TL;DR

Probably not, but there are things you can put in place quickly that certainly won’t do any harm – one of which is a “just in case”, while the other will (we would estimate) be considered best practice in the next couple of years.

What is agentic commerce?

The ‘ideal’ of agentic commerce is that LLMs or other AI search surfaces are going to make purchases on their user’s behalf. What it really is? That’s going to depend on who you talk to – but as we were initially taken in by AMP (accelerated mobile pages) despite the sage advice of some seasoned SEOs at the time, we have to say that we’re a lot more cynical these days.

The way this is supposed to work – and where it will likely progress (but we’re talking at several years or more, as an estimate) – is that your personal chatbot will begin to establish a picture of your wants, needs and preferences which will, over time, allow it to make suggestions for products and services that are tailored to your personality and profile. The way we’ve described this in the past is that they will serve as a gentleman’s gentleman (a butler) or personal shopper, and would be permitted to bring you options (minimum) or purchase on your behalf (maximum).

When is agentic commerce likely to be a thing?

There have been multiple movements in this general direction over the last year – the Google and Shopify collaboration on Universal Commerce Protocol (UCP) probably being the most advanced, which includes:

• Catalog discovery: allowing machines to query real-time product specs, pricing, and inventory.
• Cart and session management: allows agents to build and manage multi-item customer shopping carts.
• Identity linking: maps user loyalty rewards, regional context, and member pricing into agent led transactions.
• Flexible payments: interoperates with checkout tools like Google Pay and Shop Pay, but also keeps merchants as the record holder.

We’re extremely early in the development of agentic commerce, however, and the main reason for this is probably because LLM businesses have primarily focused on hyperscaling and stock market valuation inflation rather than on trying to create actual, useful products, but either way, we’re probably going to have to wait for the ‘AI’ bubble to pop and the technology to face a future of commodification before we have locally hosted personal assistants. 

We can see the trouble as trial after trial of AI led commerce are abruptly ended quickly after starting (c.f. McDonald’s AI Drive-throughs, Walmart cancelling their OpenAI experiment, which led to OpenAI scaling back its own commerce ambitions, and eBay banning AI agents). 

That doesn’t mean that this is something brands can ignore – it just means that we’re much further out from the advent of agentic commerce than various tech CEOs would have you believe.

That’s why we mentioned several years at least and probably more before this becomes even a small concern for most ecommerce brands. However, as a group of increasingly old people – we also know that these years can pass in the blink of an eye, and also that a lot can change.

Should anything be done to prepare for agentic commerce?

This is a tricky one – the answer here is pretty much ‘no’, but there are also things that it wouldn’t hurt to do and which may help with general SEO performance even if the actual advent of agentic commerce is some way down the road. They are as follows:

• LLMs.txt: – we say this with the following proviso: only if it is very easy to do or automated. We very much doubt there’s much benefit to it, and the research we’ve read suggests that there’s very little correlation with inclusion in LLM generated answers, but it won’t hurt to have it on your site (Shopify briefly auto-generated it, but then it seemed to be replaced by a sitemap_agentic_discovery.xml which likely does the same thing. The key thing to remember is that we wouldn’t predict any improvement in search, this is basically a just in case exercise.

• EntityMap: – is an open source project and, from the site, is an: “entity-first index of website knowledge for AI systems, retrieval pipelines, and language-model-based applications”. Again, this is a fairly low effort endeavour, but is far more likely to be useful to your visibility in LLMs. Unlike LLMs.txt, EntityMap provides structured context about your business, products and relationships – this makes it easier for AI systems to understand what your site is about rather than simply crawling it. It’s also relatively simple to generate and can be implemented quickly..

Final thoughts

As things stand, there’s a lot of happening yet to happen in the world of AI. Today, for example (as I write, but probably today as you read, also, whenever that is), we read that Sam Altman has declared that the AI singularity has arrived. This is supposed to be an inflection point that leads to the exponential and rapid growth of machine intelligence all the way to Super Intelligence. He is almost certainly talking nonsense, but we’ll see. 

The chances are that LLMs in their current form are completely unsustainable and that smaller models will be required that will power locally hosted tools while tech giants like Microsoft and Google continue to deliver LLMs at scale via search, but without any real commerce layer (think the purchase power of the Amazon Alexa – it’s perfectly capable, but we’d be very surprised if you’ve ordered from it). 

As such, SEO and what we’ve classified as a ‘search everywhere’ approach will remain key – so don’t worry if you haven’t the time or the inclination to begin a markdown revolution just yet. Keep doing the right things well and, for now at least, that will be enough. If your site and product catalogue is well structured, your schema is comprehensive, your product data is accurate, and your site is easy for machines to understand, you’re already doing most of what future agentic shopping is likely to require.