# What Is RAG? 7 Conditions for Being Chosen as a Source in AI Search

Author: Fatih Emin Çakıroğlu (https://fatihemincakiroglu.com)
Source: https://fatihemincakiroglu.com/en/rag-nedir
Published: 2026-10-16
Updated: 2026-10-16

RAG (retrieval-augmented generation) is when a language model finds relevant documents from outside sources before writing an answer, and then bases the answer on those documents. The model does not rely only on its training data; it reads the pages it finds at that moment and grounds its answer in them.

The sites shown as sources in AI powered search are the pages picked during this retrieval step. In this article I explain RAG without drowning it in technical detail, then list 7 conditions your content should meet to be picked.

> Short answer: RAG is a method where an AI model searches for relevant documents before generating an answer and grounds its answer in them. To be chosen as a source, your page needs to be crawlable and indexed, keep key information in text, and contain clear passages that answer a question on their own.

## How does RAG work?

Details vary by system, but the basic flow looks like this. Companies do not publish exactly which infrastructure each assistant uses, so read the diagram as a general model.

**The general RAG flow**
1. **Question**: The user asks a question or runs a search.
2. **Query generation**: The system turns the question into one or more search queries.
3. **Retrieval**: Relevant pages and passages are found in an index or on the web.
4. **Selection**: The most relevant and trustworthy looking passages are passed to the model.
5. **Generation and citation**: The model writes the answer from those passages and may show sources as links.

Google's AI features documentation says AI Overviews and AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics and data sources to build a response. So a single question can become many searches behind the scenes. For a short definition, see the [AI glossary](https://fatihemincakiroglu.com/en/ai-glossary).

## A plain language model vs RAG

| Feature | Training data only | RAG |
| --- | --- | --- |
| Source of information | The data the model was trained on | Documents retrieved at that moment |
| Freshness | Limited by the training date | As fresh as the retrieved page |
| Citations | Usually none | Can link to sources |
| Your site's influence | Indirect, if it was in the training data | Direct, if your page is retrieved |

The last row is what matters for SEO. In a system that uses RAG, whether your page can be found and read at that moment decides whether you appear in the answer.

## 7 conditions for being chosen as a source in AI search

### 1. Your page must be crawlable

To enter retrieval, a page first has to be reachable. Google states that crawling for Search is controlled with robots.txt rules for Googlebot. Other assistants have their own crawlers; check their names and rules in each company's official documentation. Test your rules with the [robots.txt tester](https://fatihemincakiroglu.com/en/robots-txt-tester) and read the logic in the [robots.txt article](https://fatihemincakiroglu.com/en/robots-txt).

### 2. It must be indexed and eligible for a snippet

According to Google's documentation, a page must be indexed and eligible to be shown in Search with a snippet to appear as a supporting link in AI Overviews or AI Mode. There are no additional technical requirements. If your page is not indexed, fix the [crawled but not indexed problem](https://fatihemincakiroglu.com/en/tarandi-dizine-eklenmemis) first.

### 3. Key information must be in text

If a price table lives inside an image, the FAQ sits in a PDF, or content only loads through JavaScript after a click, a retrieval system may never reach it. Google also recommends keeping important content in text form. I cover what to watch for with JavaScript content in the [JavaScript SEO article](https://fatihemincakiroglu.com/en/javascript-seo).

### 4. Every section must make sense on its own

RAG systems usually evaluate parts of a page (passages), not the whole page. A paragraph starting with "as mentioned above" loses its meaning when read alone. The first sentence of each section should answer that heading's question directly.

**Easy to select passage**
- Clear answer in the first sentence
- Names the topic explicitly
- Contains a concrete number, threshold or step
- Focuses on one question

**Hard to select passage**
- Answer buried at the end
- References like "this" or "that" to earlier sections
- Vague, decorative wording
- Mixes several topics

### 5. Headings should mirror the question

h2 and h3 headings that resemble the user's question help the retrieval system match the passage to the right topic and help readers find what they need. Use clear headings like "How many URLs can an XML sitemap contain?" instead of "General information".

### 6. Information must be verifiable and current

Author details, an update date, thresholds backed by official sources and consistent figures make a page look trustworthy. A page with conflicting or outdated facts may be skipped when a clearer competitor exists. The checklist in the [content for E-E-A-T article](https://fatihemincakiroglu.com/en/eeat-icin-icerik) helps here.

### 7. It should cover the sub questions too

Given how query fan-out works, a page and a site that also answer a topic's sub questions can match several of the underlying searches. An article on boiler servicing should not stop at a definition; it should cover how often, how long it takes and which symptoms call for a service. I explain topic clusters in the [topical authority article](https://fatihemincakiroglu.com/en/topical-authority).

> **No special file or markup needed** According to Google's documentation, AI Overviews and AI Mode need no special schema.org markup and no AI specific text file. Other platforms may differ; I discuss the llms.txt debate in the [llms.txt guide](https://fatihemincakiroglu.com/en/llms-txt-rehberi).

## Checklist for making your content RAG ready

- [ ] Are key pages open to crawling in robots.txt and CDN settings?
- [ ] Does Search Console show the page as indexed?
- [ ] Does the page have nosnippet or a very short max-snippet value?
- [ ] Is the core information in HTML text, not in an image, PDF or hidden tab?
- [ ] Does the first sentence of each h2 section answer the question directly?
- [ ] Are the author and update date visible?
- [ ] Are the most common sub questions answered on the page or on linked pages?

## What changes for AI Overviews, ChatGPT and Perplexity?

The core conditions are shared, but each platform has its own crawler, index and citation style. I look at the Google side in the [Google AI Overviews optimisation article](https://fatihemincakiroglu.com/en/google-ai-overview-optimizasyonu) and at Perplexity's citation behaviour in the [Perplexity citation article](https://fatihemincakiroglu.com/en/perplexity-kaynak-gosterilme).

Google reports traffic from sites appearing in AI Overviews and AI Mode in the Search Console Performance report under the "Web" search type, together with overall traffic. To separate visits from other assistants in GA4, see the [GA4 AI traffic article](https://fatihemincakiroglu.com/en/ga4-yapay-zeka-trafigi). For the full implementation, start with the [GEO implementation guide](https://fatihemincakiroglu.com/en/geo-uygulama) and, for the technical foundation, the [technical SEO guide](https://fatihemincakiroglu.com/en/teknik-seo).

## Help with RAG friendly content

Understanding why your site is not cited in AI answers means looking at crawling, indexing and passage structure together. I run this analysis as part of [GEO consulting](https://fatihemincakiroglu.com/en/geo-consulting); [book a free intro call](https://fatihemincakiroglu.com/en/book-a-call) to get started.

## Frequently Asked Questions

### What is the difference between RAG and fine-tuning?

Fine-tuning retrains the model itself on extra data, so the knowledge is built into the model. RAG leaves the model unchanged and retrieves outside documents for each question, grounding the answer in them. That is why RAG works more easily with current information.

### Can I opt out of AI Overviews without blocking Googlebot?

According to Google's documentation, you can limit what is shown in Search with nosnippet, data-nosnippet, max-snippet or noindex. These controls also affect snippets in classic results, so be clear about what you are willing to give up on each page first.

### Does blocking Google-Extended affect AI Overviews?

Google's documentation says the Google-Extended control limits AI training and grounding in other Google systems, while crawling for Search is managed with Googlebot rules. Check the current version of the documentation before deciding.

### How do I track whether I am being cited?

List your target questions, ask them at regular intervals in AI Overviews, ChatGPT and Perplexity, and record the sources shown. Answers can vary by person and time, so decide based on trends rather than a single test.

### Are short or long pages chosen more often?

Length is not the deciding factor; how clearly a passage answers the question is. A long guide with good sections can be a source for many different questions, while a messy short page may fit none.

---
Please attribute as: "Fatih Emin Çakıroğlu (fatihemincakiroglu.com)". Original: https://fatihemincakiroglu.com/en/rag-nedir
