# What AgentBlog is

Source: https://docs.agentblog.dev/
Summary: A blog for your Next.js app that search engines rank and AI assistants cite, installed with one command, with agent skills that know how to write for it.



AgentBlog is a blog you install into a Next.js app you already have. One command
writes 71 files into your repository: the routes, the structured data, the
sitemap, the feed, the social images, and a set of skills your coding agent uses
to write the posts. The files are yours after that, the same way shadcn/ui
components are yours.

## Who this is for [#who-this-is-for]

You have a product with a website, and you know a blog would bring people to it.
You would rather not spend a week reading about canonical tags to find out
whether yours are right.

That is the whole audience. You do not need to know what JSON-LD is. You do need
a Next.js app and about ten minutes.

<Cards>
  <Card title="Quickstart" href="/quickstart" description="Install it and see a post render. About five minutes." />

  <Card title="Full installation" href="/installation" description="Both install paths, the four files you edit, and what each one does." />
</Cards>

## Why it exists [#why-it-exists]

Three things go wrong between wanting a blog and having one that works.

### You are not sure how to build one [#you-are-not-sure-how-to-build-one]

The usual answers are a separate WordPress site, a Webflow or Framer page, or a
hosted platform on a subdomain. Each of those splits your site in two: your
product on one domain, your writing on another, two design systems, two bills,
and search engines treating them as unrelated.

A blog inside your Next.js app has none of that. It is a route in your existing
project. It reads your design tokens, deploys with your app, and lives on your
domain, where the links people give you help the pages that sell your product.

### You cannot tell whether it is set up correctly [#you-cannot-tell-whether-it-is-set-up-correctly]

This is the expensive one. A blog can look completely finished and be invisible.

The page renders in your browser, so you assume a crawler sees the same thing.
It usually does not. Content that only appears after JavaScript runs is missing
for AI crawlers, which do not run JavaScript at all. A new post never reaches
`sitemap.xml` because that route was cached. The author field is a plain string,
so nothing connects the post to a person. None of that shows up as an error.
Nothing turns red. You find out months later when nothing ranks and nothing
cites you.

AgentBlog was built from the current guidance Google and the AI search vendors
publish, and the parts that cannot be verified by reading are verified by
fetching. `agentblog doctor` reads your config and reports what is wrong.
`agentblog doctor --url` fetches your live site as GPTBot, ClaudeBot,
PerplexityBot, OAI-SearchBot, and Googlebot, and tells you which of them your
CDN is turning away.

### You do not want to write the posts [#you-do-not-want-to-write-the-posts]

Posts are MDX files in your repository, which makes them something a coding
agent can genuinely maintain. Installing AgentBlog also writes four skills into
`.claude/skills/`, so your agent already knows the format: the direct answer
under each heading that an AI assistant can lift, the question-shaped headings,
the tables, the citations, and the internal links that keep a new post from
being an orphan.

You ask for a post. The agent writes a draft in your voice, using your existing
posts as the reference. You edit it and merge it.

## How it installs [#how-it-installs]

The same way you install a shadcn/ui component, because it uses the same
mechanism.

```bash
npx agentblog@latest init
```

That command checks your project meets the requirements, adds the AgentBlog
registry to your `components.json`, copies the files in, patches the two config
files a file copier cannot reach, and asks you for your site URL and brand name.
It prints everything it is about to do first, backs up every file it changes,
and does nothing when you run it twice.

There is no package of ours in your `package.json` afterwards. Nothing to
upgrade around, nothing that breaks when we ship a change you did not ask for.

## What you actually get [#what-you-actually-get]

| Piece                | What it means for you                                                                         |
| -------------------- | --------------------------------------------------------------------------------------------- |
| Blog routes          | An index, post pages, category and tag pages, and author pages, all prerendered as plain HTML |
| Structured data      | The machine-readable summary Google and the assistants read, connected and typed              |
| Sitemap, feed, pings | Search engines find new posts on the day you publish rather than the week after               |
| Social images        | A generated card per post, so a shared link is not a bare URL                                 |
| Your design system   | The blog uses your colours, your fonts, and your components. It ships no theme of its own     |
| Agent skills         | Four skills that write posts, refresh them, finish the setup, and audit before you publish    |
| `agentblog doctor`   | One command that tells you whether any of the above is actually working                       |

## What it is not [#what-it-is-not]

It is not a hosted platform, and there is no dashboard. It is not a template you
clone, because your app already exists. It is not a theme: the components carry
no colours of their own, so `/blog` looks like the rest of your product on the
day you install it.

<Cards>
  <Card title="Start installing" href="/quickstart" description="The five minute path." />

  <Card title="How search and AI read a site" href="/concepts/how-ai-search-reads-your-site" description="The background, if you want it before you commit." />
</Cards>