GETTING STARTED

Publish your first AI-verified article.

From topic to publish-ready in under six minutes. A single POST triggers the seven-stage pipeline; you watch every inner step over Server-Sent Events.

UNDER 6 MIN·5 steps·Updated 2026-04-20

PRELUDE · AUTH

Sign up and get a JWT.

Every pipeline run needs a Bearer JWT that carries your workspace_id. Sign-up creates the personal workspace in the same transaction; sign-in returns the token.

# 1. Sign up. Personal workspace auto-created in the same transaction.
$ curl -X POST http://localhost:7878/api/auth/sign-up \
    -H "Content-Type: application/json" \
    -d '{"email":"jane@acme.co","name":"Jane","password":"S3cure!pw"}'

→ 201 Created (empty body; verification email sent)

# 2. Sign in. accessToken carries your workspace_id.
$ curl -X POST http://localhost:7878/api/auth/sign-in \
    -H "Content-Type: application/json" \
    -d '{"email":"jane@acme.co","password":"S3cure!pw"}'

→ 200 OK
  { "user": {...}, "accessToken": "eyJhbGciOi...", "refreshToken": "..." }

$ export JWT=eyJhbGciOi...

TIP

No verification-click required in dev. Sign-in returns tokens immediately. Verification email is sent via Mailpit for realism, but sign-in does not block on verification status in code.

STEP 01 · STRATEGY + OUTLINE

Submit your topic.

Pass your topic as user_message. The Strategy stage turns it into a position, researches what already ranks, and scopes the supporting claims. The Outline stage normalises the plan into H2/H3 sections ready for Research. You watch both run via SSE.

# Trigger the pipeline. JWT from /api/auth/sign-in carries workspace_id.
$ curl -X POST http://localhost:7878/api/article-jobs \
    -H "Authorization: Bearer $JWT" \
    -H "Content-Type: application/json" \
    -d '{
      "brand_voice_id": "33333333-...-...",
      "user_message": "How AI coding agents change SDLC",
      "audience_level": "intermediate",
      "audience_type": "technical",
      "tone_preset": "professional",
      "article_word_count_preset": "RANGE_2500_3500"
    }'

→ 201 Created
  { "id": "a9c3b1e4-..." }

SSE emits:

generating_seedsfetching_keywordsclustering_keywordsfetching_serpfetching_competitor_contentparsing_competitor_contentanalyzing_competitorsanalyzing_gapsplanning_articlebuilding_content_briefgenerating_outlinevalidating_outline

TIP

Reuse the same position across the cluster. Pillar article and sub-articles keep one angle. Depth differs between them.

STEP 02 · RESEARCH

Watch the research populate.

Research stage extracts topics from the outline, runs SERP queries per topic, and captures facts. The result is a 4-table tree in Postgres (topics → queries → SERP results → facts). You follow along via SSE; the stage takes 40 to 90 seconds for typical articles.

# Subscribe to the SSE stream for live per-step updates.
$ curl -N -H "Authorization: Bearer $JWT" \
    http://localhost:7878/api/article-jobs/a9c3b1e4-.../status

→ data: { step: "extracting_topics",      status: "processing" }
  data: { step: "researching_topics",     status: "processing" }
  data: { step: "consolidating_research", status: "processing" }
  ...

SSE emits:

extracting_topicsresearching_topicsconsolidating_research

TIP

Expect 8 or more tier-A sources for a YMYL topic. Below that, verify-time claim failures climb sharply.

STEP 03 · GENERATE

Generate with your voice.

The Generate stage loads outline + plan + research from Postgres and drafts the article section by section. Your brand_voice_id fingerprint is applied at draft time through Brand Voice; Brand Guard filters forbidden phrasing inline. The result is article_md + FAQ + JSON-LD.

# SSE continues to stream. These step names appear during Generate.
data: { step: "preparing_generation", status: "processing" }
data: { step: "generating_sections",  status: "processing" }
data: { step: "generating_faq",       status: "processing" }
data: { step: "assembling_article",   status: "processing" }

SSE emits:

preparing_generationgenerating_sectionsgenerating_faqassembling_article

TIP

Voice-score under 0.80 means the draft does not sound like you yet. Re-fingerprint on 15 to 30 of your most recent articles and re-run.

STEP 04 · VERIFY

Verify every factual claim.

The Verify stage walks the draft sentence by sentence. For every factual assertion, Perplexity sonar-pro checks the cited source and scores the match. Unverified claims are flagged and rewritten before the stage closes. Typical run: 40 to 60 claims checked, 2 to 6 flagged on a strong source base.

data: { step: "extracting_claims", status: "processing" }
data: { step: "verifying_claims",  status: "processing" }
data: { step: "fixing_claims",     status: "processing" }

# article_verification_results row written:
#   claims_extracted: 58
#   claims_verified:  54
#   fixes_applied:    4

SSE emits:

extracting_claimsverifying_claimsfixing_claims

TIP

Fix flagged claims by rewriting the draft. The cited source is ground truth; the draft bends to match.

STEP 05 · OPTIMIZE + QUALITY

Ship through the quality gate.

Optimize normalises formatting, runs SEO and GEO audits, scores the article on combined_score. Quality (the finalize orchestrator) runs an 8-phase pure-function audit: burstiness CV above 0.55, opener diversity above 58 percent, AI pattern ratio under 40 percent, HIGH-severity tells fewer than 10. Fail any one and the job terminates with status FAIL.

data: { step: "normalizing_article",   status: "processing" }
data: { step: "running_seo_audit",     status: "processing" }
data: { step: "running_geo_audit",     status: "processing" }
data: { step: "running_quality_audit", status: "processing" }

data: { step: "running_quality_audit", status: "completed" }

# article_quality_results row:
#   quality_score:    0.87
#   status:           PASS
#   needs_humanization: false

SSE emits:

normalizing_articlerunning_seo_auditrunning_geo_auditrunning_quality_audit

TIP

On FAIL, the Editor engine fixes patterns without full rewrites. It targets the tells, not the voice.

AFTER THE GATE

Retrieve the finished article.

Endpoint in development

The public endpoint for fetching a completed article by job id is in development. Once live, a single GET /api/article-jobs/:id will return the status, the final article_md, and the quality verdict.

Until then, the article is written to Postgres in article_generation_results.article_md (post-generation) or the rewritten version in article_verification_results (post-verify). Workspace-level preview is available on request.

ROADMAP

Per-stage artefact endpoints are also in development. When GET /api/article-jobs/:id/artifacts/research ships, this tutorial gains an optional step for inspecting research sources, claim verifications, and quality-audit breakdowns before publication.

Verify an existing draft (coming soon)

Try it in AI Checker