# How to get your content cited by AI assistants (a GEO guide)

_2026-07-05 · agents, referrals, geo_

Search is quietly turning into answers. When Google shows an AI summary, people click a traditional result in just 8 percent of searches, against 15 percent when there's no summary ([Pew Research Center](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/), 2025). The visit you used to win with a blue link now depends on being the source the assistant quotes. Generative engine optimization, or GEO, is how you earn that spot. This guide explains what GEO is, what content assistants actually cite, and how to tell whether it's working.

> **Key takeaways**
>
> - GEO is optimizing your content to be cited by AI assistants like ChatGPT, Perplexity, and Google AI Overviews, not just ranked in a list of links.
> - Adding statistics, quotations, and citations to sources lifted visibility in generative answers by up to 40 percent in the first peer-reviewed GEO study ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), 2024).
> - It matters because answers are replacing clicks: only 8 percent of searches with an AI summary led to a link click ([Pew Research Center](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/), 2025).
> - You can't improve what you can't see. Most assistants read your pages without a browser, so only server-side measurement shows which engines cite you.

## What is generative engine optimization?

Generative engine optimization is shaping your content so AI assistants quote and cite it in their answers. In 2024, the first peer-reviewed study on this found that the right content changes can boost a source's visibility in generative responses by up to 40 percent ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), 2024). Where classic SEO competes for a rank in a list, GEO competes to be the passage an assistant lifts into its reply.

The two are related but not the same. SEO earns a position; GEO earns a citation. A page can rank well and still never get quoted, because the assistant skimmed it, found nothing quotable, and cited a clearer source instead. GEO is about being that clearer source.

> **In one line:** SEO gets you into the list of links. GEO gets you into the answer that replaced the list.

## Why does getting cited matter now?

Because the click you used to earn is disappearing into the answer. As of 2025, people clicked a search result in only 8 percent of searches that showed an AI summary, nearly half the 15 percent rate without one, and just 1 percent clicked a link inside the summary itself ([Pew Research Center](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/), 2025). When the answer is on the page, fewer people travel to find it.

This isn't one dataset's quirk. Ahrefs put the drop in organic click-through for the top result at about 58 percent when an AI Overview is present ([Ahrefs](https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/), 2026). And the reach is enormous: ChatGPT alone reached 800 million weekly active users by late 2025 ([TechCrunch](https://techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly-active-users/), 2025). The audience that once scanned ten links now reads one answer.

## Are AI referrals worth winning?

Yes, more per visit than search clicks. Visits referred by ChatGPT have converted around 15.9 percent against 1.76 percent for search organic on one B2B site ([Seer Interactive](https://www.seerinteractive.com/insights/case-study-6-learnings-about-how-traffic-from-chatgpt-converts), 2025), and in 2026 Adobe reported AI-referred shoppers convert about 42 percent better ([Adobe](https://business.adobe.com/blog/the-explosive-rise-of-generative-ai-referral-traffic), 2026). The clicks are rarer, but each one arrives pre-vetted by an assistant that already recommended you.

That's the GEO bargain in two numbers: fewer clicks overall, but the ones an assistant sends convert far above search. The citation is what starts those visits. For capturing them once they arrive, see [how to track AI referral traffic](/blog/how-to-track-ai-referral-traffic).

## What content does AI actually cite?

Specific, sourced, quotable passages. The Princeton GEO study tested nine content changes across a benchmark of diverse queries and found that adding relevant statistics, direct quotations, and citations to outside sources were the highest-impact tactics, lifting a source's visibility in generative answers by up to 40 percent ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), 2024). Keyword stuffing, the old reflex, did almost nothing.

There's a pattern under that result. An assistant assembling an answer wants passages it can drop in with attribution and defend. A sentence that pairs a concrete figure with a named source is easy to quote and hard to dispute. A vague paragraph of adjectives is neither, so it gets passed over for a cleaner source nearby.

> **The quotability test:** read any paragraph on its own, out of context. If it states a clear claim with a number and a source, an assistant can cite it. If it only makes sense after the three paragraphs above it, it won't.

## How do you make a page citable?

You write passages an assistant can lift whole, then make the page easy for it to parse. The first half is editorial, the second is technical. Here is the editorial checklist, drawn straight from what the research rewards.

1. **Answer the question in the first two sentences.** Lead each section with a direct answer, not a windup. Assistants extract the passage that resolves the query fastest.
2. **Pair every claim with a number and a named source.** "Conversion improved" is unquotable. "Conversion rose to 15.9 percent ([Seer Interactive](https://www.seerinteractive.com/), 2025)" is a citation waiting to happen.
3. **Write self-contained passages.** Each paragraph should make sense pulled out of the page, since that's exactly how an assistant uses it. Avoid "as mentioned above."
4. **Add a short FAQ.** Question-and-answer blocks map cleanly onto how people prompt assistants, so they're disproportionately quoted.
5. **Keep it current.** Date your claims and refresh them. A 2026 figure beats an undated one when the assistant picks between two sources.

According to the Princeton GEO study, these changes are not cosmetic. Sources that added statistics and citations gained up to 40 percent more visibility in generative answers than the same content without them ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), 2024). Quotable writing isn't a style preference here. It's the ranking signal.

## Can the assistant even read your page?

Often, no, and that erases every editorial gain above. Major AI crawlers, including GPTBot and ClaudeBot, fetch JavaScript files but execute none of them ([Vercel and MERJ](https://vercel.com/blog/the-rise-of-the-ai-crawler), 2024). If your best, most quotable passage is drawn on the client after load, the assistant never sees it, and you can't be cited for content that wasn't read.

So the technical half of GEO is making your content reachable without a browser. Serve it server-rendered, and offer a clean text version for the machines that want one. A markdown twin does exactly this: the same page as plain markdown when an agent asks, which cut token use by about 80 percent on one real page ([Cloudflare](https://blog.cloudflare.com/markdown-for-agents/), 2026). Cheaper to read and easier to parse means likelier to be quoted. The full how-to is in [the markdown twin guide](/blog/hello-agentic-web).

> **The order that matters:** make the page machine-readable first, then make the passages quotable. Brilliant writing the crawler can't execute is invisible, and invisible content is never cited.

## How do you know if it's working?

You measure which engines read you on the server, because nothing else can. AI crawlers were about 20 percent of verified bot traffic in 2025 ([Cloudflare Radar](https://blog.cloudflare.com/radar-2025-year-in-review/), 2025), and since they run no JavaScript, browser analytics counts almost none of them. GEO without measurement is guessing whether the assistants ever showed up.

There are two questions worth tracking, and they're different. First, who is reading you: which crawlers fetch which pages, and how often. Second, who is sending people back: the human referrals from ChatGPT, Perplexity, and the rest, which are the downstream proof a citation turned into a visit. A page that crawlers read heavily but that sends no referrals is a candidate for more quotable rewriting.

> **What we keep seeing:** a crawl is not a citation, and a citation is not a click. Treat them as three separate signals. Caprail classifies every request on the server, so you can watch crawls, grade how each page is served, and attribute the referrals separately rather than guessing from a flat "direct" bucket.

Caprail measures this side of GEO. It records each agent request before any script runs, shows which engines read each page, and attributes the human referrals those answers send back. Adding it is two lines in your middleware:

```ts
import { createCaprailMiddleware } from "@caprail-dev/analytics/next";

export const middleware = createCaprailMiddleware();
```

For the wider picture of what server-side agent analytics covers, see [analytics for AI agents](/blog/analytics-for-ai-agents) and [the best new analytics tool](/blog/the-best-new-analytics-tool).

## FAQ

### What is the difference between GEO and SEO?

SEO optimizes for a rank in a list of links. GEO optimizes to be the source an AI assistant quotes in its answer. They overlap, but a page can rank well and never be cited. The first GEO study found content changes can lift visibility in generative answers by up to 40 percent ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), 2024).

### How do I get cited by ChatGPT or Perplexity?

Write passages an assistant can lift whole: lead with the answer, pair claims with a number and a named source, and keep paragraphs self-contained. Then make the page server-readable, since most crawlers run no JavaScript ([Vercel and MERJ](https://vercel.com/blog/the-rise-of-the-ai-crawler), 2024). Quotable content the crawler can actually read is what gets cited.

### Does GEO replace SEO?

No, it extends it. Classic ranking still feeds the assistants their candidate sources, so on-page basics still matter. GEO adds a second goal on top: being quotable once you're in the running. With only 8 percent of AI-summary searches leading to a click ([Pew Research Center](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/), 2025), the citation is increasingly the prize.

### How do I measure whether GEO is working?

On the server, where every request lands. Browser analytics misses AI crawlers because they run no scripts, and those crawlers were about 20 percent of verified bot traffic in 2025 ([Cloudflare Radar](https://blog.cloudflare.com/radar-2025-year-in-review/), 2025). Server-side measurement shows which engines read each page and attributes the human referrals the answers send back.

### Is GEO only about adding statistics?

No, but data helps a lot. Adding statistics, quotations, and citations to sources were the strongest content changes in the GEO study ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), 2024). The deeper rule is quotability: write passages that stand alone with a clear claim and attribution, so an assistant can drop them into an answer and defend them.

## Conclusion

The web is shifting from links to answers, and the page that wins is the one the assistant quotes. GEO is how you become that page: lead with the answer, back every claim with a number and a named source, and make the content readable without a browser. The research is blunt about the payoff, up to 40 percent more visibility in generative responses for sources that do this ([Aggarwal et al.](https://arxiv.org/abs/2311.09735), 2024).

Then watch what happens. Make a few pages more quotable, serve them cleanly to agents, and measure which engines start reading and citing you. Caprail is free during the private beta, with no credit card required, and most sites find an agent they never knew was reading them within the first hour.

## Sources

- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande, GEO: Generative Engine Optimization, KDD 2024, retrieved 2026-06-29, https://arxiv.org/abs/2311.09735
- Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, retrieved 2026-06-29, https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
- Ahrefs, AI Overviews Reduce Clicks by 58% (update), retrieved 2026-06-29, https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/
- TechCrunch, Sam Altman says ChatGPT has hit 800M weekly active users, retrieved 2026-06-29, https://techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly-active-users/
- Vercel and MERJ, The rise of the AI crawler, retrieved 2026-06-29, https://vercel.com/blog/the-rise-of-the-ai-crawler
- Cloudflare, Introducing Markdown for Agents, retrieved 2026-06-29, https://blog.cloudflare.com/markdown-for-agents/
- Cloudflare Radar, 2025 Year in Review, retrieved 2026-06-29, https://blog.cloudflare.com/radar-2025-year-in-review/
- Seer Interactive, Case Study: 6 Learnings about how Traffic from ChatGPT Converts, retrieved 2026-06-29, https://www.seerinteractive.com/insights/case-study-6-learnings-about-how-traffic-from-chatgpt-converts
- Adobe, The explosive rise of generative AI referral traffic, retrieved 2026-06-29, https://business.adobe.com/blog/the-explosive-rise-of-generative-ai-referral-traffic