ChatGPT, Claude or Gemini: a guide for notaries
The three major AI assistants compared for notarial practice: each family's lineup, what they do well, and above all what happens to what you share with them depending on the plan you choose.
Read first. This comparison is current as of October 9, 2026. Models change every two or three months: check versions and terms on the official pages before subscribing. This guide is not legal advice.
ChatGPT, Claude and Gemini are the three most widely used artificial intelligence assistants. All three draft, summarize, translate and analyze documents. For a notary, the question is not only "which one is best?" but also "what happens to what I share with it?". The answer depends less on the model than on the plan.
Three families, one logic
Each provider offers a lineup rather than a single model:
- a fast model, for simple questions and short answers;
- a general-purpose model, used by default;
- a reasoning model, slower and more expensive, which "thinks" before answering and suits long analyses.
| Family | Provider | Lineup as of October 9, 2026 | Plans |
|---|---|---|---|
| ChatGPT | OpenAI | GPT-6, in two variants since October 7, 2026: Sol for paid plans, Luna for the Free and Go plans | Free, Go, Plus, Pro (personal); Business, Enterprise (business) |
| Claude | Anthropic | Haiku (fast), Sonnet (general-purpose, default), Opus (most powerful, paid plans) | Free, Pro, Max (personal); Team, Enterprise (business) |
| Gemini | Flash-Lite and Flash (fast), Pro (most powerful); Gemini 4 announced, first for the Ultra plan | Free, AI Plus, AI Pro, AI Ultra (personal); Google Workspace (business) |
Version numbers matter little: they will change before the end of the year. Remember the lineup instead, and the fact that the free plan gives access to the least powerful models. At Google, since October 9, 2026, the free plan is limited to the lightest model.
What they do well, and less well
For an office's everyday tasks (rewording a letter, summarizing a contract, explaining a concept, preparing a checklist), the three families give comparable results in their paid versions. The differences come down mostly to each one's style and habits: try them on your own texts.
A few things they have in common:
- They can make things up. A statute section that does not exist, a decision cited wrongly, an incorrect deadline: every model does this occasionally, with confidence. Every reference must be checked at the source.
- They do not consult the land register. They know neither the status of a lot nor the registrations that affect it. They work only on what you give them, and on what they learned during training.
- Handwriting remains hard. Old property indexes and scanned deeds are difficult for all three. In our own tests on handwritten land register indexes, the gaps between models were clear: if you read old documents, compare on your own documents rather than relying on reputation.
- Long documents work, with limits. All three read PDFs of several dozen pages; the accepted length and number of files vary by plan.
The real difference for a notary: the plan
A file contains personal information on every page, and you are bound by professional secrecy. Before pasting a deed into an assistant, you need to know whether your text can be used to train models, whether people can read it, and how long it is kept.
On these three points, the dividing line is not between ChatGPT, Claude and Gemini: it runs between personal plans and business plans.
| Personal plans (free or paid) | Business plans | |
|---|---|---|
| ChatGPT | Your conversations may be used for training by default. You can turn this off in Settings › Data controls › Improve the model for everyone, for later conversations only. | Business, Enterprise and API: no training on your data by default. |
| Claude | The user chooses whether to allow their conversations to be used for training. If allowed, retention can last up to five years; if declined, conversations are deleted within 30 days, with exceptions (safety, legal obligations). | Team, Enterprise and API: governed by the commercial terms, no training on your data. |
| Gemini | Activity is kept by default and some conversations are read by human reviewers; reviewed conversations are kept for up to three years, even after you delete your history. Turning activity off stops training and review going forward. | Google Workspace: no human review or training outside your organization without your permission. |
Three findings stand out:
- Paying is not enough. A paid personal plan (ChatGPT Plus, Claude Pro, Google AI Pro) gives better models and more usage. It does not, in itself, change how your data is handled.
- Settings matter. On a personal plan, the privacy options are often on by default. They can be turned off, but only for future conversations.
- "No training" does not mean "no retention". Even in business offerings, providers generally keep exchanges for a while, for example to detect abuse. Ask for the exact duration.
What about Quebec's Law 25?
These services process data mainly outside Quebec, often in the United States. Entrusting personal information to such a service means communicating it outside Quebec: Law 25 then requires a privacy impact assessment and contractual safeguards. A personal plan, subscribed in a few clicks, gives you neither.
We have gathered the questions to ask a provider in Loi 25 et IA : 10 questions à poser avant de confier vos dossiers à un logiciel (in French).
So, which one should you choose?
A simple rule, whatever the family:
- To learn, draft a generic text or explore a concept: any of the three, in its personal version, will do. Do not put any client information in it.
- To work on real files: only a business plan, with a contract you have read, a known retention period and a Law 25 assessment. The choice of family comes next, based on your tests.
- For a task specific to notarial practice, such as a title search, a general-purpose assistant is not enough: it does not consult the register. That is the job of a specialized tool, which we explain in AI agent or AI model: what notaries need to know.
A habit to adopt now: never paste a deed, a client's name or address into an assistant opened with a personal account.
Sources consulted on October 9, 2026: announcements and help centers of OpenAI (GPT-6, data controls, Business and Enterprise data policy), Anthropic (plans, consumer and commercial terms) and Google (Gemini plans, Gemini Apps Privacy Hub, Google Workspace Privacy Hub). Transparency: NotairIA itself uses models from OpenAI and Anthropic, through Microsoft Azure. None of these providers reviewed or funded this guide.