Time needed
2 hours
Difficulty
Advanced
Cost
Three tools: Check available plans
Tools
ChatGPT + Claude + Gemini
Check
July 20, 2026

When I actually use three AIs

I use three AIs only when the decision is important enough to merit multiple checks and when I have materials, sources and data to compare. I don't do it to write a caption or find ten titles. The cost I consider first is not the subscription: it is the time needed to keep briefs, versions and decisions consistent.

Using multiple tools does not eliminate bias. However, it allows you to separate the functions: research, criticism and planning. It is this division that produces value.

How to create a common dossier?

Prepare a source file with objective, KPI, audience, offer, constraints, historical data, decisions already made and glossary. Number the evidence: E01, E02, E03. Each AI must work on the same version.

Add a five-line employment contract:

  • do not invent data or sources;
  • separate facts, inferences and proposals;
  • cite the IDs of the evidence used;
  • point out what is missing;
  • do not change an approved decision without stating so.

What roles to assign?

Operational schemeThree AIs, one direction
  1. Gemini researchSources, signals and limits of evidence
  2. Human verificationHe opens the sources and approves the dossier
  3. Claude criticizesContradictions, risks and assumptions
  4. Human decidesHe chooses direction and sacrifices
  5. ChatGPT schedulesActions, responsibilities, KPIs and thresholds
Arrows represent approved steps, not automatic copies. Responsibility remains at the center.

This assignment is not a universal ranking. It is a working architecture. You can switch roles, but you don't have to collapse them into the same prompt.

What are the passage prompts?

Prompt for Gemini: Research Package

Prompt to useResearcher with sources
Decision to support: [decision].
Internal dossier: [paste version and ID of evidence].
Confini: [paese, segmento, periodo, fonti ammesse].

Create a research package. For each external signal indicate URL, date, scope, observed fact, limit and ID. Connect each signal to a decision question. If you cannot find an adequate primary source, state so. Don't propose the plan yet.

Why do I use this prompt: The first prompt broadens the field, but must deliver verifiable cards and not a premature strategy.

Before continuing, a person opens the links, removes what doesn't work and assigns the status "verified", "partial" or "discarded".

Prompt for Claude: stress test

Prompt to useHypothesis reviewer
Analyze the verified dossier below. Do not add sources.

For each possible direction identify:
- evidence for and against ID;
- more fragile assumption;
- required business capacity;
- opportunity cost;
- reputational or operational risk;
- minimal evidence to reduce uncertainty.

Then formulate 5 objections that a finance, sales and operations manager could raise. Don't choose the direction: prepare the decision.

Why do I use this prompt: The second prompt restricts: it puts pressure on the approved dossier and makes the price of each choice visible.

Prompt for ChatGPT: Constrained Plan

Prompt to useExecutor of the decision
DECISION APPROVED
[Paste direction, motivation and waivers]

EVIDENCE ADMITTED
[Paste verified IDs]

Create a 90-day plan. For each task indicate objective, responsible, deadline, dependency, metric and continue/fix/stop threshold. Enter a review every 30 days. Mark as “out of strategy” any activity that does not result from the approved decision. Don't introduce new channels or audiences.

Why do I use this prompt: The third prompt converges: it does not reopen the strategy, but makes an already recorded human choice executable.

How to stay in control?

The decision log is the bridge between the tools. It must contain date, decision, evidence used, alternatives discarded, responsible person and next review. Also save the version of the dossier and prompts: otherwise you won't know why two results differ.

What works

Advantages

  • Separate research, criticism and planning
  • Introduces checks before irreversible decisions
  • It makes disagreements and assumptions more readable
What to check

Limits

  • Requires versioning discipline
  • Increase coordination time and steps
  • Three mediocre outputs do not become a good strategy

To learn more about the individual roles, use the guides dedicated to ChatGPT, Claude and Gemini. The same architecture is also expressed in the Instagram strategy with three AI.

Frequently asked questions

Does it make sense to use three AIs for a single strategy?

Yes, if you assign different roles and keep a single brief. Three parallel answers without method only increases the noise.

What role should you assign to each AI?

In the proposed workflow Gemini collects evidence, Claude criticizes hypotheses and contradictions, ChatGPT organizes the plan. Roles can change, but they must remain distinct.

How to avoid three inconsistent versions?

It uses the same source dossier, a decision log, and a mandatory human shift between research, critique, and plan.