Before you build a single template, score every page as migrate, rewrite, or kill using four cold signals: does it earn traffic or links, is it a duplicate of something better, can machines actually read it, and will its URL have a clean home on the new site. Pages that earn and read well migrate. Pages with good intent but tired bodies get rewritten. Pages that do nothing but take up space get killed. You do all of this in a spreadsheet, on paper, with your shoes still on — not in a sandbox at 11pm.
Here's the thing nobody tells you until it's too late. You can rebuild a page beautifully in the new CMS and still be wrong, because the page never deserved to exist in the first place.
The building is already standing. Don't start swinging.
Imagine you've been handed the keys to an old office building and told to renovate it. The temptation is enormous: grab the wrecking ball, knock through a wall, see what happens. Feels productive. Feels like progress.
It is not progress. It is how you discover, three rooms in, that the wall you flattened was holding up the floor above.
A migration is the same building. Every page is a room. Some rooms have people working in them (traffic). Some have load-bearing pipes running through (inbound links and redirects pointing at them). Some are identical empty cubicles duplicated eleven times because someone kept copying the floor plan. And a few are that storage closet full of 2016 holiday banners nobody has opened since.
You don't find out which is which by demolishing. You find out by walking the whole place first with a clipboard, writing a verdict on the door of every room. Boring? Yes. It is also the single highest-leverage hour in the entire project, because the clipboard pass decides how much you have to build at all.
The four questions you ask at every door
You're not grading on vibes. You're grading on signals that are either true or false, and most of them are already sitting in tools you own.
First: does it earn anything? Pull sessions, entrances, and inbound links from the last twelve months (twelve, so you catch seasonal pages that look dead in July). A page with real organic entrances or external links is load-bearing. Touch it carefully.
Second: is it a duplicate? Not just literal copies, but near-duplicates where three pages all answer the same question slightly worse than one good page could. The near-dupes are the quiet killers. They split your authority and confuse both Google and the AI answer engines about which room to send people to.
Third: can a machine read it? Is the content real HTML text, or is it trapped in a PDF, an image, a script-rendered widget, or a tab that only loads on click? If an answer engine can't parse it, migrating it faithfully just moves the problem to a nicer address.
Fourth: does its URL have a clean home? If the new information architecture has an obvious slot for this page, good. If it doesn't, you have a redirect decision to make now, not on cutover night when everyone's tired and the coffee's cold.
Turning four answers into one verdict
Four signals, three piles. Here's roughly how they combine.
| Verdict | What the signals say | What it costs you |
|---|---|---|
| Migrate | Earns traffic or links, reads cleanly, has a home. The body still holds up. | A redirect and a careful content transfer. Cheap. |
| Rewrite | Good intent and some equity, but duplicated, thin, unreadable by machines, or just old. The topic deserves a page; this one isn't it. | Writing time. The expensive pile, so size it honestly before you promise a date. |
| Kill | No traffic, no links, no unique purpose, no future slot. A cubicle nobody sits in. | A 410, or a 301 to the nearest relevant parent. Deleting is the win. |
One rule that will save you an argument later: a page is allowed to be in exactly one pile. If someone insists a page is both "important" and "we should probably rewrite it someday," that's a rewrite with a deadline, not a migrate. Vagueness is how dead pages sneak onto the truck.
Why on paper, and why now
Because the sandbox lies to you. In a sandbox, a rebuilt page looks finished, which feels like an answer. But it can't tell you whether the page should have been rebuilt at all. The clipboard can. A spreadsheet with one row per URL and a verdict column will cull twenty, thirty, sometimes fifty percent of a site before a developer touches a template, and every killed row is a template you never have to build, QA, or redirect.
This is the part worth automating, because doing it by hand on ten thousand URLs is how good intentions die. A pre-migration crawl can pull the traffic, the links, the duplicate clusters, the unreadable assets, and the orphaned files into one scored view, so the humans spend their judgment on the genuinely close calls instead of copy-pasting numbers. That scored inventory is exactly what the pre-migration triage exists to produce. The verdicts are still yours. The tedium isn't.
So walk the whole building first. Write on every door. Then, and only then, decide where the wall comes down.
The wrecking ball is a terrible planning tool. It only knows one verdict, and it writes it on everything.