Written by Jeremy Souffir Founder, JTS Tech Services
The short version: on 24 August Narvar published its 2026 Holiday Shopping Report, surveying 100 retail decision-makers and 1,348 US consumers, and the headline pair is 65% of consumers planning to use AI for at least one part of their holiday shopping against 8% of retailers describing themselves as very confident using AI to improve the shopping experience. Treat that as a readiness gap and you will reach for the wrong fix, because "confident using AI" invites you to go and buy an AI thing. The number we would actually act on is further down: only 14% of retailers expect greater use of AI shopping assistants to be the biggest behavioural change this season, and most say their attention is on shipping costs, discounts and operational execution. That is not a confidence problem. It is a planning problem, and it has a calendar attached, because a good part of what an assistant tells a shopper about you in November is settled by pages it read weeks earlier.
What did the report actually find?
The useful detail is not the 65% headline, it is the breakdown of what people say they will use AI for. Shoppers were not describing one behaviour, they were describing four, and they sit at very different distances from anything you control.
- 43% say they will use AI for gift discovery. This is the open-ended question — "something for my brother who cooks" — and it is the one where being absent from the answer costs you a sale you never see in analytics.
- 33% say they will use it for comparing products and summarising reviews before buying. This is the highest-intent use on the list and the one most directly decided by what is written on your own product and category pages.
- 29% say they will use it for budgeting and planning their holiday spend. Mostly outside your reach, and worth ignoring in your own planning.
- 15% say they will use it for managing what happens after checkout — tracking, returns, reminders. Small, but it lands on the part of your operation that almost nobody has written for a machine.
- 78% say they would use AI-powered tools if doing so made the experience feel more personalised, which is the softest number in the set and the one we would lean on least.
Put the 43% and the 33% together and roughly three-quarters of the stated AI usage happens before the shopper reaches you, in a conversation you are not in the room for. That is the part worth planning around. It is also, usefully, the part that depends on assets you already own rather than on anything you would have to buy.
Why is this a calendar problem rather than a confidence problem?
Because of how the answers get built. We went through this in detail a week ago and will not repeat it here: when an assistant answers a shopping question on the web path, it is very often not opening your page in that moment. It is working from a compressed snapshot of your page taken at index time, and re-reads are neither instant nor guaranteed. That mechanism is boring on any ordinary Tuesday. It stops being boring in the second week of November, because it means a page you rewrite on 5 November may or may not be the page an assistant is quoting on 20 November, and you have no control over which. Anything that depends on being crawled again is not a November lever. It is an August, September and October lever that pays out in November.
This is why we would resist reading the 8% as "retailers need to get more confident with AI". The retailers in that survey are not wrong that shipping costs and delivery accuracy matter — the same report has 49% of consumers saying reliable delivery dates influence their purchase decisions, and 51% of retailers naming delivery-date accuracy as their top conversion strategy, which is a rare case of both sides agreeing. The problem is sequencing. Discount structures and shipping promises can be set in late October and still work. Being present in an assistant's answer is not that kind of task, and it is currently sitting at the bottom of most Q4 lists precisely because it does not feel urgent yet.
Whose numbers these are
This is a single vendor's survey, run by a company that sells post-purchase and delivery software, and every figure here is stated intent rather than observed behaviour. People routinely over-report what they intend to do with new technology, so we would treat 65% as an upper bound and not as a forecast. Two things keep it worth acting on anyway. The first is that the sample is disclosed — 100 retail decision-makers and 1,348 US consumers — and the consumer and retailer questions were run against each other, which is what makes the gap meaningful rather than the level. The second is that the report contradicts itself in a way that is actually reassuring about its honesty: 76% of consumers said they would accept slower delivery in exchange for free shipping, and yet when asked what timeframe felt acceptable on a free-shipping order, nearly 30% said same-day (11%) or next-day (18%) should still be the standard. Both cannot be comfortably true. A report that publishes both is showing you its working, and the correct response to that pair is not to pick the number you like — it is to notice that shoppers answer differently depending on how you ask, which is exactly why you should weight the direction of these findings and not the decimal places.
What about the shoppers who arrive earlier than you expect?
There is one more pairing in the report that we think is underrated, and it compounds the timing problem rather than sitting beside it. 58% of shoppers said they plan to start their holiday shopping earlier than last year, while 77% of retailers said they have no plans to incentivise early shopping. Read those two together and the season you are planning for starts before the season you are staffed for. If a meaningful share of discovery moves into October, and discovery is the use case 43% of shoppers named, then the window in which your pages need to already be readable is earlier still. None of that requires you to believe the 65% figure at face value. It only requires the direction to be right.
The two conclusions that both get this wrong
The first wrong conclusion is "survey intent, ignore it". You can discount 65% heavily and the plan still changes, because the actions this points to are cheap, they are things you should do for human readers anyway, and they have no downside if the number turns out to be 40%. Dismissing a finding because it is imprecise is only rational when acting on it is expensive, and this one is not. The second wrong conclusion is louder and costs more: "we need an AI shopping assistant on the site before Q4". Look again at what shoppers said they would use AI for. Gift discovery and comparing products happen in someone else's interface, before they have ever heard of you, and a chat widget on your own site does not appear in that conversation. Buying a widget to answer a demand that is being expressed somewhere else is the most expensive way to respond to this report. Both readings share a mistake: they treat this as a question about AI features, when the finding is about where the buying decision is now being made and how early the inputs to it are fixed.
What would we actually do between now and October?
- Ask the assistants what they currently say about you. Take your ten highest-margin products and the three or four gift-shaped questions a customer would really ask, run them through ChatGPT, Gemini and Perplexity, and write down what comes back. Do it now rather than in November, because this is your baseline and it stops being measurable once the season starts moving.
- Fix the first two sentences of your best product and category pages. Not the meta description — the opening of the page body. Write it so that someone in a hurry would have their question answered by it: what this is, who it suits, what it costs, what makes it different. That sentence is doing more work than the rest of the page in the 33% comparison case.
- Make the gift-shaped pages exist at all. "Gifts under $75", "for someone who bakes", "for a new apartment" — the 43% discovery use case is answered from pages that address the question in the shopper's language. If your only navigation is by product category, you are not in that answer.
- Get your review content on the page as text. Review summarisation was named by a third of shoppers. Reviews that load into the page from a widget after the fact are far less certain to be read than reviews that are in the delivered page, and this is usually a configuration choice rather than a rebuild.
- Make your delivery promise legible and true. 49% of consumers said reliable delivery dates influence their decisions and the report also found 56% have delayed or avoided a purchase because refund timelines were unclear. Put the real cut-off dates and the real returns window in plain text on the page, not only in a policy PDF.
- Write the post-purchase pages for a machine as well as a person. The 15% who will use AI to chase tracking and returns are asking a question your help pages could answer directly, if those pages said the actual numbers rather than "contact us".
- Do not block the assistants at the edge. Worth re-checking every quarter and especially before a peak season, because a bot rule added for good reasons in June is the cheapest possible way to be absent in November.
The genuinely encouraging part
Almost everything on that list is writing, and writing is the one part of retail technology that has not got more expensive this year. There is no platform migration in it, no new subscription, no integration project and nothing that needs to go into a release train. A small team can do most of it in a fortnight, and every item improves the page for human shoppers too, which means the work is not a bet on a survey being right. That is a genuinely unusual position to be in. The 8% figure is being read across the industry as evidence that retailers are behind on AI, and the more useful reading is that the bar is currently very low: when only 14% of your competitors think this is the season's biggest behavioural shift, being one of the few whose pages actually answer the question is not a heroic effort. It is a fortnight in September that most people will spend arguing about discount depth instead.
Where we fit
The hard part of this is not knowing what to do — it is that nobody inside a growing business has the time to find out what six assistants currently say about eighty products, decide which of the gaps are worth closing before October, and then actually rewrite the pages while the rest of Q4 is happening. That is the work an AI Shopping Visibility engagement is for. Retaining JTS means somebody runs the baseline for you: what ChatGPT, Gemini, Perplexity and Google's AI surfaces say today about your best-selling and highest-margin products, where a competitor is being recommended in your place, and which of those answers is caused by something you can fix in your own pages versus something you cannot. We then do the fixing in priority order — page openings, the gift-shaped and comparison pages that do not exist yet, review text that is actually in the delivered page, delivery and returns facts stated plainly — and re-run the baseline afterwards so you can see what moved rather than take our word for it. We also check the unglamorous thing first, which is whether your own edge is quietly turning those crawlers away, because that single misconfiguration makes everything else on this page pointless. The deadline is not ours and it is not Narvar's. It is that the pages need to be readable before the season, not during it.
Sources
- CPA Practice Advisor — 65% of consumers say they will use AI for holiday shopping (coverage of the Narvar 2026 Holiday Shopping Report published 24 August 2026, with the sample size of 100 retail decision-makers and 1,348 US consumers, the 65% / 8% pair, the 43% / 33% / 29% / 15% usage breakdown and the 78% personalisation figure)
- Chain Drug Review — New Narvar data reveals retailers aren't ready for AI holiday shoppers (independent coverage confirming the same figures, plus the 14% behavioural-change number, the 76% and same-day/next-day delivery contradiction, the 49% and 51% delivery-accuracy pair, the 58% early-shopping and 77% no-incentive figures, the 56% refund-uncertainty figure, and the quoted remarks from Narvar CEO Anisa Kumar)
- Narvar — Press room (the publisher of the report, for the primary release and for checking the methodology yourself)
- The DCX Newsletter — Holiday shoppers aren't waiting for your roadmap (an independent read of the same report, useful mainly as a second interpretation of the gap rather than as a source of new figures)
- JTS Tech Services — What ChatGPT actually reads from your site (the mechanism behind the timing argument above: what is read, how much of it, and when it is frozen)
- JTS Tech Services — Roughly half of AI-referred sessions land straight on a product page (where this traffic actually arrives when it does reach you, and why the homepage is not the page to fix first)
- JTS Tech Services — You can do the structured data perfectly and still be invisible (the edge misconfiguration to rule out before spending a fortnight on page copy)
