Please install Yoast or RankMath to use breadcrumbs.

Behind the Scenes: How AIOX Processes a Single Blog Post

“What actually happens when I click Process on a blog post?” is the question we get most from new AIOX operators. This article walks through the pipeline step-by-step, with what each stage does and what comes out.

Stage 1 — Fetch

The AIOX dashboard sends a signed REST request to the Suite plugin’s /wp-json/aiox/v1/post/{id} endpoint. The plugin returns the canonical post data: title, slug, body HTML, taxonomies, custom fields, featured image URL, author, published / modified dates. Roughly 200ms.

Stage 2 — Preprocessing

HTML is stripped to canonical text — paragraphs, headings, lists, code blocks preserved; layout cruft (theme-injected wrappers, ad slots, related-post widgets) discarded. Whitespace normalised. Embedded media (videos, tweets, audio) catalogued separately so they can be referenced by Capsule without bloating the text body. Roughly 100ms.

Stage 3 — Content-type detection

A combination of heuristics and a small classifier picks the most appropriate Schema.org type:

  • Heuristic: heading pattern “Step 1, Step 2…” → likely HowTo.
  • Heuristic: presence of an ingredient list + cooking time → Recipe.
  • Heuristic: question/answer pattern → FAQPage.
  • Heuristic: structured product fields (price, SKU) → Product.
  • Fallback: Article.

Below a 0.7 confidence threshold the post is flagged for human review rather than guessed. Roughly 400ms.

Stage 4 — Entity extraction

Named entities (people, organisations, products, places, events) extracted via Gemini. Each entity is linked to Wikidata where possible (so AI systems can disambiguate “Apple” the company from “apple” the fruit). Confidence scores attached to each. Low-confidence entities are stored but not emitted. Roughly 800ms.

Stage 5 — Intent inference

The classifier assigns one or more intents: informational, commercial, navigational, transactional. Multi-label — a single tutorial can be both informational (“how to do X”) and commercial (“buy our tool to do X faster”). Useful for AI systems deciding whether to surface the content in answer-mode vs. shopping-mode. Roughly 400ms.

Stage 6 — Audience inference

Who is this for? Detected from vocabulary, depth, and topic. Tags drawn from a controlled vocabulary: developers, marketers, ops, executives, general consumers, students. Used by some AI systems to filter content by reader sophistication. Roughly 300ms.

Stage 7 — Schema assembly

Full JSON-LD per the detected type. Author, dates, publisher, image, and AIOX-namespaced extensions (intent, audience, license, capsuleId). Validated against Schema.org and AIOX v1 before emission. Validation failures abort the pipeline with a specific error pointing to which field needs fixing. Roughly 200ms.

Stage 8 — License merge

The sitewide license config (from Content Licensing app) is merged with any per-post override metadata. The result becomes the aiox:license field. Roughly 50ms.

Stage 9 — Signing

The Capsule (minus the signature itself) is hashed and signed with the domain’s Ed25519 private key. The key never leaves the server. Roughly 30ms.

Stage 10 — Push back

The signed Capsule is sent back to the Suite plugin via the same REST channel. The plugin stores it locally and starts serving it at /aiox-capsule/{slug}/ with proper cache headers. The site’s /aiox.json manifest is updated. Roughly 500ms (depending on plugin response time).

Stage 11 — Notifications

Webhook fires (if configured). Processing log row added in the dashboard. Token usage recorded against the customer’s account. Roughly 50ms.

Total elapsed

For a typical 1,000-word post: 3–5 seconds end-to-end. Token cost: 1,200–1,800. The Free plan covers about 7-8 such posts per month.