The rules of the game for e-commerce are currently being completely rewritten. Until
now, we’ve optimized online stores with visually appealing user interfaces,
straightforward purchase paths, and traditional search engine optimization (SEO), but going forward, we’ll
have a new target audience that demands new approaches: autonomous AI agents and
generative search engines.
In this article, we’ll take a closer look at what AI-powered discoverability (GEO) and
agentic commerce (Agentic Commerce) mean for your business and how to safely prepare
your online store for this new era—in good time, but without rushing unnecessarily.
Two Ongoing Changes
The first is discoverability. When a customer asks ChatGPT, Gemini, or Perplexity
for a recommendation, the model compiles its response from search results, product feeds, and the content it
has indexed. You can start working on improving discoverability right away; in fact, much of this work also
benefits traditional search engine visibility.
The second is the purchase transaction itself: the agent fills the shopping cart and pays for it on the customer’s behalf.
This is truly new; technical standards are still being developed, and it’s best to wait to make decisions
about this until we know a little more. The year 2027 is a realistic timeframe for
the first use cases of agent-based shopping to become part of Finnish e-commerce.
In short: get your discoverability in order now, because it will pay for itself either
way. It’s a good idea to start preparing for the advent of agent-based purchasing by making
a plan and taking the first practical steps.
First, check to see how your store is doing
Before you decide to launch a GEO project, it’s a good idea to assess your current situation. You can
do this in about ten minutes with three tests.
Ask the AI a purchase-related question for your category, just as your customers would phrase it. Do the same
with ChatGPT, Gemini, and Perplexity, and make a note of which ones are mentioned. If a competitor is on the list,
open their product page and see what they’re doing differently from you.
Work with your e-commerce partner to identify which AI bots are visiting your store.
The lists may include, for example, GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, and
ClaudeBot. It’s important to ensure that there are no signs, for example, that your robots.txt file,
firewall, or bot protection is blocking AI agents from accessing your site. This is a common finding
and almost always an accident. Few people have specifically decided to block AI agents; rather
, bot protection was likely enabled at some point on a broad scale.
Also ask your technical team to check what your product page looks like without the scripts running in
the browser—when requested directly from the server. Are the price, availability information, and
delivery time visible? If not, then even an agent won’t be able to access them. In many modern
online stores, the price doesn’t appear on the page until after JavaScript has run, at which point the AI
tells your customer a price it has deduced from somewhere else.
If the results of these three quick tests look good, you’re in
pretty good shape when it comes to buying through an agent and finding products!
GEO is not a new field of expertise
AI-powered discoverability is often marketed as a new area of expertise. In practice, it is
an extension of search engine optimization, and its technical foundation is familiar.
Product information must be formatted in a machine-readable format (structured data) so that the price, availability, and
product features can be read without interpretation. Product feeds must be up to date,
because an outdated feed displays an incorrect price right in front of the customer. The essential content of
the page must be displayed without scripts running in the browser (server-side
rendering). And search engine crawlers must be able to access the site.
Be sure to clearly publish your shipping and return policies on your own website. When a customer asks
the assistant whether a product can be returned and how quickly it will arrive, the chatbot quotes the terms directly.
Unclear terms and conditions lead to misinformation or result in the recommendation going to a competitor whose terms
and conditions were easier to find.
The Standard for Agent-Based Procurement: Follow and Be Prepared
Standards for agent-based purchasing are still in the development phase. ACP (Agentic Commerce
Protocol) is a protocol developed by OpenAI and Stripe, on which ChatGPT’s Instant Checkout operates.
UCP (Universal Commerce Protocol) is a similar open
protocol developed by Google, Shopify, and their partners. AP2 (Agent Payments Protocol), on the other hand, defines how an agent verifies
its authority to make purchases on behalf of a customer.
None of these are established standards yet, but it’s likely that they will
eventually converge in terms of implementation. Our prediction is that the UCP (Universal Commerce
Protocol) will become the most widely adopted standard.
As for agent-based purchasing protocols, one concrete step can already be taken: under
the UCP, the merchant publishes a machine-readable manifest at the address /.well-known/ucp. From there, the agent
can determine for itself which services, purchase functions, and payment methods the merchant supports. Publishing the manifest
does not impose any obligations, as it is merely a declaration of supported features.
On platforms like Shopify, protocol support is built into the platform. In that case, your job is to keep
the product information up to date and monitor the platform’s development. With your own platform, the implementation is up to you
and the technical team building the online store. This topic should already be on the agenda, at least
for your monthly meetings!
In the B2B sector, the transition is slower, but the benefits are significant
In B2B commerce, an independent purchasing agent faces three obstacles. Prices are often
customer-specific and not visible without logging in; orders may involve approval processes; and
purchasing involves more responsibility and less “consumer protection.” Therefore, the agent’s role will
be supportive in the coming years, and that is why authorization protocols established by standards are more interesting for
B2B than payment-related functions.
There’s still plenty to do, though. It’s important to address missing product data and ensure its consistency, both
from the perspective of AI and human buyers. PIM is now also an AI initiative, and its importance
is particularly evident in the B2B sector. We’re seeing growing demand for AI-assisted custom
PIM solutions, Akeneo implementations, and a few lighter SaaS-based
PIM products.
Once the product information is in order, the next two steps will pay for themselves without
the need for agents. Semantic search—which understands natural language—finds the product even when
the buyer doesn’t know the exact name. The online store’s built-in AI assistant compares specifications on
the buyer’s behalf and puts together a draft order. Once
the friction in the purchasing process has been eliminated, a human can easily place the final order.
The Road to 2027
It’s a good idea to focus on the basics in late 2026. Test, fix, and add the right development items
to the roadmap. Get your product information, availability details, and shipping terms in order.
The winning standards for agent-based purchasing will likely be established during 2027.
Make sure that the core of your e-commerce platform is API-based and capable of supporting agents as well
. Make sure that agents can access your store.
Let’s work together to make the Finnish e-commerce sector the most agent-friendly in the world!
We have compiled these steps into a single checklist, where you can see the level of criticality separately
for B2C and B2B commerce, as well as for each stage of the timeline.
Download the checklist here: https://www.crasman.fi/agenttivalmius
