A stack of printed magazine pages with an image grid, one image blue, next to a printing press

How We Built Our Blog for SEO: Static Generator, Schema Markup and Covers

You're reading this in a blog we built for ourselves. The goal sounded simple: articles should be found in search, read well on a phone, and go live without hand-coding each page. Behind it lies a set of decisions, from how the text is stored to how search engines learn about a new article on the day it's published.

Below is how a company blog is put together, using the Sphere blog as the example. Everything described comes from the project's code: the generator, the template, the feed, the files for search engines and the deployment step. There are no traffic results here: the blog launched recently, and we'll share the numbers later, once they've built up.

Why an agency needs a company blog

We need a company blog for two things. The first is SEO: someone searches for how much a website costs, how to write a project spec or where to start with AI, and lands on our article. The second is answering the questions clients ask us at first meetings. If an answer is already covered in an article, we can just send the link.

A blog like this has a catch: an article has to work for both search engines and people. A search engine needs a clean structure, markup and fast indexing. A person needs short paragraphs, clear navigation and a summary up top to decide in a minute whether to keep reading.

Requirements for the blog

Before development, we wrote down what the blog had to do. The list looked like this:

The last point matters more than it seems. Once there are dozens of articles, "we don't do that" agreements stop working if nothing actually enforces them.

How the static site generator works

The blog is a static website with no database and no admin panel. Articles live in the repository as JSON files, the layout for every article is a single HTML template, and the order and publication dates are kept in a list inside the build script. The static site generator is written in Python and runs with one command.

  1. Loading articles. The build reads each article's JSON: title, meta title and description, lead, primary and secondary queries, body text, questions and answers. It immediately checks that the URL matches the file name and that the category is on the allowed list.
  2. Markup validation. The text may contain only paragraphs, subheadings, lists, emphasis, links, tables and a few blocks from our formatting guide. If it hits any other tag, an extra class or a link to a blog article inside a paragraph, the build fails and names the article.
  3. Building the page. The text goes into the template, subheadings get anchors, a table of contents is built from them, and "Read also" blocks are filled with the title, category and date of the linked article.
  4. Feed and categories. A single feed template produces the main feed and five category pages, each with its own title, description, H1 and structured data.
  5. Files for search engines. The sitemap, RSS feed and llms.txt are updated, and a 1200×630 social image is drawn for each article.

The build script creates the social image itself: it takes the article's cover, darkens it toward the bottom, adds the logo, the category with the date, and the title in caps. The font size is picked so the title fits in four lines.

The SEO side: markup, categories and indexing

A blog built for SEO depends on small details that are easy to forget when pages are coded by hand. So everything that can be generated is generated, the same way for every page.

Schema.org article markup

Every article gets BlogPosting markup: title, description, publication and modification dates, category, target queries, word count, reading time, cover and social image, and publisher. Next to it is a BreadcrumbList with the trail "Home → Blog → category → article." If the article has questions, FAQPage is added.

The FAQ markup is generated from the same field as the FAQ section at the end of the text, so the questions on the page and in the markup never drift apart. The feed and category pages are marked up as CollectionPage with an ItemList that lists every article in order.

Table of contents, summary and typography

Subheading anchors aren't just for readers: search engines can show links to sections right in the results. The table of contents sits after the summary block, a three-to-five-point recap at the top of the article.

A typography script adds non-breaking spaces after short words, keeps dashes attached to the preceding word, and stops short particles from wrapping onto a new line on their own. A small thing, but without it text looks sloppy on a narrow screen.

Category pages targeting their own queries

Categories aren't just a filter; they're separate pages. The development category has its own heading about building websites, web services and apps, and the cases category is about breakdowns of our projects. That way each category can rank for its own group of queries instead of competing with the main feed.

Sitemap, RSS, llms.txt and IndexNow

The build updates the blog section of the sitemap: the feed, categories and articles, with modification dates and cover images. The sitemap URL is listed in robots.txt. Aggregators read the RSS feed, and it helps search engines discover new articles faster. llms.txt gets a list of articles grouped by category with descriptions; this file is read by AI-powered search engines.

The biggest boost to indexing speed is IndexNow. On every deployment, search engines receive the list of changed pages instead of waiting for the next crawl. A script compares commits, takes the changed site files and keeps only indexable pages. Deleted articles also go on the list so the search engine re-checks the URL. A failed notification doesn't block the deployment.

What we builtWhy it matters for SEO
Static pages generated from JSONFast loading; the crawler sees the full text right away
BlogPosting, BreadcrumbList, FAQPageRich results and a clear structure for search engines
Anchors and table of contentsSection links in search results, easy navigation
Category pages with their own title and H1Separate landing pages for query groups
Sitemap, RSS, llms.txtSearch engines and AI models find every article
IndexNow on deploymentNew and changed pages get indexed faster
1200×630 imagesClean previews on social media and in messengers

Designing the feed and the article page

We didn't want the feed to be a standard grid of identical tiles. On desktop it runs in five columns, on tablets in three, on phones in two.

On desktop, a script rebuilds the layout on every page load. A row holds one to five cards, but any two adjacent rows together hold at least six articles, and any three adjacent rows at least nine. That way a half-empty row never sits next to another half-empty one. Within a row, cards land in random columns, while the left-to-right order of articles is preserved.

The feed layout is new on every load, but the article order and the rules for adjacent rows always hold.

Category filtering without a page reload

Categories in the feed are ordinary links to category pages, so the search crawler can follow them. In the main feed, the click is intercepted: cards are filtered in place, and the browser's address changes to the category URL. Outgoing cards fade, the remaining ones glide to their new positions, and new ones fade in.

Cover images and the focal point

All covers share one style: a black-and-white shot with an accent in our signature blue. Each is stored as JPEG and WebP in several sizes: a large frame for the article header and a smaller one for cards. The browser picks the right size for the screen width.

A card's height is set by an aspect ratio we choose to suit the cover's subject. To keep the key detail from being cropped, each cover has a recorded focal point, and the build shifts the frame so that point lands in the center of the card. In the article header, the cover is shown full-screen, in black and white with heavier grain.

On hover, the cover turns into a negative: an SVG filter desaturates it and inverts the tones along a "white → brand blue → black" scale. Lights go black, darks go white, and midtones turn blue.

The article page

Below the header come breadcrumbs with the category and reading time, then the text with a table of contents and formatting blocks. At the end there's a contact block and five "More on this topic" cards: first articles from the same category, then the latest from the others.

How an article gets published

Code solves only part of the problem. The other part is the process an article goes through on its way to publication.

  1. Brief. It sets out the tone, accuracy rules, the approved facts about the company, SEO requirements and the article file format.
  2. Topic. Topics are agreed in advance, and each article has one primary query and several secondary ones.
  3. Fact-checking. Anything that needs confirmation goes into a needs_check field, and from there into a shared list of questions for the team. The facts are filled in before publication.
  4. Validation and build. A separate script checks the fields, markup and length of the article using the same code as the build. Then the generator rebuilds the blog.
  5. Deployment. Once the changes are published, the site updates on the server and IndexNow sends search engines the list of changed pages.

What's next

The blog launched recently, so it's too early to talk about results. Once we have data on indexing, search queries and inbound leads, we'll share it in a separate post.

The same approach (a static site, a generator, automatic markup and search engine notifications) isn't only for agencies. If you need a blog for your company website or a whole new site, you'll find examples in our portfolio, and every article in the blog feed.

FAQ

How do you build a company blog that works for SEO?

Build in what a search engine needs on every page: a clean structure with subheadings, article and FAQ structured data, category pages targeting their own queries, a sitemap, and fast notification of search engines about new pages. It's better when all of this is generated automatically rather than by hand for each article.

Why a static site for a blog instead of a CMS?

Static pages load fast and give the search crawler the full text right away. We don't need an admin panel: articles live in the repository as files, and the generator rebuilds the blog with one command. For a team that wants an in-browser editor, a CMS with the same markup rules works too.

What does FAQ structured data do?

FAQPage markup tells the search engine that the page contains questions and answers. It helps the engine understand the structure and can affect how the snippet looks. The key is that the questions in the markup match what the reader sees; ours are generated from a single field.

What is IndexNow and why use it?

IndexNow is a protocol that lets a website notify search engines about changed pages itself. Bing and several other search engines support it. Instead of waiting for the next crawl, the search engine gets the list of URLs right after deployment.

When will you share the blog's traffic results?

The blog launched recently, and there isn't enough data to draw conclusions yet. We'll share indexing and search traffic results once they build up.

Let’s discuss your project
Related reading