Generative Engine Optimisation has accumulated a significant amount of confident, vague advice. “Be authoritative.” “Write clearly.” “Use structured data.” None of it is wrong, exactly, but none of it is specific enough to act on. What actually determines whether your content gets cited in an AI-generated answer depends on which AI engine you’re optimising for — because they work differently.

How Google AI Overviews Cite Sources

Google AI Overviews are the most important GEO target for most sites because they appear at the top of Google SERPs and directly impact click-through rates for the queries they appear on.

The sources Google cites in AI Overviews are strongly correlated with its existing ranking signals. Pages that rank in the top three organic positions for a query are cited far more often than pages ranked lower. The implication: AI Overviews are not a separate optimisation target — they’re an extension of organic ranking. The best path to AI Overview citation is ranking organically.

That said, there are content-level factors that increase citation likelihood independently of rank position. Google AI Overviews tend to cite content that:

  • Directly answers the query with a clear, extractable statement in the first 200 words
  • Uses structured formats (definition → explanation → implication) rather than narrative prose for factual topics
  • Contains specific data points, named sources, or quantified claims that can be surfaced verbatim
  • Is demonstrably expert — bylined, published by an organisation Google recognises as authoritative in the domain

The single most reliable GEO optimisation for AI Overviews is answering the question directly and early, before contextualising. AI systems extract the most useful excerpt; if the direct answer is buried in paragraph four, it’s less likely to be used.

How Perplexity Cites Sources

Perplexity operates differently from Google. Its retrieval mechanism is closer to a research tool than a search engine — it actively crawls and synthesises from multiple sources for each query, rather than drawing from a pre-computed index.

The implications for GEO: Perplexity is more likely to cite newer content, more likely to cite content it can freshly crawl, and less dependent on domain authority than Google. A high-quality article on a specialist site can be cited alongside a Wikipedia entry or a major publication if it contains useful, specific information the query requires.

Perplexity tends to cite content that:

  • Is recent (published or updated within the last 12–18 months)
  • Contains specific, named, and verifiable claims
  • Covers the topic with enough depth that the AI can construct a synthesis from it
  • Is crawlable — pages behind login walls, heavy JavaScript renders, or bot-blocking are excluded

For Perplexity specifically, keeping content updated and ensuring clean crawlability are high-priority GEO factors that are irrelevant to Google AI Overviews.

How ChatGPT Cites Sources

ChatGPT’s citation behaviour (in its search-enabled mode) is the most variable of the three. Its web search integration retrieves pages in real-time but the citation logic is opaque. Empirically, ChatGPT with browsing enabled tends to favour:

  • Authoritative domains with high brand recognition
  • Content that matches the query phrasing closely
  • Pages with clear authorship signals and publication dates
  • Structured, scannable formats (headers, lists, tables) over dense narrative prose

The key GEO insight for ChatGPT is that it responds well to question-framing within content. Content that explicitly asks and then answers the question a user might pose (“What is X? X is…”) is more likely to be surfaced as a citation than content that assumes the reader’s context.

The Overlapping Principles

Despite their differences, there are GEO principles that apply across all three:

Answer directly. AI systems are trying to construct useful responses. Content that buries its answers in preamble is less useful as a source than content that leads with the answer.

Be specific. Vague assertions (“SEO is important for businesses”) are less citable than specific claims (“Sites in the top three positions receive 68% of clicks on a given SERP, per Backlinko’s 2024 CTR study”). Named data, specific percentages, and attributed research are citation-worthy in a way that general claims are not.

Use structural formatting. Headers, lists, and tables help AI systems extract the information they need. Dense paragraphs of continuous prose are less parseable than structured content.

Keep content current. All three systems penalise outdated information. A page with a 2022 publication date and no update signal is less likely to be cited for current topics than the same content published or updated recently.

The GEO opportunity is real. AI search answers are increasingly the first thing users see for a large class of queries. But it’s not a separate discipline from SEO — it’s an extension of producing genuinely useful, authoritative content that search systems of all kinds can extract value from.