We would like to warmly thank everyone who joined us and those who took the time to engage with this key topic for the future of search.
For those who couldn’t attend (or who want a recap), here are the main takeaways.
Key Takeaways
1. Traditional attribution is broken: AI assistants now influence 20–50% of purchase decisions without leaving measurable traces in Google Analytics, creating a major blind spot for brands.
2. Mint’s 5-step framework: A structured methodology to master AI search, from fixing attribution to executing a concrete optimization roadmap.
3. The new essential metrics: Visibility (brand mentions), sentiment (AI-generated score), and alignment (accuracy of positioning) are replacing traditional SEO KPIs.
Why Do Brands Need to Rethink Their Search Strategy with AI?
The transformation of the user journey by answer engines
AI answer engines are fundamentally changing how users discover and evaluate brands. The bottom line: consumers now get synthesized answers directly, without visiting websites. This represents a paradigm shift where AI becomes the main intermediary between brands and their audiences.
“…this topic of recommendations in AI is the key issue of the moment that must be mastered, since it is increasingly an important acquisition lever and also a branding lever…” — Joan Burkovic
The Limits of Google Analytics in the Face of AI Assistants
Google Analytics captures only 1–2% of referral traffic coming from AI assistants, while these actually influence 20–50% of decisions. This gap creates a critical measurement issue for marketing teams, who can no longer rely solely on traditional metrics.
The Rise of a New Discovery Ecosystem
A complex ecosystem is emerging, where ChatGPT, Gemini, Perplexity and others become acquisition channels in their own right. Each platform has its own mechanisms for selecting and citing sources, requiring a differentiated approach.
Why AI Doesn’t Click Like a Human: New Attribution Rules
AIs synthesize information without generating traditional clicks. This reality forces a complete rethink of attribution models, integrating qualitative signals such as brand mentions and the context of recommendation.
The Mint AI Search Framework: A Structured 5-Step Approach
Step 1: Fix Attribution and Capture AI Signals
The first step is to implement explicit capture mechanisms. For product teams, this means enriching existing forms (e.g., “How did you hear about us?”) by adding AI models (ChatGPT, Gemini, Perplexity, etc.) as response options. B2B teams should integrate these questions into their CRM to track AI influence on the sales pipeline.
“…all discussions start with: ‘it’s 2% of our traffic’. […] When we do all this […] we realize that […] the range is broad, but 20–30–50% in some cases.” — Matthieu Poitrimolt
Step 2: Build Your AI Intent Database
Developing an intent index specific to AI queries is crucial: it should leverage analysis of long-tail queries, support tickets, and sales conversations to identify real questions asked to assistants, while also drawing from Google keywords to capture SEO-proven formulations and enrich the intent database.
Step 3: Redefine Your KPIs for the AI Era
The new KPIs are structured around three pillars:
- Visibility (frequency of mentions),
- Sentiment (tone of recommendations),
- Alignment (accuracy of positioning).
These metrics provide a holistic view of performance in the AI ecosystem.
“In AI models, the goal is to be mentioned, but also to be mentioned well.” — Matthieu Poitrimolt
Step 4 — Analyze and Prioritize the Sources Consulted by AIs
The analysis of 16,000 sources reveals distinct patterns depending on the models. Understanding which sources influence each AI allows optimization efforts to be prioritized on the most impactful platforms.
Step 5 — Take Action: Define a Pragmatic and Prioritized Roadmap
The final roadmap combines three approaches:
- influence external sources,
- refresh existing content,
- create unique assets.
This mixed strategy maximizes the chances of being positively cited by AIs.
How Do AI Answer Engines Actually Work?
The Technical Process: From Query to Synthesis
The technical journey begins with analyzing and expanding the user’s query. The system then deploys several retrieval modes in parallel before scoring and synthesizing the results with citations.
The Different Modes of Information Retrieval
AIs simultaneously use keyword search, vectors, code, and SQL. This multi-modal approach ensures comprehensive coverage of relevant sources.
The Crucial Importance of Sources and Citations
Citations are the new currency of trust. Models prioritize authoritative and recent sources, creating an implicit hierarchy of influence.
Which Sources Most Influence AI Answers?
The Analysis of 16,000 Sources: ChatGPT vs Gemini vs Perplexity
Our study reveals distinct preferences: ChatGPT favors academic sources, Gemini prioritizes the Google ecosystem, while Perplexity diversifies its references more.
Wikipedia & Wikidata Cases: Opportunities and Constraints
Wikipedia remains a major source for all models. Key point: maintaining an updated presence on these platforms is non-negotiable for AI visibility.
Reddit Case: Massive US Adoption, Emerging in Europe
Reddit strongly influences B2C recommendations in the United States. Its growing adoption in Europe represents an opportunity to seize quickly.
The Importance of External Sources
Third-party mentions carry more weight than proprietary content. This reality requires developing an AI-oriented PR and partnerships strategy.
Differences by Language, Market, and Question Type
Sources vary significantly by language and market. Technical questions favor specialized forums, while commercial queries rely more on review sites.
The Three-Pillar Optimization Strategy
Influence: Gain Positive Off-Site Mentions
Developing strategic partnerships and encouraging user-generated content becomes a priority. PR must integrate the AI dimension in their approach.
Refresh: Adapt Your Existing Content for AI
Updating existing content with conversational language and direct answers significantly improves the likelihood of citation.
Create to Own: Produce Unique and Citable Assets
Creating proprietary studies, exclusive data, and differentiated insights establishes your authority in the AI ecosystem.
How to Structure Your Content to Maximize AI Visibility?
The 5 Essential Page Structure Pillars
Optimized pages follow five principles: up-to-date content, clear H2-H3 hierarchy, answers at the beginning of sections, enriched FAQs, and structured data with schema markup.
The Importance of Conversational Language and Direct Answers
Adopting a conversational tone that reflects real questions improves relevance. Direct answers at the start of paragraphs facilitate AI extraction.
Adding Differentiated Value: Data, Studies, Unique Insights
Content that includes proprietary data and unique analysis receives a marked preference in AI citations.
The Role of Schema Markup and Structured Data
Schema markup facilitates content understanding by AIs. Proper implementation of structured data becomes a factor of technical differentiation.
Watch the GetMint Webinar Replay and Access the Presentation
Missed the live event or want to rewatch certain parts?
We invite you to view it and share it with your teams: it’s an essential resource to understand how to adapt your strategy today to the era of AI answer engines.
Frequently Asked Questions
1. Can the Impact of AI on Our Traffic Really Be Measured? 
Yes, but not with Google Analytics alone. AI referral shows only 1–2% of traffic, while real influence is 20–50%. The key is to explicitly capture signals: fields in your forms (“Did you use an AI model?”), questions asked on calls, dedicated CRM fields. Mint then automates the tracking of these signals.
2. What’s the Difference Between SEO and AEO?
- Traditional SEO: optimize for clicks and Google ranking.
- AEO (Answer Engine Optimization): optimize to be mentioned, cited, and accurately described in AI answers.
In short: SEO targets the click, AEO targets the recommendation.
3. What Are the New KPIs to Track?
Three key indicators replace classic SEO metrics:
- Visibility: is your brand mentioned? How often?
- Sentiment: are the answers positive, neutral, or negative?
- Alignment: does the wording correctly reflect your values and differentiators?
4. How to Build a Relevant Prompt Database?
Work from your business priorities: purchase intentions, top-performing SEO keywords, long queries in Google Search Console, support tickets, sales call transcripts, customer reviews.
Focus first on bottom-of-funnel prompts (comparison, decision, recommendation) to capture actionable signals.
5. Which Sources Most Influence AI Models?
It depends:
- ChatGPT: often academic + Wikipedia.
- Gemini: strongly tied to the Google ecosystem.
- Perplexity: more diversified.
In France, AIs rely heavily on comparison sites, marketplaces, verified reviews. Internationally, G2, Gartner, Wikipedia, Reddit carry more weight.
👉 The key is to map your own influence sources rather than relying on generic lists.
6. Should You Invest in Wikipedia and Wikidata?
Yes, where possible. Wikipedia remains a major source for all models, but rules must be respected (declared conflict of interest, reliable third-party sources). Wikidata, easier to enrich, is widely used in LLM training phases.
7. Does Reddit Really Have an Impact?
In the United States, yes: Reddit is massive, especially in B2C and authentic discussions.
In Europe, its influence is growing but still limited depending on the sector. Partnerships between Reddit and OpenAI are increasing its importance: better to monitor it now.
8. What Concrete Strategy Should Be Applied to Optimize AI Visibility?
Combine three pillars:
- Influence: get positive mentions in relevant third-party sources (PR, partnerships, UGC).
- Refresh: update existing content to make it “AI-ready” (FAQ, clarity, freshness).
- Create to Own: produce unique content (studies, proprietary data, exclusive insights).
9. Is AI Changing the SEO/UX Approach?
Yes: we’re moving to a hybrid model. Your site must serve both humans and AIs:
- clear, fast structure, accessible HTML,
- fresh, conversational content,
- direct answers to questions,
- structured data (schema markup).
10. Do LLM.txt Files Matter?
Not today. Unlike
robots.txt, models almost never read LLM.txt. Better to focus on:- the readability of your pages (HTML vs heavy JavaScript),
- content accessibility,
- quality and freshness of information.
