How to win at AEO in 2026

When I built the CAVIS framework, I did something no one else in the Nordic AEO field has done. I analyzed 150,000 real AI conversations to find out how and when AI actually cites brands.

The findings changed how I think about AEO. They also disproved several of the recommendations that dominate AEO guides right now. Here are the seven most important ones.

Finding 1. Citation happens in conversation turns 2 to 4, not the first prompt

The first prompt in an AI conversation is almost always generic ("what is X", "explain Y"). The model responds with general definitions and rarely cites specific brands. It is on turns 2 to 4, when the user has moved from exploratory to evaluative ("which is best", "recommend something specific", "compare A and B"), that citations explode.

Implication. If your content is optimized for "what is a combination loan", you win in AI answers no one clicks through from. If it is optimized for "which provider should I choose for a combination loan", you win in AI answers that actually lead to a purchase. Move your focus away from the top of the funnel and toward the evaluation and purchase phase. The same pattern recurs in our analysis of commercial intent in AI conversations.

Finding 2. AI cites less from perfectly SEO-optimized content than expected

Pages that read like textbook examples of answer-first formatting were surprisingly often passed over in favor of pages with denser, more fact-driven language. The hypothesis is that the models detect how heavily a page has been rewritten for SEO, and prefer what they perceive as more authoritative source text.

Implication. Stop rebuilding everything into formulaic SEO format. Write as if you are explaining something to a colleague who knows the subject. Dense prose, concrete facts, less filler.

Finding 3. Purchase intent shows up in three question types

Three types of questions triggered the majority of commercial citations across the conversations I analyzed:

• Recommendation questions ("recommend me an X")
• Comparison questions ("X vs Y, which is best")
• Experience questions ("anyone with experience with X")

Implication. Build content that directly answers these three question types for your categories. That is where conversions happen in AI search. Generic "what is" articles do not win commercially.

Finding 4. The models prefer older established pages over new optimized ones

Authority signals such as age, link profile, and mentions on authority platforms almost always beat recency. A three-year-old article from an established site beat a brand-new article that was better optimized.

Implication. If you start from zero, you have to invest in authority building in parallel with content production. Schema and answer-first without authority is like building a roof before the walls. Earn mentions in established trade media, get interviewed on industry podcasts, and publish consistently on the same platform over time.

Finding 5. Your competitors get mentioned alongside you, not instead of you

In most commercial AI answers, three to five players are mentioned at once. The user gets a comparison, not a single answer. That means "owning the answer" is the wrong mental model. The right model is being included in the comparison, ideally with the best angle.

Implication. Co-mention patterns are an underrated metric. If you are always mentioned alongside the same three competitors, that says something about where the market places you. If you are mentioned alongside a different group than the one you want to belong to, AI has learned an incorrect categorization you need to actively correct. CAVIS measures co-mention patterns continuously.

Finding 6. Schema is a hygiene factor, not a competitive advantage

Pages with correct schema markup were not favored over pages without it, in cases where both existed in the category. But pages with no schema at all were systematically passed over. Schema alone decides nothing, while the absence of schema disqualifies you.

Implication. Implement schema as a mandatory baseline investment, but do not expect it alone to lift you. It prevents you from being filtered out. Content strength and authority decide who wins among the remaining candidates.

Finding 7. Sentiment matters more than citation count

A brand with 50 neutral citations performed worse commercially than one with 15 positive ones. Two negative citations ("too expensive", "disappointing customer service") erased the effect of ten positive ones. Sentiment is the multiplier no AEO guide talks about.

Implication. Measure sentiment, not just frequency. Address negative citations actively through customer service, communication, and where relevant product changes. A negative citation in ChatGPT lives on because the model learns from it.

What this means for you

Most AEO guides say the same thing. Schema, FAQ, answer-first, authority, measurement. The advice holds. It just stops at the surface. What actually decides whether AI cites you or your competitor is the underlying evidence about what triggers citations in real conversations, and which angles they get. That is also why good AEO work has to be built on data, not assumptions.

The CAVIS framework is CitationLab's attempt to make these findings operational. It measures the four dimensions that actually count. They are Share of Voice, Citation Count, Sentiment, and Co-mention Patterns. And it builds a continuous feedback loop to improve them.

If you want to see how your brand performs against the seven findings above, you can start with an AI visibility analysis.