Generative engine optimization (GEO): a practical guide

That matters because every one of these engines retrieves from a search index first, so a page they haven't crawled can't be cited at all.

This guide is part of our AI search optimization series.

What is generative engine optimization?

Generative engine optimization is the work of improving how often, and how prominently, your pages are used and cited when an AI system generates an answer. The term comes from a 2023 research paper, "GEO: Generative Engine Optimization" by Pranjal Aggarwal and co-authors from Princeton University, IIT Delhi and independent research, published at KDD 2024.

GEO sits next to two older ideas: classic SEO, which targets rankings, and answer engine optimization, which grew out of featured snippets and voice answers. The three share most of their work; for the side-by-side, read GEO vs SEO.

Google's own position is blunt. Its AI optimization guide says "optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Treat GEO as a sharper focus inside SEO, not a separate discipline.

How generative engines retrieve, choose and cite sources

Generative engines answer in two stages: they retrieve candidate pages from a search index, then a language model reads those candidates and writes an answer that cites some of them. Your page has to survive both stages.

1. Retrieval. The engine rewrites the question into one or more searches and pulls candidates from an index. Google says AI Overviews and AI Mode may use "query fan-out," which means "issuing multiple related searches across subtopics and data sources" (Google Search Central). Fan-out widens the pool beyond the top results for the literal query.
2. Selection and synthesis. The model reads the retrieved passages and decides which to use. Lab work shows models can be swayed by cues such as visible dates (Fang et al., arXiv 2509.11353), but production engines don't publish their selection rules.

Which index each engine uses differs:

For more on the crawlers behind each engine, see our guide to LLM SEO.

GEO starts with crawling and indexing

GEO starts with crawling and indexing because no engine can cite a page its index hasn't stored. Google states it directly: "To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet" (Google Search Central).

The same page says there are "no additional technical requirements," and that "indexing and serving isn't guaranteed." So the first GEO task is unglamorous:

• Confirm that robots.txt, your CDN and your firewall allow the crawlers you care about (Googlebot, Bingbot, OAI-SearchBot, PerplexityBot).
• Confirm that the pages you want cited are actually indexed, not just published.
• Get new and updated pages crawled promptly. Google says crawling "can take anywhere from a few days to a few weeks" (Google: Ask Google to recrawl), and until then the page doesn't exist for AI Overviews.

The same applies off your site. A guest post or review that mentions you can only feed an AI answer once the page it sits on has been crawled. That's why getting backlinks indexed belongs in a GEO plan, not just a link-building one. We cover it in getting backlinks indexed.

Content that gets cited: specific claims, clear sources, original information, clean structure

Content that gets cited makes a specific claim, backs it with a source, offers something other pages don't, and puts the answer where a model can lift it in one piece. The best evidence comes from the original GEO paper and a later benchmark that tempers it.

What the GEO paper found. Aggarwal et al. built GEO-bench, 10,000 queries across 25 domains, and tested nine page edits on a simulated generative engine built on GPT-3.5. Measured by their position-adjusted word count, adding quotations raised visibility by about 41%, adding statistics by about 31%, and citing sources by about 28% (arXiv 2311.09735). Keyword stuffing lowered it. The gains were largest for pages that started lower in the retrieved set, while the top-ranked source sometimes lost ground.

What it didn't test. The experiment used a fixed set of five retrieved pages per query. It shows what helps a page that has already been retrieved get used in the answer. It says nothing about getting retrieved, and none of its nine methods involved dates or freshness.

The counterweight. C-SEO Bench, a NeurIPS 2025 benchmark, found most "conversational SEO" rewrites were ineffective or harmful, and that a document's position in the model's context, which traditional search ranking drives, mattered far more (arXiv 2506.11097). Gains also shrank as more sites used the same tricks.

What that means for your pages:

• Lead each section with a direct answer that names its subject, so the passage makes sense on its own.
• Use real numbers with a source and a date. "Google says crawling can take days to weeks" beats "indexing can be slow."
• Add something original: your data, a worked example, a tested process, a clear opinion.
• Use tables for comparisons and numbered lists for processes.
• Don't chunk for the sake of it. Google says "there's no requirement to break your content into tiny pieces" (Google AI optimization guide).

Freshness signals and re-crawl speed

Freshness helps GEO mainly on topics where current information is expected, and only once the engine has crawled the updated version. Vendor studies show AI-cited pages skew toward recently updated pages, but authority and relevance still predict citation better than recency.

The evidence, read carefully:

• Seer Interactive (vendor study, July 2026; 7,683 pages and 47,097 citations on ChatGPT, Gemini and Perplexity for four brands) found 75% of cited pages had been updated within a year, but only 42% of pages with both dates readable had been published within a year (Seer Interactive). Maintained, established URLs win more than brand-new ones.
• Ahrefs (vendor study, July 2025; about 17 million citations) found AI assistants cite somewhat fresher pages than organic results, except Google AI Overviews, which cited slightly older pages (Ahrefs).
• Google's "query deserves freshness" systems are designed to show fresher content "for queries where it would be expected," such as reviews of a newly released movie or news after a recent earthquake (Google ranking systems guide).

Re-crawl speed is the part you can act on. Microsoft says "freshness signals directly influence how quickly updates are reflected in search results and AI generated answers," and recommends accurate sitemap lastmod plus IndexNow "for real-time URL submission" (Bing Webmaster Blog, July 2025). IndexNow shares each submission with participating engines such as Bing, Yandex, Seznam.cz, Naver and Yep, and through Bing it reaches Copilot. Google doesn't take part (IndexNow). We cover the Bing side in Bing and IndexNow.

The short version: AI answers cite what's been crawled, and they lean toward pages that are kept current. Make real updates, then make sure the engines have seen them. Changing a date without changing the content isn't freshness, and Google's guidance asks whether you're doing exactly that.

Brand mentions and off-site signals: backlinks, reviews, directories

Off-site signals matter for GEO because generative engines lean heavily on what other sites say about you, not just on your own pages. Brand mentions show the strongest correlation of any signal studied so far.

• Brand mentions. In an Ahrefs study of 75,000 brands (vendor study, May 2025, AI Overviews only), brand web mentions correlated with AI Overview visibility at ρ = 0.664, against 0.218 for backlinks (Ahrefs). That's correlation, not proof of cause.
• Earned media. A University of Toronto preprint (September 2025) found AI search cites earned media (independent coverage) far more than Google does: 69–92% of AI search citations across the product categories and countries studied, against 32–54% of Google's results (Chen et al., arXiv 2509.08919).
• Backlinks. Links still help search engines discover and trust your pages. In SE Ranking's ChatGPT model (vendor study, November 2025; 216,524 pages), the number of referring domains ranked as the top factor (SE Ranking).
• Reviews and directories. These are places where independent mentions of your brand live. We haven't seen a study that isolates their effect on AI citations, so treat them as part of the wider mentions picture rather than a separate lever.

Every one of those signals depends on a crawled page. A mention on a page no engine has indexed is invisible to AI answers.

A 30-day GEO plan

A 30-day GEO plan should fix eligibility in week one, improve your most important pages in weeks two and three, and set up measurement in week four. It assumes a site with existing content.

1. Days 1–3: pick 20–30 priority questions. Use real customer questions, sales calls and Search Console queries. Map each to the page that should answer it.
2. Days 4–7: check access and indexing. Review robots.txt and CDN rules for Googlebot, Bingbot, OAI-SearchBot and PerplexityBot. Check which priority pages Google has actually indexed, and fix or push the ones it hasn't.
3. Days 8–14: rewrite for citation. Add answer-first openings, sourced numbers, original examples and comparison tables to the top 10 pages.
4. Days 15–18: get the updates crawled. Update sitemaps with honest lastmod values and request recrawls. Confirm Google has seen the new versions.
5. Days 19–25: work the off-site layer. List backlinks, guest posts and mentions from the last six months, check which pages are indexed, and pursue two or three earned mentions.
6. Days 26–30: baseline your measurement. Run your question set in each engine and record who gets cited. Repeat on a schedule from here.

The GEO checklist

The GEO checklist below covers what to confirm for each page you want cited, in the order that matters.

• The page is allowed in robots.txt and isn't blocked by your CDN or firewall.
• The page is indexed by Google and Bing, and the indexed version is the current one.
• The opening sentences answer the main question directly and name the subject.
• Each section opens with a self-contained answer.
• Claims carry sources, dates and specific numbers.
• The page offers something original: data, examples or a tested method.
• Comparisons are in tables; processes are numbered steps.
• Structured data, if used, matches the visible text.
• The update date reflects a real change to the content.
• Important off-site mentions and backlinks pointing to the page are indexed too.

Measuring GEO: citations, mentions and referral traffic

Measuring GEO means tracking three things over time: whether engines cite your pages, whether they mention your brand, and whether AI assistants send you visits. One check is a snapshot, because AI answers vary by wording, location and day.

• Citations. Run a fixed set of prompts on a schedule across the engines you care about and log which URLs are cited. Citations are volatile: in a MaxAEO vendor study (July 2026; 1,240 sources across 132 brands) only 24% of sources that won a first citation were still being cited 30 days later (MaxAEO).
• Mentions. Track whether your brand is named in answers, even without a link.
• Referral traffic. Watch analytics for visits from chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. Google reports AI feature traffic inside Search Console's Performance report, "within the 'Web' search type," not as a separate report (Google Search Central).
• Eligibility. When a priority page isn't cited, check first whether it has been crawled in its current version.

That last check is where FreshlyCited fits. FreshlyCited helps Google find and index your good new and updated pages: it checks which pages and links Google has indexed, pushes only the ones it hasn't, and re-checks them on days 1, 3 and 7 so you see the before and after. Indexed pages are eligible for AI Overviews and AI Mode; whether they're cited still depends on quality, relevance and authority. For sites you verify with an IndexNow key file, the same URLs are also sent to Bing, free, and Microsoft says Copilot's web answers use the Bing search service (Microsoft Learn). ChatGPT, Perplexity and other assistants use their own crawlers and search partners. We don't submit to them, and we don't claim to. FreshlyCited doesn't write content or track AI citations, so pair it with whatever you use for prompt tracking.

If your priority pages or backlinks aren't showing up in AI answers, start by confirming Google has indexed them. When they aren't, push fresh pages to Google and measure again. Check your own links free.