“We Fixed The Whole Site And New Customers Still Won’t Budge”
You swapped the product photos. You rebuilt the abandoned-cart emails. You shaved another half-second off the load time.
And the new-customer number sat exactly where it sat last year.
I know that feeling, and my instinct is always the same: there must be something else inside the site left to fix. Conversion rate, checkout friction, one more test.
But when you keep fixing things inside the site and the number won’t move, it’s worth asking whether the problem happens before anyone gets there.
Somebody went and asked shoppers directly about that “before.” So let’s look.
How Many Shoppers Now Start A Purchase With An AI Tool
41.4% of shoppers said an AI tool is where they start, against 38% who go straight to a brand’s own website (the survey, 4,040 US and UK adults). That’s the first time AI has come out ahead of going direct in this question.
The survey comes from Bloomreach, which sells commerce personalization software, and was run by Propeller Insights. Fieldwork ran May 11-21, 2026, and the results landed on July 22, 2026 in a report called “Is AI Usurping Your Brand?”.
Using the tools this way isn’t a fringe habit anymore, either.
75.4% of respondents said they’ve used ChatGPT or something like it to help with shopping or a purchase decision. 80.4% said the experience met or beat what they expected, and 55.6% plan to use it more over the next twelve months.
The Thing That Moved Was The Direct-To-Site Number
Here’s the part that’s easy to read wrong. In the same survey the year before, AI was at 41.1% and going direct to the brand’s site was at 58.9%.
So the AI side went from 41.1% to 41.4%. That’s 0.3 points in a year — you’d have a hard time arguing it’s more than noise.
The side that moved was the other one. Direct-to-site fell from 58.9% to 38%, roughly 21 points.
Which means “the AI boom pulled everyone over” isn’t what these numbers say. What they say is that the habit of opening the brand’s own site first got weaker. The ranking flipped because the brand sites slowed down, not because AI sped up — it’s the marathon thing, where you move up a place and then find out you ran the same pace and the runner ahead of you cracked.
That distinction decides where you spend. Before chasing growth in the AI channel, the sharper question is where the people who used to type your URL are picking their first candidates now.
What People Actually Use AI For While Shopping
The biggest use is comparing features and prices, at 48%. Hunting for sales and deals comes next at 46%, then looking for ideas at 41%.
Comparison being the top job means something pretty concrete for a brand. Comparing is just lining up a few candidates and narrowing down, so the model isn’t playing conversation partner here; it’s assembling the shortlist itself.
And a shortlist is not a place a brand can talk its way onto after the fact.
The behavioral numbers point the same way. 60.8% said they feel more confident in their purchase decisions, 39.5% said they shop more often, and 37.5% said they spend more overall.
Put 80.4% “met or beat expectations” next to 60.8% “more confident” and the novelty reading gets hard to defend.
How Much Weight These Numbers Can Hold
Worth being straight about this. Bloomreach sells personalization and AI search for commerce. “AI commerce is going mainstream” is a conclusion the company profits from, and that’s a real commercial interest sitting behind the questionnaire.
It’s also self-reported, collected from an online panel, and the respondents are US and UK consumers. Nobody measured actual behavior, and nobody measured your market unless your market is those two.
So treat levels like 41.4% and 75.4% as the optimistic end of the range.
But the difference between two years of the same questionnaire is sturdier than either level. Whatever bias the sponsor introduces, it presses on both years the same way.
And that difference is the one above: AI flat, direct-to-site down about 21 points. That part survives the discount.
Measure Whether You’re Still On The Shortlist
None of this makes site work a waste. Everything you built to keep the people who arrive is still doing its job. What changed sits one step earlier: comparison is moving into the model’s answer, and if you’re not in that answer, how good your site is never gets evaluated at all.
The awkward bit is that “are we still a candidate” isn’t a number your analytics can show you. Site analytics only starts counting after somebody already picked you. AI’s fingerprints also tend to get filed under branded search and direct, which I dug into in the study where a ChatGPT mention drove 2.5x the visits. So I’d narrow it to two moves:
- Write down about ten comparison questions a real buyer would type about your category
- Ask them monthly, under the same conditions, and count what share of answers name you
The point isn’t asking once when it occurs to you. It’s asking the same questions on the same schedule, because a single “we showed up!” tells you nothing about a trend.
And if what’s bugging you is whether you can even receive the people AI sends, the piece on product pages AI can’t read is closer to that.
Maybe I’m wrong about how fast this lands outside the US and UK. But writing down your ten comparison questions costs an afternoon, and not knowing costs a year.
Sources
- Bloomreach, “From Novelty to Necessity: New Bloomreach Consumer Survey Finds Shoppers Now Prefer AI Tools to Shop”, July 22, 2026, bloomreach.com (the report “Is AI Usurping Your Brand?” is at bloomreach.com/en/agentic-commerce-report. Method: online panel survey of 4,040 US and UK adults aged 18+, fielded May 11-21, 2026 by Propeller Insights on behalf of Bloomreach. Caveats: Bloomreach sells commerce personalization software and AI search, so it has a commercial interest in the conclusion that AI commerce is going mainstream. Answers are self-reported, and respondents are US and UK consumers, not a measurement of any other market.)
