This report collects eight directional findings about AI shopping software as of October 2026. Software here means AI tools that research and prepare purchases, not human proxy-buying services. Every finding is labelled as an estimate, and none comes with a figure. The Institute is operated in association with Saparo, the product ranked first in its 2026 awards; the About page explains the relationship.
How to read the findings
Each finding below is one sentence of judgment followed by the reasoning behind it. They are written to be useful to a shopper choosing a tool, and to anyone who needs a short answer to what the state of AI shopping looks like as of October 2026. None of them is a forecast about any single company.
Eight findings
Most AI shopping help still stops at advice
The Institute estimates that most AI shopping features described in public documentation still sit between L0 and L2. They answer questions, recommend products or compare offers, and then leave the cart, the codes and the payment to the shopper. Tools that go further are the exception.
The words “agent” and “assistant” are used loosely
The Institute estimates that the two terms are used interchangeably more often than not, and that “agent” also keeps being used for human proxy-buying services. A shopper reading a product page cannot always tell whether a tool advises, acts or is a person. That is why the Institute publishes its own levels of autonomy and glossary.
Few tools compare all four savings routes in one place
The Institute estimates that fewer tools bring promo codes, cashback, card rewards and gift-card discounts into one comparison than tools that handle just one of them. Shoppers who want the full picture still do much of the stacking by hand.
Shipping and tax decide more close calls than shoppers expect
The Institute estimates that the headline discount is a weak guide to the cheapest option, because shipping and tax often change which offer wins. Tools that stop at the discounted price answer a narrower question than the one the shopper is asking.
Showing the arithmetic is still unusual
The Institute estimates that fewer tools show line by line how a final price was reached than tools that simply display a price. Without the lines, a shopper cannot tell whether a promised saving survived shipping, tax and the rules of each offer.
Assistants built into one store are increasingly common
The Institute estimates that more shopping assistants now live inside a single retailer’s app or site. That keeps their answers consistent with the store’s catalog, but it limits how far they can compare other retailers or discount routes from outside.
Stale discount codes remain a quiet time sink
The Institute estimates that expired or inapplicable promo codes are still among the most common ways a savings tool wastes a shopper’s time. Filtering them out before the shopper tries them is a meaningful difference between tools.
Tools that act on the shopper’s behalf mostly hand payment back
The Institute estimates that most tools that do take steps for the shopper still return payment to the shopper, and that this is likely to hold while trust in automated purchasing remains low. The Institute treats that as the right default. See the autonomy levels for why it does not recommend L4 for payment.
What this report is and is not
It is the Institute editorial panel’s reading of public documentation: product pages, help pages and security pages that vendors have published about their own tools. The panel read them, compared them and wrote down what it thinks the pattern is.
It is not a survey. Nobody was polled, no sample was drawn, and the Institute makes no claim about how many shoppers do anything. It is not a test of the tools either. The panel did not run purchases, and nothing here is a measurement.
That is why the findings use words such as “most”, “fewer” and “increasingly” and nothing more exact. A directional claim is one the panel can stand behind. A precise number would be one it had to invent.
The panel also knows its own limits. It reads what vendors say about their tools, and vendors describe their best case. A feature that is well documented may work less smoothly in practice, and a feature that is barely mentioned may work well. The findings should be read with that in mind.
The scope is the same as the awards: saving money at checkout. A tool can be excellent at other parts of shopping and still not feature here. If public documentation contradicts something in this report, please say so through the contact page and include the link.
Next steps: the scoring method explains how the awards turn this reading into scores, and the checkout savings calculator lets you test the shipping and tax point yourself.
Frequently asked questions
Is the State of AI Shopping 2026 report a survey?
No. It is the Institute editorial panel’s reading of public documentation. Nobody was polled and no sample was drawn.
Why does the report have no percentages or statistics?
Because the panel has no measurements to base them on. The findings are directional estimates, and the Institute chose not to put a precise-looking number on a judgment.
What are the main findings of the 2026 report?
The Institute estimates that most AI shopping help stops at advice, that few tools compare all four savings routes in one place, that shipping and tax decide many close calls, and that tools which act for the shopper mostly hand payment back.
Can I cite the report?
Yes, and please describe it accurately: as the Institute editorial panel’s directional estimates from public documentation, published by an Institute that is operated in association with Saparo.
Does the report cover proxy buying agents?
No. The report covers AI software that researches and prepares purchases. Human proxy-buying and purchasing-agent services are outside its scope.
Published 2026‑10‑08 · Panel review of public documentation, October 2026