Article in collaboration with Getfluence
On March 20, 2026, Julien Bismuth, SEO/GEO consultant at Getfluence, and Olivier de Segonzac, co-founding partner of Resoneo, co-hosted a 30-minute session at the SEO Summit. Their topic: moving beyond merely measuring visibility in AI engines to taking action, with concrete practices and observable results. Here is the detailed report of their presentation!
The starting observation: being visible is no longer enough
2026. AI visibility reports are multiplying. But between noting that your brand appears in ChatGPT or Gemini and understanding why it is selected, and how to actually influence the response, the connection is often hard to make.
56% of AI response sources come from third-party content (press, editorial publications, reviews, forums, social networks). This data from the Yext AI Citation Analysis, conducted in Q4 2025 on 17.2 million citations generated by ChatGPT, Gemini, Perplexity and Claude, is particularly revealing: only 44% of cited sources come from brand sites or blogs. The rest — more than half — comes from third-party content perceived as neutral and trustworthy.
In other words: SEO remains essential, but it is no longer enough. To be chosen by AI models, you need to build your brand footprint in the right places—where LLMs look for and select their sources.
How LLMs work – an important reminder
Before getting into practice, Julien and Olivier set a technical framework. LLMs do not reason: they calculate probabilities. Faced with a query, the model first decides whether to activate its web search capability:
- Without web activation : the response relies solely on the model's knowledge up to its cut-off date.
- With web activation : the model performs retrieval, that is it builds a pool of candidate URLs (the grounding phase), then selects and synthesizes.
The model chooses and selects for the user based on probability calculations. It’s a mechanism, not intelligence. But it can be influenced if you adapt to how it works.
The 10 practices to influence AI responses
1. Identify high-potential questions for the brand
Traditional search volumes are no longer a reliable indicator for queries posed to AIs. The method presented is structured in three steps.
First, start from current keywords and rankings to build Search Personas: profile, intentions, journey, barriers and opportunities for each audience segment.
Next, feed those personas to ChatGPT to generate the decision-making, comparative and commercial questions they actually ask: value searches, needs, purchase experience, decision support.
Finally, the last step is to filter high-potential questions according to three cumulative criteria:
- Absence of brand mention in current AI responses
- Presence of competitor mentions
- Persistence rate above 25%
2. Secure the best position on Query Fan-Out
When ChatGPT or Gemini activate their search engine, they automatically generate derived queries (Query Fan-Outs). One crucial point to remember: more than 50% of these fan-outs are phrased in English, even for a French-speaking user. Gemini, for its part, relies directly on the Google index.
Three concrete levers were presented:
Fan-out markers
Include in your on-site and off-site content the typical fan-out terms: best, top, best (meilleurs), comparisons, reviews, 2026…
English version of the site
At minimum, provide an English version of corporate content, best-sellers, and responses to frequent criticisms (e.g., corp.domain.com).
Meta descriptions and URL slugs
These are the elements the model reads first during the retrieval phase.
3. Spot the most frequent source sites and articles on a topic
Domain Rating or authority metrics alone are no longer sufficient; neither in SEO in 2026 nor in GEO. What matters is identifying the domains and source articles that AI models perceive as trustworthy on a given topic.
Good news : the models themselves provide the list of sources used to build their answers. The problem is volatility: to get a representative view, each question must be asked dozens of times to the target model.
This is where the tool-assisted GEO methods such as Getfluence (Spot Finder & Mentions feature) which make it possible to influence AI responses in more than 60% of cases.
Pay attention to identifying sources via the API, which may differ from the sources actually displayed in the platform's user interface.
4. Recognize good and excellent LLM-friendly spots
Once the sources have been identified, they need to be qualifiedThe method consists of analyzing two complementary dimensions:
- Citation frequency : how many times this domain or article appears in responses to test prompts.
- Source sentiment : are mentions of your brand (and your competitors) in these contents positive, neutral, or negative?
The tools presented allow you to precisely filter sources that mention competitors without mentioning your brand. These are the priority targets for placement or relinking actions.
5. Build Ambassador URLs and Review URLs
This is one of the most powerful strategies presented during the session: the relinking strategy, which produces a simultaneous triple impact.
- Source URL : First identify press articles, product tests, expert reviews, or comparison guides that mention your brand positively.
- Relinking : Publish new content that cites the source article and create backlinks to that URL (guest posts, partner articles, press releases).
The result achieved: a triple impact
- SEO : better ranking for the source URL
- Online reputation : increase in positive mentions
- GEO : higher likelihood of being cited by LLMs
As the two experts summarize: links feed Google. Mentions feed LLMs. With a single relinking piece of content, you achieve both objectives simultaneously.
6. Strengthen trust and freshness from within (E-E-A-T signals)
Google has 27 years of experience in assessing content quality. ChatGPT has 3, and it quickly understood the value of learning from its elder. The E-E-A-T signals (Experience, Expertise, Authoritativeness, Trust) are now integrated into how models evaluate sources. Page 27 of the Google Search Quality Rater Guidelines (updated September 2025) is unambiguous: Trust is the most important member of the E-E-A-T family.
7. Embrace neutrality to be highlighted
This point was probably one of the most counterintuitive revelations of the session. In March 2026, perceived neutrality became an increasingly decisive selection criterion for models — a criterion that would have been considered less central six months earlier.
The logic is simple: the LLM must be able to compare multiple options to construct a credible answer. Content that mentions only a single brand or solution is perceived as promotional, and models avoid it as a primary source.
 Single-brand contentDetected as promotional by the models. AIs prefer guides, documentation and educational articles. They avoid advertisements and biased comparisons. |  Multi-brand content (AI-First format) )LLMs love pages with lists, comparisons and tool rankings. An article that mentions multiple solutions significantly increases its probability of being cited. |
A real test was presented : an AI-First article published in February 2026 by Getfluence on monimmeuble.com, comparing six electronic signature solutions for real estate. Result before the campaign: no mention or citation of the client brand (Oodrive) in AI responses. Result after the campaign: the brand appears in the top position among recommended solutions, with the article cited as a source.
8. Merge your link-building and mention-building efforts
Why separate SEO netlinking campaigns and brand mention GEO actions when a single piece of content can achieve both objectives?
The recommendation is simple : in all your backlink purchase campaigns, systematically include mentions of your brand and all its entities (products, executives, certifications, use cases…). An article about interior decoration for an e-commerce site that mentions the brand can at the same time generate an SEO signal and be cited as a source in an AI answer on the topic.
The example shown illustrated a ChatGPT answer about decorating a living room with a velvet sofa, directly citing an article from Frenchyfancy — a piece that had integrated a client's brand while addressing the topic editorially.
9. Produce an up-to-date format and make sure the door is open
A practice often overlooked but decisive: AI crawling bots must be able to access your content. If your robots.txt file blocks LLM crawlers, no content, however well optimized, can be used to build an answer.
The main bots not to block were listed:
- ChatGPT / OpenAI : OAI-SearchBot (real-time search bot), GPTBot (training bot)
- Google / Gemini : Googlebot, Google-Extended
- Claude / Anthropic : Claude-SearchBot, Claude-User (note: Claude-Web no longer exists)
- Perplexity : Perplexitybot, Perplexity-User
The Getfluence platform includes an AI accessibility module that automatically analyzes each domain's robots.txt file and flags partial or total blocks by LLMs; a considerable time-saver for auditing your network of partner sites.
10. Provide new information – Information Gain
The last point may be the most strategic in the long term. Julien and Olivier presented Google patent US12013887B2 (granted in June 2024), which assigns each piece of content a Information Gain score between 0 and 1, measuring the amount of truly new information it brings.
| Score → 0: generic contentRewriting what already exists (top 10 of 2026, generic comparisons…), even when produced by AI. The model has already seen this information dozens of times. | Score → 1: original informationProprietary studies, internal test data, quotes from field experts, verified trial results, first-hand customer testimonials. |
Keep in mind that if an LLM has already seen the information 50 times, it won't cite it a 51st time. In other words: bring what no one else can say.
On product pages, this principle translates into a gap to fill: brands list technical specifications, but users describe situations and constraints. It is in this gap that the LLM chooses its sources. Two concrete actions were proposed:
- On listing pages (PLP) : create situational facets based on real use cases ("Fits through metro turnstiles", "Scratch-resistant", "Compatible with glasses").
- On product pages (PDP) : add situational paragraphs and FAQs drawn from customer reviews, customer service, and forums; the contextual layer that allows the LLM to recommend a specific product.
Key takeaways
During this presentation, Julien Bismuth and Olivier de Segonzac managed to establish a clear methodological framework to move from measurement to action in GEO. The central logic: AI models are probability machines. You can influence those probabilities by simultaneously working on content quality and perceived trust, presence in third-party LLM-friendly sources, and the neutrality required by AI models.
The common thread of the whole session is that SEO and GEO are not opposed; they reinforce each other when you adopt the right formats and distribution strategies. The AI-First format Developed by Getfluence, it embodies this convergence: structured to meet AI engines' selection criteria while respecting the editorial standards that make content credible and linkable.
In 2026, being visible is no longer enough. You must be chosen.
The article “The 10 practices that truly influence AI responses” was published on the site Abundance.
