Keyword documents do not fix a site that serves two titles, two identities, or a homepage that 404s its own proof. That is architecture.
I have done enough technical growth work to know where engineers actually help: crawl, canonicals, redirects, and one intent per URL. Keywords are a downstream choice. If the URL graph is a mess, the copy does not matter.
This site is the example I can show. Three GitHub Pages projects used to compete at /, /cv/, and /seo/. The consulting homepage was a second product. Unpublishing it was not a marketing campaign. It was a platform decision: one Person, one title, one graph.
One intent per URL
A URL that is a résumé, a services page, and a portfolio is three pages sharing a path. Crawlers pick one. Humans bounce.
I want:
/for who I am and what I lead/work/for evidence/cv/for dates and print/blog/for notes/seo/as a redirect, not a second homepage
If a path still sells a different identity, it is a bug, even if it ranks.
Canonicals and redirects
GitHub Pages will not give you a real HTTP 301 on every host. That is a constraint, not an excuse. I still ship a canonical, a meta refresh, and noindex on the trampoline so the old consulting URL cannot come back as a competing document.
The same pattern applies to unpublished posts. Keep the slug, point it at the index, drop it from the sitemap. Deleting a URL without a successor trains crawlers to distrust the host.
Canonicals are for duplicates you meant to have (www vs apex, trailing slash). They are not for two products you have not chosen between.
Crawl budget as an engineering limit
Empty tag archives, empty category pages, and parameter URLs waste crawl. I noindex empty archives. I do not generate a thousand tag stubs for posts I have not written.
A homepage that fetches GitHub at runtime and renders “Unable to load repositories” is also a crawl problem. The bot stored a failure. Static flagship work is slower to edit and safer to index.
What this is not
This is not a pitch for retainers, percentage lifts, or Ecuador SERP packages. Those claims do not belong on a lead-engineer site unless I can show the measurement. I will not.
Engineering SEO is making the site say one thing, at one URL, with a title that matches the document. The rest is content, and content is another queue.