AEO Step-by-Step Extraction Patterns for How-To Citations

This framework defines ten extraction patterns — covering HowTo schema, step language, prerequisites, verification, and troubleshooting — that help AI engines cite individual how-to steps cleanly in answers and overviews.

TL;DR

AI engines cite step-by-step content when each step is a self-contained, action-first statement wrapped in semantic markup. Use HowTo schema for the page envelope, ordered lists with explicit step numbers, prerequisites and time estimates above the steps, and a verification line per step. Pair text with diagrams or screenshots labeled with step numbers, and add a troubleshooting subsection per step where common errors block completion.

When to use this framework

Apply these patterns to any content that answers a how-to query: tutorials, runbooks, troubleshooting guides, and configuration references. The patterns are platform-agnostic — they help Perplexity, ChatGPT, Google AI Overviews, and Bing Copilot extract steps cleanly even where the SERP rich result has been deprecated.

The 10 patterns

Pattern 1 — HowTo schema as the page envelope

Wrap the page in schema.org/HowTo with name, description, totalTime (ISO 8601 duration), estimatedCost, tool[], supply[], and step[] arrays. Each step is a HowToStep with name, text, optional image, and an absolute url anchored to the in-page step heading. Even with Google's deprecation of HowTo rich results in mainstream SERPs (announced 2023), the schema continues to feed AI extraction layers.

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "Configure HTTPS on Nginx",
  "totalTime": "PT15M",
  "step": [
    {"@type": "HowToStep", "name": "Install certbot", "text": "Run apt install certbot.", "url": "https://example.com/guide#step-1"}
  ]
}

Pattern 2 — Action-first step language

Start every step with a present-tense verb: "Install," "Run," "Configure," "Verify." Avoid lead-ins ("Now you should," "Next you'll want to"). Action-first phrasing matches the imperative mood AI engines look for when extracting instructions.

Pattern 3 — Numbered ordered lists, not bulleted

Use

Pattern 4 — Prerequisites disclosed above the steps

List prerequisites in a dedicated H2 section before step 1: software versions, accounts, permissions, prior knowledge. AI engines often cite this block separately when the user query is "what do I need to…" rather than "how do I…". Include version constraints with explicit ranges (Node ≥ 18.17).

Pattern 5 — Time estimate disclosed above the steps

Declare both a total time and per-step time when steps differ materially. Format the total time in ISO 8601 inside totalTime and again as human prose ("about 15 minutes") in body text. Time estimates rank well in featured snippets and frequently appear as a side card in AI overviews.

Pattern 6 — Verification line per step

Close every step with a one-sentence verification: "Confirm that…," "Run nginx -t and expect syntax is ok." Verification text is highly extractable because it pairs an expected outcome with a concrete check, which is exactly the shape AI engines reuse when answering "how do I know it worked?" follow-ups.

Pattern 7 — Step images named by step number

If you include screenshots or diagrams, name and caption them with the step number ("Figure 1.3 — Step 3 output"). Add alt text that restates the step's verification line, not the action. AI engines that perform image-text alignment use captions and alt text to bind a screenshot to the correct step.

Pattern 8 — Troubleshooting subsection per failure-prone step

Where a step commonly fails, add a ### Troubleshooting subsection inside that step's H3 block. List two to four error symptoms with one-line fixes. AI engines lift these symptom-fix pairs into answers when the user query is an error string.

Pattern 9 — Video step alignment with chapter markers

If you embed a video, publish chapter markers that align to the textual steps (Chapter 3 — Step 3 — Configure HTTPS). Use Clip schema to expose the chapter list. Aligned video boosts retention and gives AI engines an alternative source when text extraction is partial.

Pattern 10 — Completion criteria block at the end

Close the page with a "You're done when…" block listing the success state in two to four bullets. This block is heavily favored as an answer terminator because it confirms task completion in a single, citable chunk.

Worked examples

Example A — Nginx HTTPS: ten steps, total time PT15M, troubleshooting under steps 4 and 7, completion block: "curl -I https://yourdomain returns HTTP/2 200."
Example B — Stripe webhook setup: prerequisites list (Stripe account, public URL, signing secret), eight steps, verification line per step using stripe trigger.
Example C — Postgres point-in-time recovery: prerequisites for backup retention, ten steps, video chapter alignment for the restore phase.
Example D — GitHub Actions release workflow: seven steps, troubleshooting for permission errors, completion block: "a release tag appears under Releases."
Example E — Notion API page creation: prerequisites for OAuth scopes, six steps, screenshots labeled by step number.
Example F — Cloudflare DNS migration: ten steps, time estimate PT45M, verification with dig +short.
Example G — Docker Compose deployment: prerequisites for Docker Engine, eight steps, troubleshooting under volume-mount step.
Example H — Terraform module publish: prerequisites for registry account, seven steps, verification by registry URL fetch.
Example I — Slack app slash command: prerequisites for workspace admin, six steps, completion block citing the /command test.
Example J — macOS launchd agent install: prerequisites for plist permissions, five steps, verification using launchctl list.

Common implementation mistakes

Mixing imperative and descriptive voice across steps. Pick one (imperative) and hold it for every step.
Putting prerequisites inside step 1. Move them above the step list so they can be cited independently.
Bulleted lists for sequences. Switch to numbered ordered lists.
Image alt text restating the action instead of the verification. Use the verification line so the alt text augments the answer.
Skipping a verification line because "the step is obvious." Verification lines are disproportionately reused in AI answers; the cost of writing one is small.

FAQ

Q: Does HowTo schema still help if Google deprecated the rich result?

Yes. Google retired the HowTo rich result in mainstream SERPs in 2023, but the schema still feeds AI Overviews and other extraction layers, and other engines (Bing, Perplexity) continue to use it.

Q: Should every step have a screenshot?

No. Screenshots help where the UI is ambiguous; they hurt by adding bloat and ALT-text noise where the action is obvious from text. Aim for screenshots on roughly one in three steps.

Q: Where should troubleshooting live?

Under the failure-prone step's H3, not as a global section at the end. Per-step troubleshooting is cited cleanly when the user query is an error string; a global troubleshooting section is harder to lift.

Q: How long should each step be?

Thirty to ninety words is the sweet spot. Below thirty, AI engines often merge two steps; above ninety, they split or summarize away the verification line.