I've been on more of a content migration kick lately, and with Sitecore having a product in their world, it was time to crack it open.
(Just a note, I know it's SitecoreAI now, and I guess SitecoreAI CMS, but XM Cloud just rolls off the tongue...or keyboard...easier. So if you see XMC, it's all the same thing!)
Pathway was announced at Symposium in 2025, and it started as a "legacy" to XM Cloud concept at the time, all inside the Sitecore ecosystem. That's now expanded to being able to pull content from any site, which is handy, but there are some tradeoffs. So let me walk you through the process from the external side, as I figure that's what most people will be concerned with.
Step 1: Select CMS
This gathers the ingestion data. Pick if this is coming from Sitecore or any other website (we'll assume any other website here). Give the migration setup a name and description, then we get to the big choices.
- First, for the URLs, you can either put in a sitemap.xml file URL or individual URLs...and here's the catch, you can only go 200 URLs at a time. This seems like a potentially restrictive limit to doing some effective migrations, but I found that when I put in 195 URLs, some of the steps took a good while to run.
- Second, you can select whether you want to import images found on the scraped site. This is on by default and I wouldn't suggest changing it.
Step 2: Configure SitecoreAI Instance
Now we start getting into the meat of configuration. Here you'll select your environment and site, and then it asks for the site path. While I would think by default it'd be the Home item of whatever site you picked, I imagine you can pick a subsection if you so choose, which would make sense if you've got more than 200 URLs to go through.
Then it's going to ask for a a structure file, and I found the instructions a little confusing the way they were worded. Basically, you need to go to Content Editor and go to the site you're working with (whichever one you picked before). Right-click the site node, go to Scripts, then the Migration folder, and then Download Export Structure. The defaults it give you should be all you need, but if you need to add additional templates not in the expected location that XMC/SXA figures, you can add them here. Once you get the JSON file, upload it in this step.
Once this is all done, hit the button under Start Web Crawling and maybe get a drink, this could take a little bit.
Step 3: Content Audit
This section comes in two steps. First is the "grouping" step, which I think looks for commonalities in the URLs for purposes of the next step, which is the first of the two "big boys" in the process.
For template mapping, you'll find an AI Instructions button that will pop a modal with a bunch of language in there that you can't change. At the bottom, though, is the key - an additional section where you can give the system all the additional instructions you want. For example, if the source system has some metadata that says "this is a landing page" but you have a "landing page" template that wouldn't be the same, designate what default templates should be used. If you have a news article template, you can give instructions on when it should be used (via metadata, URL, etc). This is your chance to be very detailed!
After you run this mapping, each URL will be listed with the template information, and you can dig in to see what fields in your template will be filled in. At this point, you'll probably find things not to your liking, and then you'd go back, revise your instructions, and go again. This can become a very iterative process, but assuming you have many sites to eventually bring over, you can keep those instructions handy to reuse in the future. One thing I'll note, for example, is the mapper wants to set or unset every available field - this includes the Page Design field, which you might've set via template in standard values. Any presets you might have are likely to get erased, so adding an instruction on fields to maintain would be helpful.
Step 4: Map Content
The second half of the "big boys" and your last chance to influence things. Another AI Instructions button awaits you, and this time we're talking about components, so more detail would be useful here than even before. Make sure to guide the conversation toward what how you want the system to map what it sees to your components. You might have to dig into CSS classes in the source system, that you can tell the mapper that X class means it should use X component. If you're using a page design with breadcrumbs built in, tell the system not to include the breadcrumbs component you might have. Again, this will be an iterative process, but one you can codify for future use.
Step 5: Migration
This is it, the big one, the one we've all been waiting for. (Hat tip to Harry Potter and Oliver Wood.) It's pretty straightforward here though...hit that Migration button and sit back. You'll get a log of successes and failures, so you'll know if any URLs don't get converted. But now you wait, and then when it's all done, go hit Content Editor and Pages and see what you've wrought.
That's the good news. The bad news, this doesn't have a redo scenario, where if the final result of generated pages comes up and you're not liking it, you just go back and run it again. No iterative here, the process is done and you'll need to start all the way at the beginning. Make any notes from here to update your AI instructions, and go around again. So I'd start with a subset of the pages you're looking at, some representative samples, so you can drop the bad pages and try again if you want.
What's Next?
Hard to say, but it appears there's a new version of Pathway coming soon. The Sitecore Symposium agenda lists a session on Pathway 2.0-the agentic migration: from an application to a toolset:
Pathway 2.0 turns migration into a composable toolset with agent-callable capabilities, built-in logging, and step-level traceability. Learn how Sitecore supports Sitecore and non-Sitecore migration paths, and why observable, agentic tooling is a durable foundation for modernization.
My guess is there will be more insights into the migration process, to avoid the "bad news" that I mentioned above, and to allow for deeper refinement. This might also help folks do more with the Agentic Toolkit, as I'd wager there might be enhancements there to support this.
I will say that in the early days, Pathway had a pretty bad reputation, but it does seem to be better now in its external scraping. The 200 URL limit is likely the biggest hurdle to large-scale usage, but given the wait times at some steps, I understand the limitation for now. Hopefully Pathway 2.0 will help there as well.
Stay tuned!
Top comments (0)