How AI chooses which websites to cite: what Google, Bing, OpenAI and Perplexity say

Step one: being found

Before an AI can cite a page, its search system must be able to reach and index it. Each platform describes this in its own documentation:

If robots.txt, a firewall or a meta directive blocks these systems, the page drops out before content quality even matters. We cover the crawlers in detail in Which AI crawlers to allow in robots.txt.

Step two: being used in the answer

Here the platforms are less specific, but their published guidance points in the same direction:

• Bing: URLs are more likely to be selected for grounding and citations when content stands on its own — facts and definitions are explicit, key statements do not rely on implied content, and important information is visible on the URL itself. Content should fully satisfy user intent and be understandable without external context.
• Bing: entities should be defined clearly and consistently; images and video should support the text, not replace it.
• Google: important content should be available in text, pages should be crawlable and offer a good page experience, and structured data must match the visible text.
• OpenAI: any public website can appear in ChatGPT search if it does not block OAI-SearchBot.

What the GEO research found

The paper that introduced generative engine optimization (Aggarwal et al., arXiv 2023, KDD 2024) tested which content changes increase a source’s visibility in AI answers. Adding quotations, statistics and citations of sources worked best — up to 40% improvement on the benchmark and up to 37% on one metric on Perplexity — while keyword stuffing gave little to no improvement.

What reduces the chance of being cited

• Blocked crawlers — in robots.txt, a firewall or a CDN bot setting.
• Directives — Bing says NOARCHIVE prevents content from being used in Copilot responses; nosnippet and noindex limit Google’s use of the page.
• Thin or ad-heavy pages — Bing says they may lose grounding visibility.
• Manipulation — Bing treats prompt injection and other attempts to manipulate language models as abuse.
• Contradictory facts — in our practice, different prices or addresses on different sites lead AI to repeat errors or skip the company.

Short checklist

1. Search and AI search crawlers are allowed in robots.txt and in the firewall.
2. The site is indexed in Google and Bing.
3. Each important question has a page or section that answers it in the first sentences.
4. Facts are explicit: prices, addresses, timeframes, numbers, with sources where possible.
5. Company data is the same on the website, in Google Business Profile and in directories.