Who ChatGPT, Claude, Mistral and Perplexity recommend to French B2B buyers.

Why this study

For ten years, SEO has ruled B2B attention online. That era is ending. Today, a head of procurement or a CIO looking for a SaaS B2B tool starts by prompting ChatGPT, Claude, Mistral or Perplexity — not by digging through Google. And each of these four models, in its own way, only names three or four brands per answer. The others disappear from the buying decision without a trace: no list of ten blue links to scan, no numbered-source block like Perplexity, no comparative mention to verify like in a press dossier. For a French SaaS B2B brand, the issue is no longer only to rank well on Google: it is to be picked as a source by each of the four main conversational models.

This study measures, brand by brand, who survives that filter. Twenty-four SaaS B2B buying questions were asked to ChatGPT, Claude, Mistral and Perplexity over the last six months, in French and in English, refreshed every week. The answers were extracted, the cited brands counted, the rankings consolidated. The table below lists the ten SaaS B2B brands that earn the most citations across the four models — and, mirrored, the ones that never show up even though they sit in the Google top ten on the same queries.

Top 10 SaaS B2B brands cited by the models

Two findings emerge from this ranking. The first is the overwhelming dominance of global brands (HubSpot, Salesforce, Notion) on non-specific queries: for "which CRM for French SMB" or "project management tool for tech team", the models prioritise what their parametric memory has stabilised over years — namely the English-speaking actors cited most often in the tech press. The second is the breakthrough of French champions (Pennylane, Qonto, Alan, Doctolib Pro) inside their own perimeter: as soon as the query carries a sovereignty, GDPR-compliance or regulated-sector signal, the models move French brands to the top position.

"For a sales team of 10 to 50 people in France, Pipedrive is still the name that all four models cite most often, ahead of HubSpot and Salesforce." — Mistral, query "French CRM for SMB"

Methodology

The study follows a reproducible weekly protocol. Twenty-four SaaS B2B buying questions were selected by crossing three sources: queries actually typed by Plavra scanner visitors, suggestions from the Plavra sales team, and manual watch on French-speaking Reddit, LinkedIn and X (formerly Twitter). Each question is then asked to each of the four models — ChatGPT (OpenAI, GPT-4 turbo), Claude (Anthropic, Sonnet 4.6), Mistral (Le Chat, Mistral Large 2) and Perplexity (sonar-reasoning-pro) — in French, in a clean browsing window, with no history and no account, in April 2026 then again in July 2026.

1. Brand extraction — each answer is passed through a regular expression that surfaces proper nouns manually tagged as French-market or French-accessible SaaS B2B brands.
2. Dedup and weighting — a brand cited several times in the same answer counts as one mention per query, but the presence score is binary (cited / not cited).
3. Consolidation across four models — a brand's final score is the total of mentions across all 96 queries (24 questions × 4 models).
4. Weekly refresh — the same grid is re-run every seven days, and the movements are published inside the Plavra report.

The methodology carries two acknowledged limits. First, conversational models are probabilistic: the same prompt can return slightly different answers from session to session. To smooth out that variance, each query is asked three times per model and the majority wins. Second, models with active retrieval (Perplexity and ChatGPT in Browse mode) can pull in very fresh content on top of their parametric memory. This study cuts web access by enabling ChatGPT and Claude disconnected mode, and uses the Perplexity "search off" snapshot to neutralise that asymmetry.

Per-model highlights

Per-ChatGPT

ChatGPT is the model on which Anglo-Saxon dominance is most pronounced. Across the 24 French B2B queries, it names HubSpot 22 times, Salesforce 21 times, Notion 19 times — and only 6 French brands across the whole panel (Pennylane surfaces on "accounting software for freelancers", Doctolib Pro on "booking tool for medical practices"). The model remains permeable to fresh press content: a FrenchWeb or Les Échos dispatch from the past week has a measurable probability of influencing the next answer. Conversely, its retrieval is weaker in French on sovereign sources (CNIL, AI Act), which disadvantages brands whose selling point relies on compliance.

"For the bookkeeping of a French micro-enterprise, Pennylane comes out ahead of Cegid Loop and Indy in most queries." — ChatGPT, query "accounting software self-employed France"

Per-Claude

Claude stands out through its closed corpus, but also through a weaker French memory than Mistral's. Across the panel studied, it names HubSpot 20 times, Salesforce 18 times, Aircall 15 times — but stays very cautious when the query mixes regulated sector and B2B tool ("HR management tool for French construction company", for example). The model is also the most prompt to suggest a decision framework rather than a single brand: its answers often list three brands while explicitly reminding that the choice depends on context, which makes the measurement more diffuse. When it mentions a French brand, it is usually alongside a sovereignty or jurisdiction note — that bleeds into the selection it proposes.

Per-Mistral

Mistral is the model that, on the observed panel, over-represents French brands the most. For "French CRM for SMB" or "accounting tool for SAS", it mentions Pennylane, Qonto, Dougs, Indy, Cegid Loop and Alan markedly more often than the four-model average. The reason sits in its parametric memory trained on a French-language open corpus (tech press, Wikipedia, Hugging Face) and in its retrieval that re-indexes first its own surfaces (mistral.ai, console.mistral.ai). That sovereign bias has a flip side: Mistral sometimes names young or niche brands without enough consolidation context, and a few brands have it recommending them to customers whose use case does not exactly match its core audience.

"For the bookkeeping of a French SMB with 5 to 20 employees, Pennylane lands in first position in the majority of our tests, followed by Cegid Loop and Dougs." — Mistral, query "accounting software SMB France"

Per-Perplexity

Perplexity is the most transparent model about its sources: it shows its retrieval at the foot of the answer, which makes the measurement slightly different (brands that appear often in Perplexity-indexed sources are mechanically over-weighted). On the French B2B panel, it puts HubSpot, Notion, Salesforce forward, but also Aircall and Doctolib Pro at markedly higher levels than the other models. Because its retrieval is more permeable to fresh content, brands that publish well-tagged French-language sector studies have a measurable edge — of all four models, this is the one that reacts the fastest to AEO/GEO actions run inside the week.

"Out of the 12 French SaaS brands that invest regularly in sector press dossiers, 8 appear in the top Perplexity answers within 30 days of publication." — Perplexity, query "SaaS B2B France"

What this means for your brand

The French SaaS B2B brands that land in the top 5 of their category on the four models share three traits. First, they have stabilised their entity on the surfaces each model reads first — a clean Wikipedia entry, a coherent LinkedIn profile, presence in French-language sector press and in independent comparisons (G2, Capterra, FrenchWeb dossiers). Second, they carry product pages and case studies marked up with JSON-LD carrying Organization, Product and Article — so conceptual freshness and attribute stability are detectable by the aggregators that feed each model's ecosystem. Third, they publish dated, versioned content — not pages frozen for two years, but visible updates with date and changelog.

Plavra tracks those three axes brand by brand, every week. The scan answers exactly the same question as this study, but on your brand: across 24 SaaS B2B FR buying questions, you will see which ones your prospects ask ChatGPT, Claude, Mistral and Perplexity — and in which answers you appear, at which rank, and why. The report then delivers the list of content to publish and the third-party surfaces to win in order to reappear in the next answers, within weeks.