A rebuild quote is a guess until someone counts the archive.
Four counts that take an hour on any live site — language twins, canonical families, archive surface, template ratio — and how each one moves a rebuild quote.
A rebuild quote is a guess until someone counts what the archive will break. Four counts decide most of the variance, and all four run from outside the site with public data: language twins, canonical families, archive surface, and the ratio of pages to templates. An hour of counting tells you more about the final bill than another round of design revisions.
Every rebuild estimate I have watched go wrong failed in the same place. Not the homepage, not the component library, not the CMS selection — the URL inventory that nobody counted before quoting. The new site ships, it looks right, and six weeks later someone notices that a language tree stopped pairing, or that half the archive now shares one title, or that pages which used to rank serve a 301 to the wrong winner.
Those are not build defects. They are inventory defects, and they were priced at zero because they were never counted.
The design is the cheapest line in the quote
A rebuild is the only construction project where both sides start by discussing finishes. Design hours are easy to estimate because everyone has done it: page count times a rate, plus a component library, plus a CMS integration. Two agencies quoting the same site land in a similar range on design and build, then diverge wildly on what happens to the existing URLs.
A design estimate is a guess about work that has not happened yet, so both sides can only argue about it. A URL inventory measures work that already exists: it is either right or wrong, and it is the one line that cannot be negotiated.
Count 1 — language twins
Start with the obvious asymmetry: a bilingual site does not have two trees of equal size. It has one original tree, one translated tree, and a gap. Every page that exists in one language and not the other is a decision the rebuild has to make again.
The pairing is not cosmetic. Google's guidance on localized versions is explicit on two points: each language version must list itself as well as all other language versions, and if two pages do not both point to each other, the annotations are ignored. That second rule is the one that bites: a single missing return link does not degrade the pair, it deletes the annotation.
Now put that next to adoption. The Web Almanac's 2025 SEO chapter puts hreflang on roughly 20% of pages, concentrated on home pages, with Spanish appearing on about 3% of them. Inner pages get less consistent coverage. That is the tell: clusters get declared at the top of the site and abandoned further down, where most of the URLs live.
Three numbers:
- URLs in each language tree.
- URLs that actually pair with a twin in the other tree.
- URLs that have no twin at all.
The third number is the expensive one. A rebuild has to decide, per URL, whether it is a translation to preserve, a rewrite, or an orphan to retire. That is editorial and SEO work, not template work, and it is billed by the hour. If the new site generates the alternate-language block from a translation field, and half the archive has no translation, the build will either pair the wrong pages or emit nothing. Both outcomes are invisible in a design review and expensive in month two.
Count 2 — canonical families
Almost every site has a canonical tag, so "do you have canonicals" is a useless question. The question that changes a quote is: how many URLs claim to be the canonical version of the same content?
Google describes rel="canonical" as a strong signal and sitemap inclusion as a weak one, and warns against declaring different canonicals for the same page through different methods. The failure mode this creates is a site with two live URL patterns for one piece of content — a translated segment and a localized segment, or a category path and a flat path — where both return 200 and both point at themselves. Nothing is broken in isolation. Search engines simply get to choose, and they choose inconsistently.
A rebuild forces the choice. Someone has to pick the winner, write the redirect for the loser, and confirm the winner is the URL that actually accumulated whatever authority exists. Pick wrong and you keep the copy but 301 away the address, which is a self-inflicted migration wound that looks like an algorithm update from the inside.
The count is one number per content family: how many distinct URLs return 200 and self-canonicalize. Every answer above one is a redirect entry you must write, test and monitor.
Count 3 — archive surface
Paginated listings, tag pages, author pages, date archives, internal search results. Each is a URL with a template, usually generated by a CMS rule nobody has looked at since launch.
The cost is not storage. It is metadata. A new listing template will hard-code one title and one description unless someone parameterizes it, which means the rebuild can turn a family of pages into a family of duplicates — identical titles, identical descriptions, all self-canonical, all indexable. Google's crawl budget guidance points at the same mechanism from the other direction: perceived inventory, where too many duplicate URLs "wastes a lot of Google crawling time on your site," and the guide is written for properties at 10,000+ unique pages with fast-changing content.
The ratio is the finding:
- How many archive URLs exist.
- How many distinct titles and descriptions they carry.
Five hundred archive URLs carrying four titles is an afternoon of template work. Five hundred carrying one title means the archive was never designed.
Count 4 — the page-to-template ratio
This is the count that separates a clean rebuild from a forensic one, and it is the one nobody runs.
Pull the sitemap and group URLs by path shape. Shapes that contain one URL are hand-built pages: bespoke sections, embedded components, copy written for that page alone. Shapes that contain hundreds are generated from structured data.
The two behave in opposite ways. A generated family is cheap to carry over if the data model is explicit — the new template reads the same fields. A hand-built page is never cheap: there is no model to port, so it gets re-authored.
The dangerous case is the overlap: generated pages whose bodies were hand-edited after publication. Those edits exist only in the rendered HTML — the model does not know them, the new template will not reproduce them, and nobody notices until traffic is compared by page.
You can find them without database access. Take ten pages from the largest generated family, strip the boilerplate, and diff the body copy. If they are 90% identical, the family really is one page with variables and the rebuild is mechanical. If they diverge substantially, editors have been writing into published output, and that content has to be extracted before migration or it is gone.
The four counts, and where each one hits the quote
| Count | What it measures | What breaks in a rebuild | Where it lands in the quote |
|---|---|---|---|
| Language twins | URLs per language; share with a reciprocal twin | Alternate cluster dropped or mis-paired; translated pages read as duplicates of the original | Editorial hours plus risk to the weaker locale — usually the largest variance |
| Canonical families | 200, self-canonical URLs per content family | Redirect map points at the wrong winner; the URL that ranked is the one retired | One-time redirect map, plus QA after launch |
| Archive surface | Listing URLs vs. distinct titles and descriptions | One title cloned across the family; index bloat and duplicate metadata | Template parameterization — hours, not content |
| Page-to-template ratio | URLs per path shape; hand-edited bodies inside generated families | Bespoke pages rebuilt as templates; edited copy overwritten and unrecoverable | The highest-variance line in any rebuild; scope before signing |
Run it yourself
Four commands, no analytics access, no vendor tooling. Substitute your own host and adjust the path patterns to your URL shape.
Language twins. Count URLs per first path segment, then check how many pages declare an alternate at all:
curl -s https://example.com/sitemap.xml \
| grep -o '<loc>[^<]*</loc>' | sed 's/<[^>]*>//g' > urls.txt
awk -F/ '{print "/"$4"/"}' urls.txt | sort | uniq -c | sort -rn
while read -r u; do
echo "$u $(curl -s "$u" | grep -c 'hreflang=')"
done < urls.txt | awk '$2==0' | wc -l
If the first command returns a sitemap index rather than a URL list, fetch each child sitemap first.
Canonical families. Find every URL whose canonical points somewhere else, then look at what is left claiming to be canonical:
while read -r u; do c=$(curl -s "$u" | grep -o 'rel="canonical" href="[^"]*"' | sed 's/.*href="//;s/"//') [ -n "$c" ] && [ "$c" != "$u" ] && echo "$u -> $c" done < urls.txt | wc -l
Archive surface. Count paginated and listing URLs, then check how many titles that family carries:
grep -E '/(page|blog|category|tag)/[0-9]+/?$' urls.txt | wc -l for p in 1 2 3 4 5; do curl -s "https://example.com/blog/$p/" | grep -o '<title>[^<]*</title>' done | sort | uniq -c
Page-to-template ratio. Group by path shape and see which shapes are single pages:
awk -F/ '{print "/"$4"/"$5"/"}' urls.txt | sed -E 's#/[0-9]+/$#/#; s#/[a-z0-9-]{12,}/$#/<slug>/#' \
| sort | uniq -c | sort -rn | head -30
Bracketed shapes with a count near one are hand-built pages. Shapes with a count in the hundreds are generated families — and those are the ones to diff by eye.
What these counts cannot see
They measure surface, not quality. A site can pass all four counts and still have copy that does not convert, and the rebuild will faithfully preserve that. Internal link equity is also invisible to a sitemap read, redirect mapping still has to be written by hand for every retired URL, and none of this sees anything behind a login or a client-side route that never reaches the sitemap.
What they give you is a floor: a defensible number for work design review cannot find.
The count that changes the quote most
If I had to keep one, it would be the language-twin count — not because bilingual sites are the biggest category, but because asymmetry is the assumption nobody questions. The homepage is a mirror, so everyone assumes the tree is. It almost never is, and the gap is where a rebuild quietly reclassifies a translated page as a duplicate of its original.
The runner-up is the hand-edited-body check inside generated families. It is the only one of the four that protects work which exists nowhere else.
Both are cheap to run and both are unglamorous, which is the whole argument: they belong in the scoping call, before the design deck, not in the discovery phase after the deposit clears.
If the decision is still open between repairing, redesigning and rebuilding, the five checks that separate those three are the input above this one. Once the answer is rebuild, the migration lessons from a WordPress to Astro move cover what breaks in the move itself, and the file and deploy limits that decide a build are the constraint the archive count eventually runs into.
Count it, then price it. That is the scoping call we run before quoting a rebuild — and if you want a second opinion on an existing quote, start here.
Frequently Asked Questions
How long should a website rebuild take to quote?
The design and build hours are the fast part — a week of scoping is normal for a mid-sized site. The slow part is the URL inventory: counting language pairs, duplicate canonicals, archive pages and hand-edited content, then writing the redirect map. If a quote arrives without a redirect map line item, the inventory was skipped, and the cost will surface later as lost rankings or a change order.
What is a language twin in a website rebuild?
A language twin is the URL of the same page in another language. Twins are only twins if both URLs declare each other with hreflang; Google states that each language version must list itself as well as all other language versions, and that if two pages do not both point to each other the annotations are ignored. A rebuild that regenerates templates without carrying that pairing forward can silently unpair every bilingual URL on the site.
Should archive and pagination pages be migrated in a rebuild?
They should be decided, not defaulted. Paginated listings, tag pages, author pages and date archives each cost a template, a title, a description and crawl attention. Migrate the ones that earn links or serve navigation, consolidate the ones that only exist because a CMS generated them, and make sure the new template does not clone one title and description across the whole family.



