Procedural queries are the loudest, most underserved slice of AI search traffic right now. Anyone who has ever watched a ChatGPT session knows the rhythm: someone asks how to do a thing, the model returns a numbered list, and the user follows it. Brands keep publishing 2,000-word thought pieces and then wonder why none of it shows up when a buyer asks “how do I migrate from X to Y.” The shape of the answer is not the shape of your content.
The mechanic
There are three forces working together here, and once you see them you cannot unsee them.
First, AI chat shifts query mix. Classic Google biased toward navigational and short head terms; people gave it nouns. LLMs are conversational, so people give them verbs — “how to,” “what do I do if,” “walk me through.” That category is enormous, and procedural intent rewards a very specific content shape: enumerated, sequential, imperative.
Second, passage retrieval rewards structured steps. AI engines do not read your post; they slice it at heading boundaries, embed each chunk, and score chunks individually against the prompt. Step-numbered content slots in perfectly. Each step is already a self-contained answer-unit with a verb, an outcome, and a clear boundary. Compare that to a flowing essay where the model has to guess where one idea ends and the next begins. That is part of why 68.7% of AI-cited pages follow a strict H1→H2→H3 hierarchy, and why 44.2% of LLM citations land in the first 30% of a page — structure makes retrieval cheap.
Third, HowTo signaling still pays even when it is not “officially” rendered. Google retired the rich result for most queries, which led a lot of operators to rip the markup out of their templates. Bad call. The structured data is still parsed by AI answer-fetchers — OAI-SearchBot, ChatGPT-User, PerplexityBot — and it tells the machine “this page is a procedure, with N ordered steps, each with a name and a body.” That is metadata your prose alone cannot give them.
What to do this week
Pick the top ten procedural queries in your category — the literal “how to [verb]” and “what’s the process for [X]” prompts your buyers are already asking ChatGPT and Perplexity. Type each one into all four major engines and write down who is being cited. If the answer is competitor blog posts, third-party how-to roundups, or — in 2026, still — Reddit threads, that is your map of which surfaces to either replace or join.
Rewrite or build the matching posts in this shape. Headline names the outcome (“How to migrate from HubSpot to Customer.io without losing automations”). Lede is a 40-word “here is the short version” answer-unit, sitting in the first 30% of the page so it gets cited verbatim. Then a numbered series of H2s, one per step, where the H2 is the step name in imperative form (“Export your existing workflows”). Each step body is 50 to 150 words, restates the subject (do not write “now do this” — write “now export the workflows”), and ends with an explicit outcome sentence. Avoid cross-references like “as mentioned above” — each step has to make sense in isolation because it will be retrieved in isolation.
Add HowTo JSON-LD. `@type: HowTo`, a `name` matching the headline, an `estimatedCost` and `totalTime` where they apply, and a `step` array where each `HowToStep` has its own `name` and `text` that mirror the on-page H2 and body. Do not duplicate the prose into the schema — paraphrase tightly so the schema text is its own quotable unit. Some engines retrieve from the JSON-LD before the body.
Layer one more thing on top: a “common mistakes” or “if this fails” subsection after the steps. Those exception blocks get pulled into AI answers as caveats and add the kind of honest-trade-off texture LLMs treat as a trust signal. Bonus: long procedural posts with steps, screenshots, and gotchas naturally clear the depth bar that earns 4.3× the citations short posts do.
If you’re a brand that wants to be the answer LLMs reach for (not just rank on Google), Paris Roussos has been engineering search visibility for 30 years and now runs done-for-you AI SEO. Flat-rate, no-fuss. Email parisroussos@gmail.com.
AI search has already decided what a “how to” answer should look like — match the shape, or stay invisible while the model quotes someone who did.