Why being cited by ChatGPT changes the B2B game
As of May 2026, ChatGPT has more than 800 million weekly active users (OpenAI confirmed figure on their official blog in April 2026). A growing share of these queries is commercial or comparative — "best ERP for an industrial SME", "GEO agency Paris", "Stripe vs Adyen". When your brand gets named in the answer, you bypass Google, you bypass ads, you land straight into the prospect's consideration set.
The GEO study by Pranjal Aggarwal et al. (Princeton CS, 2024, paper "GEO: Generative Engine Optimization") demonstrated that targeted content optimisation can raise a source's visibility inside generative engine answers by 30 to 40%. That study is the academic foundation of the discipline: it frames the order of magnitude structured GEO work aims for, with actual results varying by sector, starting point and competition.
How ChatGPT actually picks its sources
ChatGPT draws from three distinct pools. The first is its training corpus, frozen at a given date (April 2024 for GPT-4o at release, more recent for GPT-5). The second is real-time browsing via Bing — sources must be indexed and considered authoritative by the engine. The third is the proprietary SearchGPT, which cross-references several indexes and favours content with high factual density.
To get cited, a brand must therefore be present in the right corpora, indexed by Bing, and structured so that passages are unambiguously "extractible". That's the exact opposite of pre-2020 SEO, where you tried to occupy the SERP. Here, you try to become a source the AI cites.
Step 1 — Citability audit (week 1)
Before any action, measure where you stand. NEXUS GEO audits your brand against the Référentiel GEO-47, the open framework it publishes under CC BY 4.0: 47 criteria organised into 8 public pillars, scored out of 100. The pillars are public; the detailed criterion list, the weighting and the scoring method are part of the paid audit deliverable.
The deliverable is a 20-page PDF with an overall score out of 100, a score per pillar, and a benchmark against three direct competitors on a panel of 30 industry prompts. Until you have that baseline, any action is blind. Our [methodology page](/methodologie) presents the approach and the 8 public pillars of the [Référentiel GEO-47](/referentiel-geo-47); the per-criterion detail belongs to the audit deliverable.
Step 2 — Schema.org: the non-negotiable technical layer
Schema.org is the vocabulary AIs use to understand your site without ambiguity. Without valid Schema.org Organization markup, ChatGPT has to guess who you are — and it can get it wrong, or worse, attribute your signals to a namesake. Google Search Central's 2025 documentation confirms that structured data is an eligibility signal for AI Overviews.
Priority types for B2B are Organization (identity), ProfessionalService or SoftwareApplication (offer), FAQPage (extractibility), Article (editorial content), Person (authors). Our [Schema.org for AI](/ressources/schema-org-pour-ia-guide-technique-2026) guide details the JSON-LD implementations line by line.
Step 3 — llms.txt: the AI welcome mat
The llms.txt standard, proposed by Answer.AI in September 2024, became dominant in less than twelve months. Anthropic, Vercel, Mintlify, Cloudflare and hundreds of tech brands have deployed it. It's a Markdown file placed at the site root that summarises your value proposition, canonical pages and identity signals — under 4,000 tokens, so any LLM can read it in one pass.
In our field experience, the effect on citation rate typically becomes visible within a few weeks of deployment, especially with Perplexity and Claude, which consult the file readily. Details in [llms.txt: the complete guide](/ressources/llms-txt-guide-complet).
Step 4 — Citable content: 5 writing rules
- One claim, one source. Every figure, every claim must be attributed to a dated source. AIs cross-check sources and favour traceable content.
- Definitions in the first sentence. LLMs extract opening sentences to answer "what is X?" queries. Put the definition under an H2 and answer in fewer than 40 words right below.
- High factual density. One number, date or proper noun every few sentences. The Princeton GEO study found that adding relevant, sourced statistics is among the optimisations that most improve visibility in generative answers.
- Scannable structure. H2 every 200-300 words, ordered lists, comparison tables. AIs read by HTML block.
- Explicit FAQ. A FAQ section with Schema.org FAQPage at the end of every commercial page. That's the format ChatGPT and Perplexity reuse most directly.
Step 5 — Domain authority: the slowest layer
Technical citability is a prerequisite; domain authority decides your frequency of appearance. LLMs weight sources with signals comparable to Google PageRank: number of referring domains, editorial quality of those domains, coherence of the Knowledge Graph (Wikipedia, Wikidata, Crunchbase, the SIRENE company registry for France).
Three levers carry most of the weight here. First, entity recognition: being present and unambiguous in the public knowledge bases AI engines lean on (Wikipedia, Wikidata, official company registries), so they can tell you apart from any namesake. Second, regular third-party coverage: dated mentions in credible sector media compound into authority over time. Third, publish proprietary studies that media will cite — the most durable lever of the three. Deciding which lever to pull first, in what order and with which supporting signals is precisely the sequencing work of a GEO engagement.
4 illustrative scenarios
The profiles below are representative scenarios, not named client records. They describe typical trajectories observed on this type of engagement, with orders of magnitude — your own results depend on your starting point, your sector and your level of competition.
Scenario 1 — French HR SaaS, small team
Starting point: only a handful of citations across roughly thirty tested industry prompts (ChatGPT, Claude, Gemini, Perplexity). After about 90 days of structured work (audit + Schema.org + llms.txt + pillar content + Wikidata entry): coverage typically rises to more than half of the prompts, the citability score climbs sharply, and AI-sourced traffic appears with a demo-request rate well above that of classic channels.
Scenario 2 — DTC fashion brand (premium footwear)
Brand already established on classic SEO, weak on GEO. After a few months: presence across a clear majority of the conversational shopping prompts tested ("best leather sneakers made in Europe", "ethical footwear brands"), where the brand was almost never cited at the start.
Scenario 3 — Regional law firm
Strong local Google presence, invisible on ChatGPT. Targeted engagement on a dozen prompts like "recommended employment lawyer [city]". Within a few months: a shift from near-total absence to a regular presence in answers, and the first inbound enquiries attributable to the AI channel. For this type of service, the high unit value per case can make the return on investment fast.
Scenario 4 — Climate deeptech B2B startup
Target: citations inside thematic analyses ("French startups decarbonising industry"). Approach centred on producing proprietary studies designed to be picked up by sector media, plus a Wikidata entry and a full llms.txt. After a few months: the startup becomes a reference regularly cited in thematic queries on Claude and Perplexity, and appears in a large share of generated sector comparisons.
Step 6 — Measure to steer
You can't steer what you don't measure. NEXUS GEO operates a proprietary tracker that queries ChatGPT, Claude, Gemini, Perplexity and Copilot monthly on 30 industry prompts per client, computing three metrics: citation rate (prompts citing the brand / total), average position (rank inside the answer), and AI share of voice (brand citations / total citations across all brands).
To go further, see our dedicated article on [measuring AI citations](/en/ressources/measure-ai-citations-tools-methods-2026) which compares free tools (Google Search Console, Bing Webmaster Copilot tab), paid tools (Profound, AthenaHQ, Goodie AI) and DIY approaches.
FAQ
Sources
- Aggarwal P. et al. — "GEO: Generative Engine Optimization" — Princeton CS, 2024 (arxiv.org/abs/2311.09735).
- OpenAI — Official blog, ChatGPT usage stats, April 2026.
- Google Search Central — "Structured data and AI Overviews" documentation, 2025.
- Profound — "AI-referred traffic conversion benchmark" study, 2025 (tryprofound.com).
- Answer.AI — llms.txt standard proposal, September 2024 (llmstxt.org).
- Similarweb — "AI chatbot traffic insights", 2024-2025.
Want to move beyond theory and apply this to your own site? [The NEXUS GEO audit](/audit/new) delivers it in 10 days. No commitment.
Want to know whether ChatGPT cites you?
NEXUS GEO measures your current citation rate on 30 industry prompts and ships an action plan in 10 days. Complete audit on 47 criteria for €1,750 — 20-page PDF, score per pillar, 6-month action plan, 60-minute debrief. No commitment: deliverables and timelines are stated upfront, and results depend on your market and execution.
