- Time needed
- 90 minutes
- Difficulty
- Intermediate
- Cost
- Can also be used without additional tools
- Tools
- ChatGPT + business data
- Check
- July 20, 2026
What do I prepare before the prompt
In my work, strategy doesn't start with "make me a digital plan". Start with the questions that the business can no longer put off. Before opening the conversation I collect an essential dossier:
- business objective and time horizon;
- supply, margins, commercial capacity and constraints;
- audiences already served and problems observed;
- active channels, budget and available skills;
- data from the last six or twelve months;
- decisions already made and aspects still open.
Always distinguish facts, hypotheses and preferences. If you write “Instagram is the best channel” as if it were a fact, the AI will tend to build the plan around that premise. If you present it as a hypothesis, he can question it.
What is the correct workflow?
- DossierFacts, data, constraints and open questions
- DiagnosisProblems, opportunities and missing information
- ChoicesAlternatives with explicit trade-offs
- SlowlyPriorities, responsibilities and times
- ReviewObjections, risks and decision thresholds
This sequence avoids two frequent mistakes: asking for ideas before having defined the problem and turning the first plausible answer into a decision. OpenAI recommends clear, specific prompts and an iterative process. Here the iteration is not used to "make the answer written better": it serves to check the reasoning one block at a time.
Which prompts to use and why are they different?
Prompt 1: Separate evidence and hypotheses
Analyze the following dossier to prepare a digital strategy.
OBJECTIVE
[Enter goal, metric and date]
DOSSIER
[Enter bid, audience, historical data, channels, budget and constraints]
Return a table with 5 columns:
1. available evidence;
2. what does it suggest;
3. hypothesis not to be confused with fact;
4. missing information;
5. how to check it.
Don't propose a strategy yet. If a piece of data is not present, write "not available". Close with the 7 questions that would most change the decision.Why do I use this prompt: This prompt is exploratory: it should highlight gaps and assumptions. It does not yet have to produce ideas or channels.
This prompt prevents the most dangerous shortcut: filling in the blanks with credible assumptions. The “don't propose yet” request keeps the conversation in the diagnosis phase.
Prompt 2: Compare real alternatives
Use only the dossier and the answers verified in the previous phase.
Propose 3 digital strategic directions that are incompatible with each other. For each indicate:
- priority audience;
- problem it solves;
- central promise;
- selected channels and excluded channels;
- necessary resources;
- main risk;
- signal observable within 30 days;
- condition which would make the direction to be discarded.
Rate each direction from 1 to 5 on impact, feasibility, speed of learning and consistency with the brand. Justify each vote with a sentence and don't invent benchmarks.Why do I use this prompt: Here the prompt becomes divergent: it asks for truly different alternatives and comparable criteria, not thirty disconnected ideas.
A strategy is an organized renunciation. Asking for incompatible options forces the model to make the choices visible, rather than adding up all channels and audiences.
Prompt 3: Convert choice to execution
The approved direction is: [paste chosen direction and human changes].
Make it a 90-day plan divided into three 30-day cycles.
For each cycle define:
- learning objective;
- maximum 3 initiatives;
- deliverables;
- responsible to be assigned;
- anticipatory metric;
- threshold continue / fix / stop;
- addictions and risks.
Add a final log with: decision, evidence, owner, review date. Do not add activities that do not follow the approved direction.Why do I use this prompt: This is a convergent prompt: it takes an already chosen direction and translates it into activities, responsibilities and control thresholds.
How to transform the output into a useful plan?
Don't copy the result into a presentation and consider it finished. Have a 30-minute review with whoever will be executing the plan. For each initiative ask: do we really have the data? Who decides? What activity do we stop doing? What signal authorizes us to change direction?
A useful plan contains a few clear relationships:
- a business objective linked to observable behavior;
- a public priority linked to a real problem;
- a channel chosen for a function, not because it "should be used";
- an experiment linked to a threshold;
- a person responsible for the next decision.
To expand the method you can compare it with the guide on how to set up a digital strategy with Claude or see the workflow with ChatGPT, Claude and Gemini.
What checks do you do before approving?
Advantages
- It makes assumptions and missing information explicit
- Compare alternatives with the same criteria
- Turn choices into a controllable plan
Limits
- It can make a flawed premise believable
- It does not automatically know internal or updated data
- It does not replace responsibility, sources and feasibility
Check numbers, quotes and proper names. Eliminate every sentence that doesn't change a decision. If two actions have the same goal but no priority, the plan is not yet a strategy. Finally, ask a person external to the work to summarize the plan: if they cannot say what you will and will not do, there is a lack of clarity.
Frequently asked questions
Can ChatGPT create a complete digital strategy?
It can organize information, surface hypotheses and transform already motivated decisions into a plan. Objectives, data, priorities and final approval must remain human.
Do you need a single, very long prompt?
No. A short chain of prompts separates diagnosis, choices, plan, and review, making it easier to check for errors and assumptions.
What data must be provided to ChatGPT?
At least objectives, audience, offer, constraints, available channels, historical data and criteria with which you will evaluate the result.