is Why the Honest Advice Points at Your Catalogue

Here is the conclusion the numbers actually support, as opposed to the one the category sells.

If AI referral traffic were already 20% of sessions, the right move would be catalogue work. Because it is under 1%, the right move is still catalogue work - because the same product data serves every channel that matters at once. The title, description, attributes, images and structured data on a product determine its organic shopping rank, its search snippet, its conversion once a human lands on it, and - per the Peec study - its eligibility for the AI carousel that reads those organic results. There is no separate "AI optimisation" to buy. There is one catalogue, read by increasingly many machines.

That ordering matters for budget. Work that only pays off if AI shopping explodes is a speculative bet. Work that pays off today through search and conversion, and also happens to be exactly what the AI pipeline reads, is not a bet at all. Complete titles with brand, model and key attributes; descriptions with extractable facts rather than adjectives; accurate prices and availability; alt text on images; no duplicate titles competing with each other. Boring, compounding, channel-independent.

We built Geopad to do precisely that work - the 13 checks we run and their exact weights are documented here - and the honest framing is the one above: catalogue quality first, AI-readiness as the free by-product.

What We Will Not Tell You

A short list of claims you will hear in this category that we decline to make, and why.

"AI search is replacing Google." The measured shares above say otherwise for commerce today. Traditional organic search remains the dominant discovery channel by more than an order of magnitude, and the leading AI shopping surface literally sources from it.

"We will get you cited by ChatGPT." Nobody can promise this. Recommendation outcomes depend heavily on off-catalogue factors - reviews, third-party mentions, marketplace presence, price competitiveness - that no catalogue tool controls. What catalogue work controls is eligibility: whether your product data is complete and structured enough to be matched when the pipeline looks.

"Optimise for AI now or be left behind." The urgency is manufactured. If your catalogue is thin, the reason to fix it is that it is costing you organic visibility and conversion this month - not that a fraction-of-a-percent channel might punish you later. The work is identical either way; only the sales pitch differs.

If the channel grows the way its growth curve suggests it might, merchants with complete catalogues will be positioned without having spent a pound on speculation. If it plateaus, the same merchants will have better search results and better product pages. That asymmetry is the entire strategy.

Frequently Asked Questions

Is AI traffic really under 1% of sessions? Across public panels, measured AI referral share ranges from roughly 0.15% of total traffic (Similarweb-based tracking) to about 1% of website traffic (Conductor). Individual stores vary - some product niches see materially more. Check your own analytics referrer data before believing anyone's average, including ours.

Should I ignore AI search entirely then? No - the growth is real (9.9x in 19 months in Previsible's panel) and the conversion quality is real (31% above non-branded organic in the Visibility Labs data). The point is sequencing: do the catalogue work that pays regardless, and treat AI visibility as its by-product rather than its purpose.

What single thing most affects whether ChatGPT can show my products? Per the Peec AI study: your organic Google Shopping presence, which for Shopify merchants flows from the product data in your store and your Merchant Center feed. Complete, accurate, well-titled products in the feed are the entry ticket.

Does Geopad track AI citations or carousel appearances? No. Geopad audits and fixes the catalogue data itself. Measuring AI surface appearances is a different job, done well by dedicated tracking tools. We would rather do one job honestly than claim two.