A page that exists is not a page that is found. Between publishing a URL and seeing it earn a click sit three separate events — discovery, crawling and indexing — and a bilingual Miami site puts twice as much traffic through that pipe as anybody planned for.
Most site owners here discover this the same way. The developer confirms the page is live, the analytics show nothing, and somebody concludes the content is bad. Usually the content was never the problem. Google simply never got around to the URL, or got to it and decided not to keep it.
What follows covers crawl budget and what consumes it, sitemaps as an instrument of discovery rather than a formality, the submission mechanics in the Indexing Hub, and the arithmetic that tells you how long a site of your size will take. Running through all of it is the local complication: a fully bilingual site doubles every address before a single new page is written.
Existing, found, crawled, kept
These are four different states and a URL can sit in any of them for weeks. Existing means the server returns the page. Found means Google knows the address. Crawled means a bot fetched it. Kept means the page survived evaluation and can appear in results. Only the last one produces traffic.
The states fail in different ways too, which is why "the page isn't ranking" is a useless diagnosis. A page nobody linked to and no sitemap lists may never be found at all. A page that is found but sits eleven clicks deep may wait months. A page that is crawled and judged a near-duplicate of another gets dropped, and on a bilingual site that judgement lands more often than owners expect.
| State | What it means | Common cause of a stall |
|---|---|---|
| Exists | The server returns the page | Live but linked from nowhere |
| Found | Google knows the address | Absent from every sitemap |
| Crawled | A bot fetched it | Buried deep, or the crawler ran out of budget |
| Kept | It survived evaluation | Judged thin, or a duplicate of its counterpart |
| Ranked | It appears for a query | Competing with your own second version |
Crawl budget, and what quietly eats it
Crawl budget is the practical ceiling on how much of your site gets fetched in a given period. It is not a number you are told, and for a twenty-page restaurant site it never matters. It starts mattering the moment a site generates more addresses than it has pages — which is most sites, and every bilingual one.
- Parameter and filter URLs. A yacht brokerage with filters for length, year, builder and price can generate thousands of crawlable combinations from forty listings. Each one is an address the crawler may spend a visit on.
- Redirect chains. Every hop is a separate fetch. A domain migration that left three-step chains in place is paying triple for every old address that still gets requested.
- Soft errors. Pages that return a success code while showing "not found" or an empty result look like real content and get crawled repeatedly, because nothing tells the bot to stop.
- Slow responses. A server that takes two seconds per page is fetched less often than one that takes two hundred milliseconds. Speed is a crawl issue before it is a ranking issue.
- Near-duplicate pairs. Two addresses with substantially the same content each consume a fetch, and one of them will eventually be discarded. This is the bilingual failure mode in its purest form.
None of this is exotic. A construction contractor with a project gallery, a clinic with a procedure page per treatment and per doctor, or a freight forwarder with a landing page per trade lane will all produce more URLs than a human would guess. Add a second language and the count doubles again.
A bilingual site doubles every address on day one
This is the structural fact of doing business here. If a firm in Brickell or Doral runs an English site of 400 pages and does the job properly in Spanish, it does not have 400 pages plus a few. It has 800 addresses, and the two trees compete for the same crawl budget, the same internal link equity and, if the setup is wrong, the same query.
The doubling itself is not a problem. It is the intended shape of a bilingual site, and in this market the Spanish tree is frequently the more valuable half rather than an accessory. The problem is how the second tree usually arrives: late, partially, and built by whoever had capacity that quarter.
The pattern is consistent enough to describe in advance. The English site launches first because the agency that built it works in English. Two years later somebody notices that most of the customer base reads Spanish, and a translation project starts. It covers the home page, the main services and perhaps a dozen articles. Then it stops, and what remains is a half tree.
Four ways the Spanish tree goes wrong
Translated to a point, then stopped
Twenty pages of a hundred exist in Spanish. The navigation still lists all hundred, so the menu links to addresses that were never built.
- Broken or missing links
- Crawl budget on dead ends
Spanish URL, English content
The address exists under the Spanish path but serves the English text when no translation was written. The bot sees two addresses with identical content.
- Guaranteed duplicate signal
- The weaker copy may win
hreflang pointing at the wrong page
Every Spanish page names the English home page as its counterpart instead of its actual twin, so the pairing is meaningless and usually ignored.
- Non-reciprocal tags
- Missing self-reference
Both versions chasing one query
Spanish-language search in Miami often returns English pages. Without correct pairing your two versions alternate, and neither settles.
- Unstable positions
- Split link equity
The fourth of these is specific to bilingual markets and easy to miss, because the symptom looks like ordinary volatility. One week the English page shows for the Spanish query, the next week the Spanish page does, and the average position in Search Console looks merely unstable rather than broken.
| Symptom | What the crawler sees | What to do |
|---|---|---|
| Spanish page shows English text | Two addresses, one content | Translate it, or remove the address entirely |
| Menu links to untranslated pages | Dead ends spending budget | Hide the link until the page exists |
| All hreflang tags point at the root | An unusable pairing | Pair each page with its actual twin |
| Positions alternate by language | Two candidates for one query | Fix pairing, then check internal links |
| Spanish pages never crawled | No path in, no sitemap entry | Give the tree its own sitemap |
A bilingual tree that a crawler can read
The rules are not complicated; they are simply skipped when the second language arrives as an afterthought. Each is easy to verify on a sample of ten pages, which is the audit worth doing before any indexing work begins.
- One address per page per language. The English page under its own path, the Spanish page under the language directory, and no third address serving either of them.
- Reciprocal pairing, with a self-reference. Each version names itself and its counterpart. If the English page names the Spanish one, the Spanish one must name the English one back. One-way tags are discarded.
- A canonical that points at itself. A Spanish page whose canonical names the English version is asking to be dropped, which is exactly what happens.
- Internal links stay inside their language. A Spanish article should link to Spanish services and Spanish articles. Mixed linking sends the crawler back and forth and splits the signal.
- No automatic redirect by location. Sending every visitor from Colombia to the Spanish version by IP means the bot only ever sees one tree. Offer the switch; do not enforce it.
Half-translated pages need a decision rather than patience, and there are only three defensible answers. Finish the translation, if the page matters. Keep the address out of the index until it is finished, if the work is scheduled. Or delete the address and point it at the language parent, if nobody is going to write it. Leaving it live in English under a Spanish path is the one option that actively costs you.
Sitemaps do one job, and do it well
A sitemap does not make Google index anything. It makes addresses discoverable, which is the first of the four states and the one most often missing on a bilingual site where the second tree has few inbound links. For that job it is the most efficient instrument available.
Structure matters more than most people assume. A single flat file listing every URL works, but it tells you nothing when something goes wrong. Split by language and then by section, with an index at the top, and a submission report becomes readable: you can see that the Spanish services sitemap has a problem while the English one is clean. That is precisely the shape the Indexing Hub is built to parse.
One index for the domain
A single entry point naming the two language indexes and nothing else. This is the address you submit.
An index per language
English and Spanish each get their own, so the two trees can be compared line by line instead of being read as one pile.
One file per section
Services, industries, projects, articles. A fault now has a location: the Spanish services file rather than "the sitemap".
A report you can act on
Counts arrive per file, so a missing tree or a broken section is visible in the submission rather than three weeks later.
Submission by upload or by URL
For sites whose address count grew past what anyone wants to paste into a form.
- Two ways in. Upload the file directly, or hand over the URL of a sitemap already published on the domain.
- Recursive parsing three levels deep. An index pointing at indexes pointing at sitemaps is followed all the way down, which is what a language-then-section structure produces.
- Up to 1,000 sitemaps per job. Enough for a large bilingual portfolio without splitting the work into several submissions.
Two jobs run concurrently and up to twenty wait in the queue, which is a design worth noticing rather than a limit to fight. It means an agency handling several Miami clients can submit everything at once and let the queue meter the work, instead of tracking which submission ran on which afternoon.
Bulk submission, IndexNow, and reading the result
The URL tracker works to a daily budget of 1,000 addresses per account, and accepts up to 10,000 in a single batch. Those two numbers are not in conflict; the batch is what you hand over, the daily figure is the rate at which it goes out. Submission runs through the IndexNow API, which notifies GoogleBot and BingBot.
Bulk submission with a per-URL log
For anyone who needs to answer "was this address ever actually visited?"
- A record per address. Bot visit with a timestamp, status, and error detail where there is one — not a single number for the whole batch.
- Live counters. Submitted, found and failed update as the job runs, so a systematic fault shows up early instead of at the end.
- Both bots covered. IndexNow notifies GoogleBot and BingBot, which matters more in a market where a share of buyers arrive from outside the country.
Reading a batch is a matter of proportions, not totals. A failed count near zero means the addresses are sound. A failed count concentrated in one part of the tree — every URL under one path, say — is a structural fault and not bad luck. And a high found count with no movement in Search Console two weeks later means the pages were reached and judged, which is a content question rather than a technical one.
This distinction is worth stating plainly because a lot of money gets spent on the wrong side of it. If pages are being crawled and dropped, more submissions change nothing; the fix is in the page. The campaign and analytics side of the panel is where that work happens, and the Hub's job ends at the door.
Questions that come up on bilingual sites
Should the English and Spanish trees have separate sitemaps?
Yes, in nearly every case. A sitemap per language, split further by section and referenced from one index, gives you a readable report rather than a single number. Recursive parsing follows the structure three levels deep, so the nesting costs nothing at submission time.
We have Spanish URLs showing English text. Delete them or leave them?
Neither indefinitely. Decide per page: translate it if it matters, keep it out of the index while translation is scheduled, or remove the address and point it at the language parent. Leaving a duplicate live is the only choice with a guaranteed cost.
How long before a submitted URL shows up?
The bot visit is often logged within days; appearing in results is a separate matter with no guaranteed timeline. Treat the log as evidence the address was reached, and Search Console as evidence it was kept. They answer different questions.
Does resubmitting a URL every week help?
No, and it spends a daily budget that a large bilingual site needs elsewhere. Submit after a real change — new page, substantially rewritten content, a fixed pairing. Repeated submission of an unchanged address adds nothing.
Should we redirect Colombian visitors straight to the Spanish version?
Offer the switch, do not enforce it. A location-based redirect means the crawler only ever sees one tree, and cross-border research is exactly the traffic a Miami exporter, clinic or law firm cannot afford to route blindly.
A closing calculation
Take a mid-sized Miami business: 600 pages in English, the same 600 properly built in Spanish. That is 1,200 addresses. Submitted as one batch it fits comfortably inside the 10,000-URL ceiling, and against a daily budget of 1,000 addresses the run takes a little over a day. A full portfolio of ten such clients is 12,000 addresses and around twelve days — which is a plan, not a wait.
Now put the same 1,200 addresses through the failure modes above. If a third of the Spanish tree serves English text, 200 of those URLs are near-duplicates that will consume crawl budget and then be discarded. Fixing the structure first turns 1,200 submitted addresses into 1,200 candidates, instead of 1,000 candidates and 200 receipts for wasted work.
The rest is routine. Site tags filter a portfolio down to one property, individual sites can be shared with another email address when a client's developer needs access to one project, and background workers keep the counters current without a manual refresh. Exports run to 10,000 rows in CSV or JSON and 250 in PDF, which is enough to hand a developer a list of every failed address. More walkthroughs sit on our blog, and the audit itself is described under the services we run.
Where automation earns its place is the repetition. AutoSEO at $149 per month per domain keeps keyword work and link building moving while the structural fixes land, and My SEO Stream takes URL lists in batches, which suits a bilingual site producing two lists for every task. First measurable movement typically arrives four to eight weeks after the work starts, and on a site with an unindexed second language that window begins when the pages become findable, not when they were written.
The bottleneck is quiet because nothing announces it. No error appears, no alert fires; a page simply sits there existing. To see which of your addresses are in which state, connect the domain and run one sitemap job through the Indexing Hub. On a bilingual Miami site the first submission usually answers a question nobody had thought to ask: how much of the Spanish tree Google has ever actually seen.