GEO 7 min read

How to get cited by AI answers: a practical GEO checklist

Assistants don't rank pages, they pick sources. That one difference changes how you should write, structure, and publish. Here's the checklist we run for every site.

When someone asks ChatGPT, Perplexity, or Google AI a question in your market, the answer arrives already written — with two or three citations attached. There is no page two. Either you’re one of those citations, or you don’t exist for that customer.

Generative engine optimization (GEO) is the work of becoming that citation. It overlaps with SEO, but it isn’t the same job: a search engine ranks documents and lets the reader choose, while an assistant reads a handful of sources and composes one answer. To be chosen, your page has to be the easiest thing in the corpus to quote accurately. That’s the whole game, and the checklist below is how you play it.

1. Answer one question per page, in the first 150 words

Assistants quote passages, not sites. A page that opens with three paragraphs of scene-setting gives the model nothing to lift. Put the direct answer at the top, in plain declarative sentences, then spend the rest of the page earning trust in it.

Quick testPaste your page's first screen into any assistant and ask "what question does this answer?" If the model hesitates, so will every engine that reads it.

2. Write quotable sentences

A quotable sentence is self-contained: it names the subject, makes one claim, and includes the number or qualifier that makes the claim useful. Compare:

  • Vague: “Our classes are great for beginners and very affordable.”
  • Quotable: “Studio Ton’s beginner course runs Tuesday evenings in Zürich Kreis 4, costs CHF 320 for six sessions, and includes wheel time.”

The second sentence survives being lifted out of context. That’s what a citation is — your sentence, alone, in someone else’s answer.

3. Ship structure the machines can read

Models lean on structure to decide what a page is. The basics are boring and they work:

  • One h1 that states the question or claim; h2s that mirror the sub-questions people actually ask.
  • Schema.org markup for what the page is — FAQPage, Product, HowTo, LocalBusiness. It’s the difference between the engine inferring and knowing.
  • Real dates, visible and in metadata. Assistants strongly prefer sources they can date; “updated July 2026” is a strong signal in a world of stale corpora.
  • Tables for anything comparative. A pricing table gets quoted; a pricing paragraph gets skimmed.

4. Be present where the models look things up

Engines corroborate. Before citing you as “the beginner pottery class in Zürich,” the model checks whether anyone else says so. That means mentions matter again — not as link juice, but as evidence:

  • Directory and review listings with consistent name, address, and description.
  • A handful of third-party mentions that repeat your key claim in their own words.
  • Consistent phrasing of what you are across your own pages. If your homepage, about page, and footer each describe you differently, the model averages you into vagueness.

5. Keep score, or you’re guessing

GEO without measurement is astrology. Track a fixed panel of questions — the ones a real customer would ask — and run them weekly against each engine. What you want on the board:

answer panel · "zurich pottery" · weekly run

ChatGPT         cited  4/10  (+1 this week)
Perplexity      cited  7/10  (—)
Google AI       cited  3/10  (+2)
Claude          cited  5/10  (+1)

Citation share moves slowly, then suddenly — a page rewrite or a new third-party mention often shows up across engines within a couple of weeks. Without the panel you’ll never connect cause to effect.

The checklist, in one place

  • Every important page answers exactly one question, in its first 150 words.
  • Key claims are written as self-contained, liftable sentences with numbers in them.
  • Headings mirror real questions; schema markup declares what each page is.
  • Dates are visible and current on anything you want cited.
  • Comparative content lives in tables, not paragraphs.
  • Your one-line description of yourself is identical everywhere it appears.
  • A fixed question panel runs weekly against every engine, and someone reads the scores.

None of this is exotic. It’s the discipline of being easy to quote, applied consistently — which is exactly the kind of work that’s tedious for a person and trivial to keep running automatically.

Spellsuite runs this checklist for you.

Every page, every engine, scored weekly. You approve the fixes.

Start free
Written by the Spellsuite team · The Spellbook All posts