How AI Models Decide Which Agency to Recommend

We track every time an AI model recommends a marketing agency in Montreal. After monitoring seven AI models simultaneously ChatGPT, Perplexity, Grok, Gemini, DeepSeek, Claude, and others we can tell you exactly how these systems decide who gets cited. The answer is less mysterious than most agencies think, and more fixable than most realize.

Here’s what we found, using our own numbers.

We’re Rank 8. Here’s What the Gap Taught Us

Let’s start with the uncomfortable truth. In our competitive set of Montreal agencies, Agence Minimal has 103 AI mentions. Lg2 has 703. Sid Lee has 594. Cossette has 522. Digitad has 396.

That gap isn’t about size or client roster alone. It’s about how AI models actually source their answers.

When a user asks ChatGPT “what are the best marketing agencies in Montreal?”, the model doesn’t recall static training data and output a memorized list. It searches. Actively. It runs web queries like a research assistant terms such as “top branding agencies Montreal Quebec 2026”, “best digital marketing agencies Montreal Canada”, and “Montreal branding agencies competitors Lg2 Sid Lee”.

We watch these exact queries happen in real time, fired by 10 different AI models every few hours, every day. What those queries find is what determines who gets recommended.

The 3 Types of Content AI Consistently Cites

After analyzing which URLs get cited in our competitive space, a clear pattern emerged across every AI model we track.

1. Roundup Articles with Structured Lists

The single most-cited URL in our niche is a page from lafusee.net titled “Top 5 agences marketing web pour les PME à Montréal.” It accumulates 373 citations. A page from knowltonquebec.ca “Meilleures agences web et marketing à Montréal en 2026” gets 250. Digitad’s “13 agences marketing incontournables à Montréal” gets 171.

Every one of these is a listicle. Numbered, structured, scannable. AI models cite these because they are citation-ready by design. Each entry is self-contained: agency name, specialty, a sentence or two of description. The model can extract a structured answer directly without interpretation.

2. Agency Homepages with Explicit, Front-Loaded Positioning

Agency pages that state their specialty in the first paragraph not buried in a philosophy statement get cited frequently. Lg2’s homepage receives 102 citations. Not because of paid placement. Because the opening paragraph says, directly, what they do and who they serve.

3. Third-Party Directories with Structured Data

Semrush, Sortlist, and helloDarwin collectively account for hundreds of citations in our market. These are trusted sources with clean, parseable attributes: ratings, specialties, location. AI models don’t have to interpret or infer — the structure does the work.

Notice what’s absent from this list: long blog posts with vague positioning, service pages written like brochures, or case studies without structured summaries.

We Had the Right Articles. We Still Got 0 Citations

Here’s the part that surprised us most. We had already written the list articles. We have a page for the best branding agencies in Montreal. One for digital marketing agencies. One for web production services. One targeting the Laval market.

Combined, these pages generate fewer than 5 AI impressions. The content existed. The citations didn’t come.

The problem wasn’t the topic it was the signal chain behind it. Lafusee.net and Knowltonquebec.ca have accumulated external references, editorial links, and third-party mentions over months or years. Our newly published list articles had none of that weight behind them.

This is the point most agencies miss when approaching AI visibility: publication alone isn’t the lever. The question is whether external sources corroborate what you’ve published.

Princeton University’s research on Generative Engine Optimization confirms this. Citing authoritative external sources in your content increases AI citation rates by up to 40%. Including specific statistics adds 37%. Quotations from identifiable experts contribute another 30%. Structure matters but so does the ecosystem pointing back at you.

The Exact Queries AI Models Fire About Agencies

This is the most actionable data we have.

When a user prompts any major AI model with a question about Montreal marketing agencies, these are the actual search queries that get run:

• “top branding agencies Montreal Quebec 2026”
• “best digital marketing agencies Montreal Canada”
• “Montreal branding agencies competitors Lg2 Sid Lee”
• “best web development agencies for startups Montreal”
• “agencies marketing numérique Montréal DDB Canada concurrents”
• “top web advertising agencies Montreal Quebec digital marketing agencies”

The same cluster fires across ChatGPT, Grok, Gemini, Perplexity, DeepSeek, and Claude sometimes within the same hour.

The implication is direct: if your agency’s name doesn’t appear in a source that ranks for these specific queries, you don’t exist in AI search regardless of your actual reputation or client results.

The 5 Factors That Actually Drive AI Recommendations

Based on what we track across seven models over several months, here is what determines whether an AI model recommends you:

1. Third-Party Mentions from Credible Sources

External sources directories, editorial roundups, industry publications that name you specifically. Each external citation is a corroborating signal the model can cross-reference.

2. Explicit Positioning on Your Own Pages

Your homepage and services page should answer “what do you do, for whom, in which city” in the first two sentences. Not in paragraph four of a brand philosophy. AI models read the beginning of each section first.

3. Content Freshness

The roundups that dominate citations are updated annually. The URL often contains the year. AI models weight recent content higher for competitive, location-specific queries where currency matters.

4. Anchor Text in External Links

When a third-party site links to you as “digital marketing agency Montreal,” that anchor establishes topical relevance. Generic links (“click here” or your brand name alone) carry significantly less signal.

5. Structured Markup

Pages with FAQPage, LocalBusiness, or Organization schema give AI models clean, unambiguous signals. Unstructured text requires the model to infer. Structured data eliminates inference and inference is where citations get lost.

What We Changed at Agence Minimal

Three concrete changes came out of this analysis.

First, we are actively building editorial references to our existing list articles from Quebec industry sources. Not generic directory submissions editorial mentions that establish topical context.

Second, we restructured our service pages to front-load positioning. The opening sentence now states what we do, where, and for whom. The philosophy paragraph moved down.

Third, we’re publishing content that has no direct equivalent in our competitive space articles AI can cite for queries nobody else is targeting. There is no roundup titled “how AI models decide which agency to recommend.” Now there is one.

These changes complement our work in SEO, content creation, and GEO optimization.

How to Monitor Your Own AI Visibility

You can track your AI citation count using tools like Otterly.ai, which monitors how often AI models surface your brand when running queries in your market. The report shows your mention count, your competitive rank, and which specific queries are or aren’t finding you.

The data is more granular than traditional SEO analytics. You see the exact queries AI fires. You see which competitor URLs are getting cited instead of yours. You can identify the gap, then close it systematically.

For agencies in particular, this data is becoming a business development metric. A growing share of agency searches now start in an AI assistant rather than Google. If you’re not measuring it, you’re operating without a full picture of how you’re being found or not found.

FAQ