SEO & indexing
Google had indexed 2 of our 25 pages. Here's what was wrong, and what got it to 16
By the Raw Code team · · 5 min read
Short answer
Seven weeks after launch, Google had indexed 2 of our 25 URLs and had never heard of 17 of them. Our tags were fine. Google simply wasn't asking for the pages: a new domain with no inbound links and thin, templated case studies. After two weeks of content and internal-linking changes, 16 of 30 URLs were indexed, and by November 18, 25 of 34.
What Search Console said on October 12
We run every URL in our sitemap through Google's URL Inspection API from a GitHub Actions job, after each deploy and on demand. On October 12, 2026, seven weeks after launch, the run came back like this:
| Verdict | URLs |
|---|---|
| Indexed | 2 (the home page and one service page) |
| URL is unknown to Google | 17 |
| Discovered – currently not indexed | 6 |
| Crawled – currently not indexed | 0 |
That last row matters most. "Crawled – currently not indexed" means Google read a page and decided it wasn't worth keeping, which is a quality problem. We had none. Google hadn't rejected our pages; it hadn't fetched them. "Unknown" means it had never queued the URL at all, even though every one was in a submitted sitemap.
We checked the tags first, and they weren't the problem
The usual advice for unindexed pages is to check the technical basics, so we did: every page returned 200, had a self-referencing canonical, carried no noindex tag or X-Robots-Tag header, was allowed by robots.txt, and was listed in a sitemap Search Console had read without errors.
Once those pass and most URLs sit in "unknown" or "discovered", editing tags further does nothing. The question becomes why Google isn't choosing to crawl.
The server logs showed one inner page fetched in 30 days
The site runs on Google Cloud Run, so every request is in Cloud Logging. Filtering on a Googlebot user-agent was useless at first: most of those hits were vulnerability scanners using Googlebot's name to probe paths like /.env. Anyone can send that user-agent string.
Filtering to Google's real crawler IP range (66.249.*) told a clear story. Over 30 days, Googlebot kept coming back for robots.txt, the home page, and sitemap.xml, and fetched one inner HTML page in total. It knew the sitemap was there and wasn't acting on it.
That pattern is the signature of low crawl demand. A brand-new domain with no other sites linking to it gets a small crawl budget, and Google spends it on the few URLs it has a reason to care about.
Why demand was low
- A new domain. rawcode.ai had been live for seven weeks.
- No inbound links. None of the four client sites we build and maintain linked back to us, so Google had no outside reason to think the pages mattered.
- Thin, templated pages. Most of our case studies were about 130 words and shared the same boilerplate line. A dozen near-identical pages make a whole domain look low-value.
What we changed
We worked down the list in order of what we could do ourselves that day:
- A body link from the home page, the one page Google was reliably crawling, to a new Toronto hub page that links on to the rest of the site.
- Facts that only belong on one page. Each client case study now lists what's on the live site, with page counts read from the client's own public sitemap and the date we checked. Those can't be copied from page to page.
- A "The problem" section on each ML case study, written from the project record, in place of the shared boilerplate.
- Rotating "more work" links, so each case study links to its neighbours rather than every page pointing at the same first three.
- Two Toronto pages for the queries we actually want, software development and website development, each built around projects we can name.
- A hand-set lastmod date in the sitemap, bumped only when content really changes, instead of the build time.
Two weeks later: 16 of 30
The October 26 run, on a sitemap that had grown to 30 URLs, looked very different:
| Verdict | Oct 12 | Oct 26 |
|---|---|---|
| Indexed | 2 of 25 | 16 of 30 |
| URL is unknown to Google | 17 | 4 |
| Discovered – currently not indexed | 6 | 10 |
| Crawled – currently not indexed | 0 | 0 |
All five service pages were indexed, along with both new Toronto pages, the AI chatbot page, and six of the fourteen case studies: three of the four client sites that now list live-site facts, and three of the nine ML projects. The hub pages (/services, /work, /team, /contact) were still out. They're mostly lists of links to other pages, which fits.
We can't separate how much of this came from the changes and how much from two more weeks of the domain existing. Both happened at once, and there was no control group. What we can say is that "Crawled – currently not indexed" stayed at zero throughout: nothing Google fetched was judged too thin to keep.
Update, November 18: 25 of 34
We re-ran the inspection after the deploy that published this post. Counting only the 34 URLs that existed before that day:
| Verdict | Oct 12 | Oct 26 | Nov 18 |
|---|---|---|---|
| Indexed | 2 of 25 | 16 of 30 | 25 of 34 |
| URL is unknown to Google | 17 | 4 | 2 |
| Discovered – currently not indexed | 6 | 10 | 7 |
| Crawled – currently not indexed | 0 | 0 | 0 |
Three of the hub pages, /services, /work, and /team, are now indexed. /contact still isn't. Neither are six of the ML case studies or two of the newer Toronto pages. None of the six pages published that day, this post included, had been fetched yet.
The last row is still zero. Every page that's out is out because Google hasn't fetched it, not because it fetched the page and turned it down. So we've kept adding detail to the thinnest pages. Thin pages lower how much of the site Google bothers to crawl, even if they're never rejected one by one.
What we'd do first on any new site
- Run URL Inspection on every sitemap URL and count the verdicts before touching a single tag.
- Read the server logs for verified Googlebot (by IP or reverse DNS), not the user-agent string.
- Get a few links from sites Google already crawls. For a studio, that's a "Built by" link from client sites to the matching case study. We still haven't got ours, and it's next.
- Link from the home page body to the pages that matter most.
- Give every templated page at least one fact that's true only of that page.
This is the same checklist we now run on SEO/GEO website builds for clients. The client sites we maintain, such as Osoba Renos & Design, are structured around the same idea: every city or service page carries details specific to it.
Related
- Osoba Renos & Design · case study
- LumiRank · case study
- SEO/GEO-Optimized Website Builds
- Website Development in Toronto
More from the blog
Have a problem like this?
Tell us about it on a free call and we'll say what we'd try first.
Book free consultation