When someone asks ChatGPT, Perplexity, or Google's AI Overviews a question, the model writes one answer and credits a handful of sources. Generative Engine Optimisation (GEO) is the work of being one of those sources. The mechanics are different from chasing a blue link, and the pages that get cited tend to share a few specific traits.

Here are seven tactics that move you from "indexed somewhere" to "quoted in the answer."

1. Answer the question in the first two sentences

How to do it: Put the direct answer at the top of the section, before any background. If the heading is "How long does it take to ferment sourdough?", the next line should say "Most sourdough needs 4 to 6 hours of bulk fermentation at 24°C, then 12 to 16 hours of cold proofing." Then explain the variables underneath.

Why it works: Language models read your page and try to extract a self-contained claim they can drop into an answer. If the claim is buried under three paragraphs of throat-clearing, the model either skips it or summarises it badly. A clean answer near the top is easy to lift verbatim, and verbatim is what gets attributed. This is the single highest-leverage change most pages need, and it costs nothing but discipline.

2. Write headings as the questions people actually ask

How to do it: Convert vague section labels into real questions. "Pricing" becomes "How much does it cost per user per month?" "Setup" becomes "How do I connect it to my existing CRM?" Match the phrasing to how a person would type or speak the query, not internal jargon.

Why it works: Retrieval systems match user prompts against the text on your page. A heading that mirrors the prompt is a strong relevance signal, and the paragraph beneath it becomes an obvious candidate to quote. Question headings also force you to answer one thing per section, which keeps each block extractable. A page of nouns ("Overview", "Features", "Benefits") gives the model nothing to grab.

3. Publish original statistics nobody else has

How to do it: Run a small survey, pull a number from your own product data, or analyse a dataset you can legally share. You do not need 5,000 respondents. "Across 312 onboarding sessions, 41% of users connected a second integration within the first week" is a citable fact because it exists nowhere else. State the sample size and the method in one line so the number is defensible.

Why it works: AI answers love numbers, and they prefer numbers with a clear origin. When you own a statistic, every model that wants to use it has to point at you. Recycled industry stats ("90% of marketers say content is important") are already attributed to whoever published them first, so repeating them sends the citation upstream. Original data is the closest thing GEO has to a moat.

4. Cite primary sources, and link them

How to do it: When you reference a study, regulation, or spec, link the original document, not a blog post that summarised it. Name the source in the sentence: "according to the 2025 ONS labour survey" reads better to a model than "studies show." Quote the exact figure and the date.

Why it works: Models weigh the trustworthiness of a page partly by what it connects to. A page that links a government dataset, a peer-reviewed paper, or an official changelog looks more reliable than one floating claims without backing. It also makes your page useful as a hub: the model can follow your citations, confirm the facts, and then credit you for assembling them clearly. Sloppy sourcing is one of the fastest ways to get filtered out.

5. Add an FAQ section with structured data

How to do it: End substantial articles with three to six genuine questions and tight answers, then mark them up with FAQPage schema in JSON-LD. Keep each answer to two or three sentences. Do not stuff the FAQ with keywords; write the questions a reader would actually ask after finishing the piece.

Why it works: FAQ blocks are pre-chunked question-and-answer pairs, which is exactly the shape retrieval systems want. The schema tells crawlers precisely where the question ends and the answer begins, removing ambiguity about what to extract. Even where rich results have been scaled back in traditional search, the structured markup still helps machines parse intent. You are doing the chunking for them, and well-chunked content gets pulled more often.

6. Keep facts current, and show the date

How to do it: Put a visible "Last updated" date on pages that contain time-sensitive facts: prices, version numbers, market figures, anything that drifts. When you revise, change the actual content and the date together. Reflect recency in the prose too: "as of June 2026, the free tier includes 5 articles."

Why it works: Many AI systems favour fresh information for anything that changes, and they read dates to judge it. A page dated three years ago with a 2023 price tag signals staleness, and the model will reach for a newer competitor even if your underlying content is better. Showing recency in both metadata and the text gives two confirming signals. Stale facts do not just lose citations; they get your page actively distrusted.

7. Build entity and topic authority, not one-off posts

How to do it: Cover a subject in depth across several connected pages rather than one thin article. Use consistent naming for products, people, and concepts so machines can resolve them to a single entity. Keep a populated About page, name your authors with real credentials, and interlink related pieces so the cluster reads as a body of work.

Why it works: Models build an internal sense of which domains are authoritative on which topics. Ten coherent, cross-linked pages on a narrow subject teach the system that you are a reliable source for it, so you get cited even on queries you did not target word for word. Consistent entity naming helps the model connect your mentions across the web into one recognisable author or brand. Authority compounds: each solid page makes the next one easier to cite.

A practical takeaway

You do not need all seven at once. Pick your three highest-traffic pages, move the direct answer to the top, rewrite the headings as questions, and add one original statistic to each. That alone changes how often a model can use you. The rest (primary sources, FAQ schema, fresh dates, topic depth) is steady maintenance you layer on over the following weeks.

If keeping a library of pages answer-first, well-sourced, and current sounds like more upkeep than you have time for, that is the kind of thing Austen handles in the background while scoring each draft for GEO before you publish. Start with the parts you can do by hand today; the citations follow the clarity.

7 Ways to Get Your Content Cited by AI Search in 2026