AI SEO Optimization: Getting Cited in AI-Generated Answers

Created by Pranjal Aggarwal et al. - https://arxiv.org/abs/2311.09735

Overview

AI SEO optimization is the practice of making your pages easy for AI answer systems to find, understand, trust and cite. The systems in scope are Google's AI Overviews and AI Mode, ChatGPT search, Perplexity and Microsoft Copilot. Each one retrieves pages from a search index and writes a response grounded in what it retrieved, the pattern the research literature calls retrieval-augmented generation. A page succeeds in this setting when it is retrieved, when a claim from it makes it into the generated answer, and when the answer credits it with a link.

The research framing comes from the paper GEO: Generative Engine Optimization by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, a team that includes researchers from Princeton and IIT Delhi alongside independent authors. It was first posted to arXiv in November 2023 and published at KDD 2024. The authors named the field, built a benchmark of diverse user queries called GEO-bench, and tested which edits to a source page change how much of it a generative engine uses. Their headline result is that these edits can boost visibility by up to 40% in generative engine responses, and that what works varies by domain.

The detail of that result shapes this method. Across the nine edits the authors tested, adding quotations from relevant sources, adding statistics and citing credible sources helped most, while keyword stuffing offered little to no improvement. The best edits improved a position-adjusted word count metric by 41% over baseline. Lower-ranked pages gained the most: citing sources raised visibility by 115.1% for pages ranked fifth in the search results, while top-ranked pages saw decreases. Evidence and specificity help. Repeating the query does not.

The platforms have since published their own guidance, and it does not all point the same way. Google says there are no additional requirements and no special schema.org markup needed to appear in AI Overviews or AI Mode. Its generative AI guide treats AEO and GEO as ordinary SEO and lists tactics to ignore, including llms.txt files, "chunking" content and chasing inauthentic mentions (Google's AI optimization guide). Microsoft puts more weight on structure: its guidance says Copilot parses a page into smaller pieces and favors concise, self-contained answers, clear headings, lists and tables (Microsoft Advertising). This method keeps the overlap: write for readers first, in a form that also survives extraction.

The traffic stakes are measurable. In a Pew Research Center study of Google searches, users who saw an AI summary clicked a traditional result in 8% of visits, against 15% without one. Only 1% of visits included a click on a link inside the summary itself. Google, for its part, says clicks from results pages with AI Overviews are higher quality, with visitors spending more time on the site. Both points lead to the same practice: measure citations and conversions alongside clicks.

Practitioners use three overlapping labels. Answer engine optimization (AEO) usually means shaping content to be selected as the answer, generative engine optimization (GEO) means improving visibility inside synthesized responses, and LLM optimization means shaping how models describe a brand. This method treats them as one practice with seven skills: auditing, query research, content structure, structured data, topical authority, citation-worthiness and measurement.

Everything here sits on top of SEO fundamentals. A page must be indexed and eligible for a snippet to appear as a supporting link in Google's AI features, and a site that blocks OAI-SearchBot will not be shown in ChatGPT search answers. The method does not replace crawlability, internal linking or helpful content. It adds a way to check whether AI systems use that work, and a set of changes to make when they do not.

Core Principles

Retrieval comes before persuasion

An AI answer can only cite a page its system retrieved. Google limits supporting links to pages that are indexed and snippet-eligible, and OpenAI and Perplexity each run a separate search crawler (OAI-SearchBot, PerplexityBot) that a robots.txt rule can shut out. Check access before rewriting anything. A blocked crawler makes every other principle irrelevant.

Give the answer something only you have

Generative engines draw on many sources at once, so restating common knowledge gives them no reason to pick your page. Google's guide asks for non-commodity content, such as first-hand experience and original analysis, over summaries of what already exists. The GEO results point the same way: statistics and quotations were among the most effective additions. Original numbers, named expert views and worked cases are what an answer engine can use and credit.

Evidence travels with the claim

A claim that carries its source is easier to reuse than one that asks the reader to trust the page. Citing credible sources was one of the strongest edits in the GEO experiments, and Bing's guidance lists supporting claims with examples, data and cited sources as a way to build trust when content is reused in AI answers. Put the source in the sentence that states the fact.

Write passages that stand alone

Retrieval systems often lift a sentence or a passage away from the rest of the page. A heading that names the question, followed by a direct answer that makes sense on its own, serves both the reader who skims and the system that extracts. Microsoft recommends exactly this pattern for snippet selection. Google adds that there is no need to split content into tiny pieces, so structure for clarity and stop there.

Structured data describes the page

Schema.org markup gives search systems explicit clues about what a page is, who wrote it and which organization stands behind it. Google says it is not required for its generative AI features and that no special markup exists for them. Rich result support also shrinks over time: HowTo rich results were deprecated in 2023 and FAQ rich results stopped appearing in May 2026. Use markup that matches visible content and helps disambiguate your entities, and expect no ranking boost from it.

Authority is earned off your own site

AI answers lean on third-party sources. A comparative study of AI search engines found a systematic bias toward earned media over brand-owned content. That makes independent coverage, reviews and expert references part of AI SEO work. Google warns that seeking inauthentic mentions is not as helpful as it seems. Coverage has to be earned through work that others choose to cite.

Measure with first-party data first

Visibility in AI answers changes as models, retrieval and demand change. Google Search Console now has a Generative AI performance report for AI Overviews and AI Mode impressions, and Bing Webmaster Tools reports citations and grounding queries for Copilot. Third-party trackers add coverage of ChatGPT and Perplexity, but Google notes that no third-party tool has access to its internal ranking data. Start from the platform reports and treat sampled prompt tracking as a directional signal.

Research and Platform Guidance Compared

The tactics in circulation come from different kinds of evidence. This table sorts the main ones by who says what, so a team can tell a tested finding from a platform statement from folklore.

Steps

1. Audit your current AI visibility Write down the questions a buyer would ask an AI assistant about your category, your brand and your competitors. Run them in the systems your audience uses and record, for each answer, whether you are mentioned, whether you are cited with a link, which sources are cited instead and whether the facts about you are right. Pull the first-party numbers too: the Search Console Generative AI performance report and the Bing AI Performance report. The output is a baseline sheet of questions, answers, citations and errors. Every later step is judged against it.

2. Research conversational queries People ask AI systems longer questions than they type into a search box; Google reports that AI Mode queries are twice as long as traditional searches on average. Collect those questions from sales calls, support tickets, community threads and the grounding queries Bing now reports. Group them by the job the asker is trying to do, and note which follow-up questions tend to come next. Google warns that building separate pages for every variant or fan-out query to manipulate answers violates its scaled content policy, so the output is a question map for a small set of strong pages.

3. Structure pages so answers stand alone For each page, name the primary question in the title and H1 and answer it directly near the top. Use headings that say what the section answers, then open each section with a sentence that would still make sense if quoted alone. Put comparisons in tables and procedures in numbered lists, because Microsoft's guidance says these formats can be lifted cleanly into answers. Keep the page organized for a human reader, since Google says there is no ideal page length and no need to chunk content.

4. Add structured data that matches the page Mark up the entities a page is about: Organization on the home page, Article with author details on editorial pages, Product where you sell things. Google's guidelines require that markup represent the main content and describe only what users can see, and they recommend JSON-LD. Use the sameAs property to point to profiles that unambiguously identify your organization. Validate before release, and do not expect markup alone to earn a citation.

5. Build topical depth around your core subjects Pick the few subjects where you have real expertise and cover each one completely: a hub page for the topic and focused pages for the questions buyers actually ask. Link the pages to each other with descriptive anchor text that says what the target page covers. Cut or merge thin pages rather than adding more; Google's scaled content abuse policy covers many pages generated without adding value. Depth shows in the questions you can answer that competitors cannot.

6. Make pages worth citing Add the material an answer engine can credit: original data with its method, named expert commentary, clear definitions and sources for every factual claim. Put a byline on each article that leads to information about the author, one of the trust signals in Google's people-first content guidance. Check crawler access for each AI search system you care about, keeping OAI-SearchBot separate from GPTBot, which governs training use. Then pursue independent coverage, since AI search engines lean toward earned media.

7. Track visibility and share of citations Rerun the baseline question set on a fixed schedule with the same wording, systems and settings, and log mentions, citations, cited URLs and accuracy. Add Bing's citation share for Copilot and the Search Console generative AI impressions for Google. Report changes against the baseline, broken out by topic and by page. Tie the numbers to outcomes such as signups or demo requests, because Google itself suggests looking past clicks to conversions.

8. Iterate on what the data shows Compare pages that gained citations with pages that did not, and look for concrete differences: fresher data, a clearer opening answer, a named source, a better internal link. Change one thing at a time on a group of pages so you can attribute the result. Re-audit after major model or product changes, because answers shift without any change on your side. Retire tactics that show no effect in your own data, even when they are popular.

When to Use

• Your Search Console data shows impressions holding while clicks fall on informational queries, a pattern consistent with Pew's finding that users click less when an AI summary appears.
• Buyers in your market research by asking assistants comparison and how-to questions, and your sales team hears competitor names that AI tools recommended.
• AI assistants describe your company or product incorrectly, for example with an old price or a discontinued feature, and you need a repeatable way to find and fix the sources.
• You publish expert content in a knowledge-heavy field and have original data or experience that commodity content cannot match.
• You are planning a content program and want each page to serve classic search and AI answers from the start.

When Not to Use

• Your pages are not crawlable or indexed yet. Fix technical SEO first, because Google's AI features only link to pages that meet the normal Search technical requirements.
• Your content sits behind a login or paywall that crawlers cannot reach, and you have decided to keep it there.
• Demand in your category is almost entirely transactional or local, where product feeds, business profiles and ads matter more than answer visibility.
• You want a shortcut such as mass-producing pages for query variants or buying mentions. Those tactics run into Google's spam policies and add little value for readers.