There is no single "best AI" for a business, only the right AI for a given use. For everyday writing and communication, a general-purpose assistant — ChatGPT, Claude, Gemini, Microsoft Copilot or Mistral's Le Chat — is enough, and the right choice depends mostly on the work environment already in place: Copilot if the company lives in Microsoft 365, Gemini if it lives in Google Workspace, Mistral if European hosting and data sovereignty come first. But as soon as the AI needs to know your prices, clients, documents or processes, no general-purpose assistant is enough: you need an AI connected to your data, meaning an agent or a custom tool.
The confusion that wastes time
The question "which AI should we use?" mixes two very different things.
The first is the assistant: an interface where an employee asks a question or pastes a document and gets an answer. It is paid per user per month, deploys in a day and helps immediately. But it only knows what it is given in the conversation.
The second is AI built into your processes: a tool that fetches information from your systems itself, applies your rules and produces a usable result without anyone copying anything. That is where five- and six-figure gains are found, and it is a project, not a subscription.
Most companies need both. The classic mistake is buying the first while hoping to get the second.
The five major assistants, compared on what matters
| Assistant | Strength | Choose it if… |
|---|---|---|
| ChatGPT (OpenAI) | The most widespread, highly versatile, large extension ecosystem | Your teams already use it and the need is general |
| Claude (Anthropic) | Long-document reading, analysis, writing quality | You work with large contracts, files and reports |
| Gemini (Google) | Built into Gmail, Docs, Sheets and Drive | Your company runs on Google Workspace |
| Microsoft Copilot | Built into Word, Excel, Outlook and Teams | Your company runs on Microsoft 365 |
| Le Chat (Mistral AI) | French company, European hosting options and models deployable on your own infrastructure | Data sovereignty and GDPR are the priority |
Transparency: Codito Ergo Sum is a certified Anthropic partner. We nonetheless regularly deploy other models for clients, because the right model depends on the task, the data and the constraints — not on the partnership.
The four criteria that actually decide
1. Where your documents already live
This is the most underestimated criterion. An assistant built into the tool your teams open all day will be used; one more tool in one more tab much less so. Adoption matters more than a few points on a leaderboard.
2. The confidentiality of what you will entrust to it
Before any rollout, check three things: whether your data can be used to train the model, where it is hosted, and how long it is retained. Business plans generally exclude using your data for training; that is not always the case for free consumer versions. An employee pasting a client contract into a personal free account is the most common risk — and the least visible. We cover these issues in our page on security and sovereignty.
3. The nature of the task
Writing a letter, summarising a meeting, rewording an offer: every assistant does well. Analysing a 200-page file, comparing contract versions, producing a long document that follows a strict template: the gaps become noticeable, and it is worth testing two models on your own documents before rolling out.
4. The need to connect to your data
This is the criterion that turns choosing an assistant into a project. If the task requires knowing your catalogue, client history, pricing or procedures, the question is no longer "which assistant" but "which tool connected to our systems". See our comparison custom AI development or SaaS.
What it costs
A general-purpose assistant is paid per user per month, with business plans whose price varies by vendor and service level. The mistake is to reason in licence cost: ten unused licences cost more than a custom tool that removes a ten-hour weekly task.
For uses connected to your data, cost is counted in projects and usage: at Codito Ergo Sum, an audit starts at €2,000 excl. VAT, monthly support sits between €499 and €799 excl. VAT per month, and custom development starts at €3,000 excl. VAT. Our audits have revealed between €109,000 and €527,000 of hidden costs per year depending on the company — that is what the investment compares against.
The three-step approach
Equip: pick an assistant consistent with your work environment and open it to teams, with clear rules on what may be pasted into it.
Train: half a day on the company's real documents changes usage more than any choice of model.
Connect: identify the recurring task that costs the most and build the tool that handles it with your data. That is the step that produces the return on investment — around two months observed with our supported clients. To find which one, see where to start with AI in your business.