Concepts and glossary

DispatchSEO invents a few words and borrows a few others. This page defines all of them. If something in the dashboard or in your agent's output doesn't make sense, it's probably here.

The core objects

Project

One managed website. A project owns its own keywords, queue, rankings, settings, and its own MCP token.

One DispatchSEO install can run any number of projects. Which one a request belongs to is resolved three different ways: the dashboard uses a cookie, the MCP server uses the bearer token (the token is the site selector), and scheduled jobs simply loop over every project.

Suggestion

One idea in the queue. It's the unit everything else revolves around.

A suggestion has a type - guide, tool, backlink, or update - and a status that moves in one direction:

pending → approved → in_progress → done, or rejected at any point.

Your agent files suggestions as pending. You approve them on the Queue page. The builder picks up approved ones in queue order. Each carries a rationale (why this is worth doing) and a spec (the brief: angle, outline, search-result notes).

Keyword

A search term you're tracking. Carries search volume, difficulty, and intent when you have a paid data source connected, and a status of tracking or paused.

Rank check

One measurement of where you placed for one keyword at one moment. Append-only

• a keyword accumulates a history you can see as a trend line. A null position means you weren't in the top 100.

Page

Something you published - a guide, a tool, a landing page. Recorded with its URL, type, target keyword, and the pull request that shipped it.

A page is marked live only once its URL actually returns HTTP 200. Merged is not the same as published, and the dashboard tells the two apart.

Backlink prospect

A domain worth trying to get a link from. Moves through new → contacted → acquired, or rejected.

Content and quality

Guide vs tool

The two things the pipeline builds. A guide is an article. A tool is a small interactive widget - a calculator, a converter, a checker - that earns links because it's useful rather than because it's persuasive.

Tools get an extra gate: a real headless-browser test that the thing actually works before the PR can merge.

Archetype

The shape of a guide - tutorial, comparison, data study, opinion, reference. You can disable archetypes you don't want on the Instructions page.

Quality bar

The rule set every research and build run obeys. A search-volume floor, an audience-fit test, and a difficulty ceiling that scales with your domain's strength - so a three-week-old site isn't sent after keywords it has no chance at, and an established one isn't held back.

Plus the hard rules: never push to main, always open a labeled pull request, never fabricate data.

Sameness gate

A check that a draft doesn't read like a re-skin of your own existing posts. It compares the opening word run, the heading skeleton, and stock phrases against everything you've already published, and fails the draft if it converges.

This is the fingerprint of scaled content that a competitor-only tool can't see, because it requires knowing your corpus.

Publishing pace

At most one guide per UTC calendar day. Flat, permanent, not tied to your site's age.

The pace isn't what protects you from looking spammy - the quality gates are. It exists so a queue of thirty approved ideas doesn't become thirty posts in a week.

Content preferences

Your overrides on the above: free-text house rules injected into every build, archetypes you've disabled, and skeleton blocks (TL;DR, comparison table, visuals, FAQ) you don't want.

Automation

Mode

How much you're in the loop:

• Semi-automatic - the agent researches and builds, you approve ideas and merge pull requests. The default.
• Automatic - hands-off. Ideas self-approve, green PRs self-merge.
• Custom - shown when your individual toggles don't match either preset.

Automation

One toggleable behaviour - the nightly rank check, the weekly research run, auto-merge, and so on. Twelve of them, each with a schedule and evidence of when it last ran, on the Automations page. Some are locked on because the product doesn't work without them.

Journey stage

Where your site sits on the honest SEO timeline: setup → foundation (months 1-2) → first signals (month 3) → traction (months 4-7) → compounding (6-12+).

It's derived only from things that can't go backwards - first-ever events and cumulative counts - so a bad week can never demote you.

The agent side

Playbook / instructions

The versioned procedure your agent follows for a given job. Ten of them: install, setup, research, build-guide, build-tool, report, backlinks, trend-scan, trend-expand, geo-scan.

These live on the backend, not in your repo. Your repo keeps only thin commands that fetch the current playbook before acting - so improvements reach every connected site on its next run without anyone updating anything.

Conventions

Your site's facts, written by the agent during setup and kept in .dispatchseo/conventions.md in your own repo: the stack, the build command, where content lives, theme tokens, voice rules, example posts to imitate.

This is what stops the agent guessing at your project's shape every time.

Site profile

Your product's identity in the lengths directories ask for - name, tagline (60 characters), short description (160), long description (300-600), categories, tags. The backlink playbook fills submission forms from it.

Pipeline pack

The set of files an install writes into your repo: the seo-* GitHub Actions workflows, the MCP configuration, the slash commands. Served from the backend so it stays current.

In-stack builder

On a self-hosted Docker install, a container running headless Claude Code that polls for scheduled work. It exists because a machine at home has no public URL for GitHub Actions to call back into, so the schedule has to run inside your own stack instead.

Data and tracking

Keyword source

Where difficulty and volume numbers come from: gsc (free, Search Console and Autocomplete only), serpapi (free tier, live search results), or dataforseo (paid, full metrics). See Keyword data sources.

GSC stats

A daily snapshot from Google Search Console - clicks, impressions, click-through rate, average position, and your top queries and pages. Free, so it's re-snapshotted every few hours.

Domain rating

A 0-100 estimate of how strong your domain is, refreshed weekly. Feeds the dynamic difficulty ceiling in the quality bar.

AI visibility

Whether AI answer engines - Google's AI Overview, ChatGPT, Claude, Perplexity, Gemini - cite your site when asked questions you should own. Sampled on a schedule and recorded per engine, along with which domains got cited instead of you.

Trend topic

A subject the trend radar caught being talked about right now - not yet a guide idea.

The radar runs in two stages on purpose. Stage one finds subjects with evidence (why now, what signals, which sources). Stage two takes one subject you picked and expands it into three to five concrete, validated angles. This keeps the noisy step cheap and the expensive step deliberate.

Cron run

A logged execution of a scheduled job. Failures raise a banner on the dashboard and send an email. Read Schedules and jobs for what runs when.

Setup

Install progress

The six-step checklist stamped while your agent installs the pipeline: workflows → adaptation → repo settings → content home → site facts → research. It's what the wizard's live progress screen is reading.

Finishing all six is not the same as being installed - a separate, stricter check verifies the work before the dashboard unlocks.

Project key

The per-project bearer token your agent uses to reach the MCP server. Viewable and rotatable on the Settings page. Anyone holding it can read and change that project's data, so treat it like a password.

Cron secret

The shared secret scheduled jobs present when calling the backend. Generated during setup and stored in the database.

Next