Module 4: How Generative Engines Find and Use Content

Module Challenge

Learning Outcomes

The Big Idea

A generative engine does not always answer a question the same way.

Sometimes the model can respond from patterns and information learned during training. Other times the question benefits from current, specific, local, niche, or detailed information, so the system may search for or retrieve outside sources.

When retrieval happens, your website enters a new kind of competition. You are no longer competing only for a blue link. You are competing to become useful source material for an answer.

Retrieval gets a source into consideration. Synthesis determines how that source contributes to the answer.

That distinction matters. A page can be discoverable but never selected. It can be selected but not cited. It can be cited but not recommended. GEO is partly about improving the odds that your content is useful at each stage.

---

How to Use This Module

This module has two parts.

Complete Before Class 1: How AI Systems Retrieve Sources before the first class period. You will explore when AI systems appear to use retrieval and what kinds of sources they surface.

Complete Before Class 2: How AI Systems Build Answers from Sources before the second class period. You will compare sources, examine how information gets synthesized, and think about what would make your own site more useful.

As usual, do more than read. Run searches. Open citations. Compare source pages. Look for patterns. GEO becomes much easier to understand when you watch the process happen.

Your exact graded Prep Ticket instructions will be in Canvas.

Visual Framework

1. User Question

A person asks a natural-language question. The wording, context, location, timing, and follow-up conversation can all affect what information is needed.

2. Interpretation

The system tries to understand the user’s intent. A complicated question may contain several smaller information needs.

3. Retrieval

If outside information is useful, the system may search the web, an index, a database, or another knowledge source.

4. Candidate Sources

Retrieval can surface many possible pages or sources. Your page may be one candidate among many.

5. Source Selection

The system must decide which sources appear useful, relevant, credible, and sufficiently clear for the task.

6. Synthesis

The generative model combines information into a new response rather than simply returning a list of links.

7. Answer + Citations

The final answer may mention brands, summarize ideas, cite sources, link to pages, recommend options, or omit sources entirely.

Key Terms

Retrieval

Finding information from an external source at the time an answer is being produced.

Retrieval-Augmented Generation (RAG)

A general approach in which a system retrieves relevant outside information and gives it to a generative model to help produce an answer.

Grounding

Connecting a generated response to external information so the answer can be based on specific sources or current facts.

Candidate source

A page, document, database result, or other source that may be considered for use in an answer.

Source selection

The process of choosing which retrieved sources are useful enough to influence the generated response.

Synthesis

Combining information from one or more sources into a coherent generated answer.

Citation

A visible reference or link that shows where information in an AI-generated answer came from.

Query reformulation

Changing or expanding a user’s question into another search or information request that may retrieve better sources.

Freshness

How current the information is relative to the user’s question.

Retrieval advantage

An opportunity for a website to add value because the answer benefits from fresh, local, specific, niche, original, or experience-based information.

Not Every Question Needs Retrieval

Try asking an AI system:

“What is photosynthesis?”

A capable model may answer that easily without searching the live web. The concept is stable, common, and heavily represented in training data.

Now compare that with:

“Which restaurants in Rexburg are open after 10:00 p.m. tonight?”

That question depends on current, local information. A useful answer is much more likely to benefit from retrieval.

This distinction matters for your semester project. A brand-new website may have a difficult time becoming necessary for a broad question that AI can already answer extremely well from established knowledge. It may have a more realistic opportunity when the question requires information that is specific, current, local, niche, original, or changing.

Questions That Often Benefit from Retrieval

• Current: What changed this week?
• Local: What is available near me?
• Specific: Which option fits these exact circumstances?
• Comparative: How do these alternatives differ right now?
• Niche: What is true for a small or specialized audience?
• Experience-based: What happened when someone actually tested or used this?
• Original: What does this business, researcher, creator, or organization uniquely know?
• Dynamic: What are the current prices, schedules, inventory, policies, or conditions?

A Useful GEO Question

Does this question create a reason for the AI system to retrieve my website?

You do not need to finalize all of your target questions yet. We will do deeper prompt and question research later. For now, start noticing which questions create a natural need for retrieval.

---

How Retrieval Can Expand a Question

A user may ask one question, but answering it well can require several pieces of information.

For example:

“What is the best beginner fly-fishing setup for Yellowstone in July for under $300?”

A system might need information about:

• Yellowstone fishing conditions in July
• beginner rod and reel recommendations
• current equipment prices
• park regulations
• appropriate flies or tackle
• the user’s budget

The system may perform multiple searches, reformulate parts of the question, or retrieve several different source types before producing one answer.

This is another reason topical clarity matters. Your page does not have to answer every possible question on the internet. It does need to be very useful for the questions it is trying to answer.

What Makes a Page Retrieval-Friendly?

There is no magic checklist that guarantees retrieval. Still, several characteristics make a page easier to discover, interpret, and use:

• The page is publicly accessible.
• The topic is obvious from the title, H1, headings, and content.
• The page directly addresses a real question or information need.
• Important facts are stated clearly rather than buried in vague marketing language.
• The content includes specific details, examples, evidence, or data when appropriate.
• The page is connected to the rest of the website through useful internal links.
• The information is current when freshness matters.
• The site gives users and systems reasons to trust the information.

Activity 4A: Watch Retrieval Happen

Choose an AI system that can search the web and run three different kinds of questions:

1. Evergreen question: a broad question that can probably be answered from general knowledge.
2. Current or local question: a question that clearly benefits from up-to-date retrieval.
3. Niche question: a specific question related to your website topic.

For each question, observe:

• Did the system appear to search or retrieve outside sources?
• What kinds of sources appeared?
• Were the sources large established sites, smaller niche sites, local sites, forums, videos, primary sources, or something else?
• Were citations visible?
• Did the answer rely heavily on one source or combine several?
• What surprised you?

Inspect the Sources

Do not stop at the AI answer. Open at least two cited or referenced pages.

For each source, ask:

• What specific part of this page probably made it useful?
• How quickly can you tell what the page is about?
• Does it provide specific facts, evidence, experience, or examples?
• How is the page structured?
• Why might this page have been chosen instead of another page?

Prepare for Class

Complete Prep Ticket 4A in Canvas. Come prepared to show one example where retrieval clearly mattered and one source that you think was especially well suited to the answer.

Retrieval Is Not the Finish Line

Suppose an AI system retrieves ten pages. It probably will not use all ten equally.

Some sources may provide one fact. Another may provide the main explanation. Another may influence the answer without receiving a visible citation. Some may be ignored completely.

So there are really two competitions:

1. Can my page get retrieved?
2. If retrieved, is my page useful enough to influence the answer?

From Sources to a New Answer

Generative search is different from a traditional list of results because the system can combine information from multiple places into one response.

That creates both an opportunity and a challenge for marketers.

The opportunity is that a smaller site does not always need to become the single top result to contribute useful information. The challenge is that the AI system may extract only the part it needs, combine it with other sources, and present a synthesized answer.

What Kind of Content Is Easy to Use?

Imagine that an AI system retrieves two pages about the same topic.

Page A begins with three paragraphs of broad promotional language before finally giving the answer.

Page B states the answer clearly, provides supporting evidence, explains exceptions, and organizes the details under descriptive headings.

Which page would you rather use if you had to build a reliable answer quickly?

AI systems face a similar information problem. Clear, specific, well-supported content is simply easier to work with.

Useful Source Characteristics

• Directness: The page actually answers the question.
• Specificity: It contains details rather than generic claims.
• Structure: Headings and sections make important information easy to locate.
• Evidence: Claims are supported when support is needed.
• Context: The page explains when, where, for whom, or under what conditions an answer applies.
• Freshness: Time-sensitive information is current.
• Originality: The page contributes something distinctive rather than repeating generic information.
• Trust: The source gives readers reasons to believe the information.

Citation Is an Outcome, Not a Guarantee

A website owner naturally wants a visible citation or link. But retrieval does not guarantee citation, and citation does not necessarily mean recommendation.

For this course, we will keep several AI visibility outcomes separate:

• Mention: your brand or content is named.
• Citation: your page is referenced or linked.
• Accurate description: the system represents your brand, product, or information correctly.
• Recommendation: the system actively suggests you as an option.
• Influence: your information helps shape the answer, even when the user may not see an obvious citation.

Later in the course we will measure these outcomes more systematically.

---

Activity 4B: Reverse-Engineer an AI Answer

Choose one question closely related to your website topic. Use an AI system that provides web sources.

Then work backward from the answer.

1. Read the generated answer carefully.
2. Identify the major claims or recommendations in the answer.
3. Open at least three sources.
4. Find where each source overlaps with the generated response.
5. Notice which source contributed facts, explanations, examples, comparisons, or recommendations.
6. Identify anything in the answer that does not appear to be clearly supported by the visible sources.

Compare the Source Pages

Create a simple comparison for the three sources:

• What question is this page especially good at answering?
• What information is easy to extract?
• What makes the page credible?
• What makes the page distinctive?
• What could the page do better?

Now Look at Your Own Website

Choose one important page on your own site and ask:

“If this page were retrieved alongside three strong competitors, why would an AI system use information from my page?”

If you cannot answer that yet, that is okay. You have just discovered an important GEO opportunity.

Write down one improvement that would make the page more useful as source material. For example:

• give a more direct answer
• add specific local information
• add first-hand experience
• add a useful comparison
• provide original examples
• add evidence or data
• make headings more descriptive
• clarify who the information applies to
• update outdated information

A Note About What We Cannot See

Be careful about claiming that we know exactly why an AI system selected a particular source. Commercial search and generative systems are complex, change frequently, and do not expose every part of their retrieval or ranking logic.

In this course, we will separate what we can observe from what we infer. We can observe which sources appear, which pages are cited, and how answers change. We should be more cautious when explaining the hidden reasons behind those outcomes.

Prepare for Class

Complete Prep Ticket 4B in Canvas. Be ready to show one AI-generated answer, explain how at least two sources contributed to it, and identify one change you want to make to your own site because of what you observed.

Save to Your GEO Portfolio

Module Summary

Optional Further Reading

For students who want to go deeper, these current resources explain retrieval, grounding, and AI search in more detail:

• Google Cloud: What is Retrieval-Augmented Generation (RAG)?
• Google Cloud: Grounding and RAG
• OpenAI: Searching the web with ChatGPT
• Perplexity: What is Pro Search?