A practical AI guide for executives (Claude, ChatGPT and Gemini)

1. What is AI useful for in an executive's work?

In company leadership, AI is useful for five families of tasks that take up a good part of the week: reading, preparing, writing, analyzing and automating. In every case the principle is the same: AI produces a first version, and you review it. For your first tests, use fictitious documents or documents you are authorized to share; section 5 explains why.

Reading and summarizing. Long reports, contracts, minutes, endless email threads. AI does not replace reading what matters, but it tells you where it is. Example: you upload a subsidiary's quarterly report and ask for one page with the key points, the risks and the open questions, each with the page of the original where you can check it.

Preparing. Executive committees, client meetings, interviews. You give the context and it returns an outline: objectives, points to cover, likely objections, questions to ask. Example: a project configured to prepare your committees, with the context, goals and limits already defined, that you reuse every month. It is detailed step by step in section 3.

Writing. Sensitive emails, internal memos, proposals, announcements. AI produces a first version in the tone you specify; you decide what stays. Example: you need to announce an organizational change to your team. You give the context, what cannot be said yet and the tone you want, and you ask for two versions: one direct, one more explanatory.

Analyzing. Data tables, supplier comparisons, scenarios. Ask it to explain its calculations: that way you can review the method it presents. But the checking is up to you: figures, units, taxes and formulas, in the original documents or in a spreadsheet. Example: you turn three supplier offers into a table with the same criteria (price, lead times, terms, risks) and ask it to point out what is missing from each one.

Automating. Tasks you repeat every week with the same pattern. Examples: an assistant that turns meeting notes into minutes and next steps, or a simple automation that sorts emails and prepares draft replies.

In my sessions, what holds executives back most is not doubting that AI can help them: it is not knowing where to start. The plan to get started is in section 6.

2. What is an AI mindset?

An AI mindset means knowing what to ask AI, which tool to ask, and where its limits are, in your own work. It is not technical knowledge: it is a work habit built on three reflexes.

Give context. AI does not know who you are, which company you run or what you are looking for. The more specific the context (industry, audience, goal, constraints), the better the result. When an answer disappoints, check your request first: did it include all the context needed? If it did and the result still falls short, the tool may have misread the document, or the task may not suit it.

• Without context: "Write me an email for a client who is complaining."
• With context: "I run a B2B services company. An important client is complaining about a two-week delay on a delivery. We want to keep them, acknowledge the mistake without blaming the team and offer a concrete solution. Sober tone, 150 words maximum."

Verify. AI can produce texts that are convincing and wrong. Figures, names, quotes and dates are always checked at the source. A good habit: ask it where each piece of data comes from, then check that the source exists and actually says what the AI claims.

Iterate. The first answer is a draft. You improve it by asking for specific changes: shorter, a different tone, a different angle, a table instead of text. If the result does not work, it is better to rephrase the request with more context than to repeat the same one.

And one rule that sums up all three:

Of the time you save on execution, you need to reinvest part in planning and reviewing.

3. Why is a configured project better than a prompt?

For a recurring task, a well-configured project is worth more than a hundred one-off prompts, because it stores your context, goals and limits, and reuses them every time.

All three tools let you create this kind of workspace, even in their free versions, with limits: Projects in Claude, Projects in ChatGPT and Gems in Gemini (Claude, ChatGPT, Gemini). The principle is the same, although the features vary from one tool to another: instead of rewriting the context in every conversation, you write it once, add the reference documents and set the rules.

Step-by-step example: a project to prepare your executive committee.

1. Context. Who you are, which company, who attends the committee, which topics come up again and again.
2. Documents. The standard agenda, the latest minutes, the indicators you track.
3. Goal. "Every month, prepare a committee outline for me: three priorities, risks, pending decisions and questions for each manager."
4. Limits. "Do not invent figures: if a piece of data is not in the documents, mark it as pending."
5. Use. Every month, you add the latest updates and ask for the outline. In a few minutes you have a first version to review.

The project instructions can look like this:

You are my assistant for preparing the monthly executive committee of [company]. Attendees: [roles]. Use only the project documents and what I tell you in each conversation. Always deliver: three priorities, risks, pending decisions and one question for each manager. If a piece of data is missing, write "pending"; never estimate it. Direct tone, one page maximum.

Before uploading documents to a project, check that you may share them and under what conditions: section 5 explains this.

An example of my own. Every week I use a configured project to prepare my LinkedIn posts:

• Context: a knowledge base with my voice (precise, results-oriented, challenging received ideas), my post formats and my topics.
• Reusable instructions: a few fixed commands: rewrite a text, turn a news item into a post, turn an idea into a post, look for new angles.
• Use: with a single instruction, for example "text based on this news item", I get a draft in my format.
• Limits: a mandatory final review that removes empty phrases, and no data without a verified source. For a post on usage-based AI pricing, I only used a quote from an industry executive after checking it in the original source.
• What I still do myself: I choose the topic, adjust the angle, rewrite the opening line and decide whether to publish. The project prepares; the post is mine.

4. Claude, ChatGPT or Gemini: what does each one bring?

All three tools work for an executive's job; the difference lies in how you use each one and in your company's environment.

Claude is at the center of my work. I use it to write, analyze long documents and work with projects configured by task. It offers projects, web search and connectors to other tools, including one for Microsoft 365, available on all plans with a work Microsoft account authorized by your company's administrator (source). With Claude Cowork and Claude Code, I also chain multi-step tasks on my files.
Choose it if your work is mostly writing and analyzing documents, and you want to organize it by project.

ChatGPT offers projects on all its plans, including projects shared with other people. Depending on the plan, it adds GPTs you can create and share, scheduled tasks and deep research for reports with sources (source).
Choose it if your company already uses it, or if you want to schedule recurring tasks and share projects with colleagues.

Gemini offers Gems and Deep Research from the free version, and Gemini Notebook (formerly NotebookLM) to work on your own sources. On paid plans from Google AI Pro upward, and in Google Workspace depending on your company's setup, it is integrated into Gmail and Docs (source).
Choose it if your company works with Google Workspace or your documents are already in Gmail and Drive: it is the closest to them.

I use all three every day and compare their answers on important tasks. The detailed comparison, with prices and a task-by-task table, is in Claude vs ChatGPT vs Gemini for executives and companies.

5. What should you not delegate to AI?

AI prepares, proposes and drafts; the executive decides, validates and signs.

Decisions. AI can rank options and arguments, and even point out the risks of each one. The decision, and the responsibility for it, are yours.

Sensitive data, before you have checked your account. What matters is not whether you pay, but the type of account and its settings:

• In personal Claude and ChatGPT accounts, free or paid, whether your conversations are used to train the models depends on an opt-out setting you need to check (Anthropic, OpenAI).
• In personal Gemini accounts, it depends on the "Keep Activity" setting: if it is on, Google may use your activity to train its models, with the help of human reviewers. Turning it off does not erase everything: new conversations are kept for up to 72 hours, and those already reviewed, as well as feedback you have sent, may be kept for up to three years (source).
• On business plans (Claude Team and Enterprise, ChatGPT Business and Enterprise, Google Workspace), the providers state that your data is not used to train the models by default (Anthropic, OpenAI, Google).

The fact that a provider does not use your data to train its models does not, on its own, guarantee that you may share it. If a document contains personal data, the GDPR requires, among other things, a legal basis, limiting the data to what is necessary, a data processing agreement when the provider acts as a processor and, if the data leaves the European Economic Area, safeguards for that transfer. And a document without personal data, such as a commercial strategy, may still be confidential. Before entering client information, check which plan you are on, how it is configured, and whether you may share that information and under what conditions. Anonymizing helps, but removing names is not always enough: for the AEPD (the Spanish data protection authority), data is only anonymous if there is no reasonable likelihood of identifying the person. The details, including what the GDPR and the EU AI Act require, are in Is it safe to use AI with company data?

Final validation. Figures, names, legal references, commitments: everything that leaves your desk, you have checked yourself.

A real case: while preparing content for this very website, an AI attributed to a circular from the Consejo General de la Abogacía Española (the representative body of the legal profession in Spain) obligations that the text does not contain. Read by a lawyer, that sentence would have ruined the page's credibility. We caught it because the rule was to check every claim against the official source.

6. How do you get started this week?

Start with a single repetitive task and configure it well, instead of trying ten things at once.

1. Choose a task you do every week that takes you at least an hour. Good candidates: preparing a recurring meeting, summarizing the reports you receive, drafting a type of email that keeps coming back.
2. Choose a tool: the one your company already has or, if there is none, the one you prefer. What matters is sticking with it for the whole trial.
3. Create a project with the context, documents and limits, as in the committee example. Before uploading documents, check that you may share them and under what conditions (section 5).
4. Use it for three weeks in a row and note what goes wrong: missing data, a tone that does not fit, a format you have to redo.
5. Adjust the context and rules based on your notes. Every adjustment stays saved in the project: you will not have to repeat it.

Once this first task works, move on to the next one with the same method.

If you are not sure where you are starting from, the AI skills test gives you your level in 3 minutes, free and without signing up.

7. Common mistakes when getting started

These mistakes are not technical but methodological, and they are fixed with the three reflexes from section 2.

Asking without context. A one-line request gets a generic answer. What to do: explain who you are, who the result is for, what you are looking for and what the constraints are.

Believing the first answer. A well-written text is not a correct text. What to do: check figures, names, quotes and dates at the source before using it.

Giving up after a bad result. A disappointing answer does not prove the tool is useless. What to do: rephrase with more context; if it still fails, try another tool or handle that task another way.

Testing ten tools at once. Switching tools every week prevents you from building a habit. What to do: choose one, configure it well and stick with it for a few weeks.

Copying prompts from the internet. Someone else's prompt knows neither your company nor your limits. What to do: use it as a starting point and turn it into a project with your own context.

Skipping planning and review. Asking straight away, and delivering without rereading. What to do: remember the rule: of the time you save on execution, you need to reinvest part in planning and reviewing.

8. Frequently asked questions

Want to apply it to your own work? The 1:1 AI training for executives is 3 sessions of 1.5 hours on your own files, at your own pace. Start with a free 30-minute intro call.

Are you a team of 3 or more? Discover AI training for teams.

Sources: Claude — plans and pricing, ChatGPT — pricing, ChatGPT — Projects, Gemini — subscriptions, Gemini — Gems, Claude — Microsoft 365 connector, Anthropic — terms for Free, Pro and Max accounts, AEPD — anonymization and pseudonymization and Gemini Apps Privacy Hub, accessed on 13 October 2026 · GDPR (Regulation (EU) 2016/679) · Anthropic — data use for training, ChatGPT Business and Generative AI in Google Workspace Privacy Hub, accessed on 9 and 10 October 2026.