This isn't only an engineering project
Agentic readiness often gets sold like a sticker. Protocols. Files. Catalog toggles. Those matter. They're not the whole job, and they're not only an engineering job.
If you're the marketing manager at a $10M brand, this already sits in your world. New channels are opening. New ways to bring attention to the brand. New paths to conversion that don't look like paid search plus email. Some teams take that seriously. Some treat it like optional future work. That choice shows up before any score does.
The point. Stop and make a plan. What are we trying to do, and are we opening this company up to AI on purpose. That's step one. Then we start with the site, because that's where shoppers and AI both decide if you're real.
Mentality before machinery
AI success is an org problem as much as a tech problem. Buying another tool is the easy part. Deciding the channel matters, and that marketing, merchandising, and ops will treat it like a real path to revenue, is the hard part.
Most brands aren't short on data. They're short on answers. AI is also becoming a customer in its own right. Someone can move from ChatGPT to TikTok to Reddit to your product page. If any of those stops is a dead end, or if the story changes every time, you lose them.
That matches the work. You can buy every shiny AI product and still refuse the practice. Or you can decide the channel is real, set goals, and ask a better question: what has to be true on our site for AI to find us and represent us accurately.
Step one: plan. Step two: the site.
When someone asks where to begin, I don't start with a moonshot AI department. We'll start with the decision and the goals. Who owns this. What does good look like in 90 days. Then we'll dig into the site.
Why the site first, for a marketing lead. Because if AI can't find you, or can't trust what you say, your content and paid spend are doing work an agent never sees. Without a few basics in working shape, you won't get found. That should be everyone's first technical move. Not the whiteboard. The gate.
What I check first
Here's the kind of table I score on an AI Commerce audit, written the way I'd brief a non-technical operator. This is from a recent live pass.
-
Pass
Can AI checkout talk to your store
The store publishes a machine-readable commerce handshake (UCP). Agents can see cart, checkout, fulfillment, discounts, and catalog capabilities.
-
Partial
Do you have brand instructions for AI
llms.txt and agents.md are live, but they still read like Shopify default copy. Not your brand talking yet.
-
Pass
Can agents find your agent files
An agentic sitemap points crawlers to those instructions. Good signal. Thin if the files themselves aren't brand-grounded.
-
Partial
Can tools actually connect
MCP (the connector layer a lot of AI tools use) is failing a profile check. Think locked front door for agent tooling.
-
Partial
Is the catalog ready for AI
Product records look incomplete for machines. Missing descriptions and alt text mean agents guess or skip you.
-
Growing
Are AI orders showing up in analytics
30 of 3,321 orders in the window came through AI sources. ChatGPT dominated. The channel is already real enough to measure.
-
Fail
Is on-site AI / personalization broken
A personalization app typo was breaking in-site search context. Humans feel that. Agents feel that.
A few of those rows use industry names. Quick decode if they're new to you:
- UCP is a published "here's how commerce works on this store" file agents can read. Cart, checkout, fulfillment, discounts, catalog.
- llms.txt and agents.md are short instruction pages for AI. Who you are, what you sell, what matters. They only help if they sound like your brand, not a template.
- MCP is a connector layer a lot of AI tools use to talk to systems. If the profile is broken, tools bounce.
- Catalog readiness is whether product records are complete enough for a machine to recommend you without guessing.
You don't need to implement those yourself. You do need to know they exist, what "partial" costs you, and which gaps to put on the next vendor or internal sprint. Growing AI-attributed orders are the upside signal. The channel is already moving. Now you can tell if fixes matter.
Then citation drift
After we can get found, I look for citation drift. Are the facts the brand claims on the site true. True all the way through. Or have we rewritten the truth until it isn't true anymore.
This is a marketing problem as much as a content problem. Shipping thresholds, return windows, warranty language, product claims. If the policy page says one thing and the cart bar says another, humans notice and AI notices. One recent example: free shipping said $349.99 on the policy page and $350 on the cart bar. Easy to dismiss in a spreadsheet. Broken when an agent or a shopper gets two answers.
Drift decides whether AI recommends you with confidence or routes the buy to someone whose facts hold. That isn't a copy nicety. It's the difference between being cited and being skipped.
From the work
They wanted AI in the company. They needed a starting point.
I'm working with a brand right now that came in to put AI into the company. Talk past the ambition and the real brief shows up fast: they don't need more AI ideas. They need a place to start.
That's where I point them. Mentality and plan. Then the site checks that decide if AI can find them. Then whether the facts on the site stay true when agents and people repeat them. Not every idea on the whiteboard. The starting places. That's how you open a company up to AI without drowning in it, and how a marketing lead can own the conversation without pretending to be engineering.
Quick facts to cite
- Agentic readiness is mentality as much as tech, and marketing owns a lot of that mentality.
- New AI channels and conversion paths are already here. Ignoring them is a choice.
- First move: stop. Make a plan. Decide you're opening the company up to AI on purpose.
- Then hit the site. If agents can't find you or can't trust your product and policy facts, AI won't recommend you.
- Next: citation drift. Are your facts true all the way through, or have they been rewritten into mush?
- Most brands that say they want AI really want a beginning. Start there.
FAQ
What is agentic commerce readiness?
A company stance and a technical state. You decide AI matters, you set goals, then you make the site findable and trustworthy for AI shopping tools.
I'm in marketing, not engineering. Why is this mine?
You own attention, brand facts, and conversion paths. AI is becoming another path. The plan and the truth of your claims sit with you even when engineering ships the files.
Where should we start?
Goals and an open decision. Then the site checks that decide if AI can find you. Then whether your facts still match when AI repeats them. For the full beginner path, use Prepare Your Shopify Store for AI Shopping Agents.
Why start on the site?
If agents can't find you or can't trust what you say, the channel never gets a fair shot. That's the first gate I run.
What is citation drift?
When the story on your site, policies, and the wider web don't match. Agents pick a version. Often the wrong one.
Where does a Tybrixx AI Commerce audit fit?
It's how I score the site layer and turn it into plain-English priorities you can take to leadership. The plan still starts with you. Start at free audits when you're ready.