# How AI recommendations work: scenarios → sources → why competitors win

Kyrylo Poltavets  
Jan 27, 2026  
10-12 min min read

## Key takeaways

- Track recommendations by scenario, not by brand-name prompts.
- Win sources: strengthen your site pages and your third-party coverage.
- Make your positioning simple. Add proof AI can cite.
- Publish “best for”, “vs”, and “alternatives” content with criteria and trade-offs.
- Measure weekly: visibility, share of voice, recommendation rate, and citation coverage.

## The 3 outputs AI can give you

AI can treat your brand in three ways.  
Track all three. They are not the same.

1. Mention
2. Citation
3. Recommendation

### Mention  
**Definition:** AI includes your brand name in the answer.  
**How to spot it:** your brand name appears.  
**How to measure it:** count mentions across your scenario set.  
**What “good” looks like:** you appear in the top options for your buyer scenarios.

### Citation  
**Definition:** AI uses your pages as a source for a claim.  
**How to spot it:** your domain or page is referenced.  
[**How to measure it:**](/content/blog/methodology-how-dabudai-extracts-links-from-ai-answers/index.html) track citation / source coverage across scenarios.  
**What “good” looks like:** AI uses your pages to justify key points about you.

### Recommendation  
**Definition:** AI tells the user to choose your brand for a specific need.  
**How to spot it:** phrases like “choose”, “best for”, “I recommend”, “pick X if…”.  
[**How to measure it:**](/content/blog/what-is-aeo-recommendation-rate/index.html) track recommendation rate across scenarios.  
**What “good” looks like:** you are suggested for the right ICP and constraints.

## Scenarios: why the question changes the answer

The question is not noise. It is the main input.

### Scenario = intent + constraints

A scenario is a repeatable buyer question.  
It has a fixed **role**, **intent**, and **constraints**.

### How to define a scenario (step by step)

1. Pick a buyer role (your ICP).  
2. Pick an intent: best, vs, alternatives, how-to, or definition.  
3. Add constraints: budget, region, team size, must-have features.  
4. Define success with a fixed format: top-3, criteria, trade-offs, and sources.

| **Scenario type** | **Example prompt (dentistry)** | **Output focus** | **What you measure** | **Best content to improve it** |
| --- | --- | --- | --- | --- |
| Definition | “What is a dental implant? Explain simply. Who is it for?” | Citation | Mentions + citations of your site/pages | Glossary pages (Implant, Crown, Root Canal) |
| Best option | “Best dental clinic in **[City]** for implants. Give top 3 and why.” | Recommendation | Recommendation rate + list position | “Best for” landing pages + city pages |
| Vs | “Dental implants vs bridge: which is better for a missing tooth?” | Recommendation | Which option AI recommends + reasons | Comparison pages with criteria + FAQ |
| Alternatives | “Alternatives to braces for adults. What are options?” | Mention | Inclusion in option lists | Treatment options hub + guides |
| How-to | “How does a root canal work? Steps, pain, recovery time.” | Citation | Citation coverage + step quality | Step-by-step procedure pages + recovery guides |

## How AI builds an answer (simple flow)

Most answer engines follow a similar flow.

### The 5-step flow

1. Step 1: Detect intent  
    AI decides if the user wants a definition, a comparison, or a recommendation.  
    What you can do: create pages for “best / vs / alternatives”, not only generic posts.

2. Step 2: Collect candidates  
    AI looks for information it can use.  
    What you can do: build strong first-party pages and grow consistent third-party coverage.

3. Step 3: Select what to trust  
    AI favors content that is clear, consistent, and supported by proof.  
    What you can do: publish proof pages and improve page structure on key pages.

4. Step 4: Synthesize an answer  
    AI combines the selected information into a single response.  
    What you can do: write short, direct statements and repeat the same wording across key pages.

5. Step 5: Output a pattern  
    The answer becomes a mention, a citation, or a recommendation.  
    What you can do: track all three outputs across a stable scenario set.

**A key point:** the same brand can look different across prompts.

## The source layer (what AI pulls from)

Your homepage is not enough. AI pulls from many sources.

### First-party sources (your site)

These are pages you control. They should be clear, structured, and consistent.

**Core pages (must-have):**
- Homepage (what you do + best for)
- About (who you are + credibility)
- Pricing or Plans (constraints, limits, clarity)

**Citable assets (high impact):**
- Glossary pages (definitions AI can reuse)
- Methodology pages (how you measure or do the work)
- Proof pages (case studies, results, policies)
- Comparison pages (criteria + trade-offs + best for)

### [Third-party sources](/content/blog/what-is-third-party-sources-aeo/index.html)

AI also uses what others say about you.

#### Start with these (in this order):
1. Reviews and ratings (highest trust signal)
2. Directories and listings (easy visibility wins)
3. Partner pages and industry guides (strong credibility)

### What makes a source usable (signals checklist)

| **Good signals:** |  |  |  |
| --- | --- | --- | --- |
| 1. Clear topic and clear brand/entity name |  |  |  |
| 2. Specific claims with visible proof |  |  |  |
| 3. Strong structure: headings, lists, FAQ, tables |  |  |  |
| 4. Consistent wording across pages and sources |  |  |  |
| 5. Updated signals when freshness matters (dates, “last updated”, clear ownership) |  |  |  |

**Bad signals:**
- Vague claims (“best”, “leading”) with no proof
- Long blocks with no headings
- Conflicting positioning across pages
- Outdated or unclear pages

## Why competitors win (7 common drivers) - improved, compact

This section explains **how ai decides which brand to recommend** — and why the same prompt can produce a different winner when sources and proof change.

### The 7 drivers

1. Clear category + “best for”
2. Proof pages that can be cited
3. Strong comparisons
4. Consistent third-party coverage
5. Better page structure
6. Reputation signals
7. Better scenario fit

## Competitor analysis in 60 seconds

Use this method first.

1. Pick 10 buyer questions. Use best / vs / alternatives.
2. Run them in 2–3 answer engines.
3. Write down who gets recommended for each question.
4. Write down which sources are cited or referenced.
5. Tag the winner’s main advantage: clarity / proof / coverage / structure / match.

### Video: What is Retrieval-Augmented Generation (RAG)?

RAG (Retrieval-Augmented Generation) is one of the main reasons AI answers can include specific facts and sources.

**Why this matters for AI recommendations:**
- AI can only cite and recommend what it can reliably retrieve. If your product pages, docs, and proof points aren’t easy to find and extract, you’ll lose visibility to competitors.

### Next steps (make this useful right now)

1. Run the Clarity / Proof / Match check on 10 buyer questions.
2. Set up weekly tracking with a fixed list of buyer questions.