CONTACT
CONTACT
1) Formulate an effective request to the AI
Properly structuring your requests to AI is essential to obtaining accurate, useful, and directly actionable results. By clearly stating the objective, context, constraints, and expected format, you reduce ambiguity and save time. Here are the essentials to include in your requests to AI.
Key idea : a specific request = a useful response.
To do : state the context (who you are, for what audience, for what objective), the expected result (“5-point plan”, “150-word text”, “summary table”) and the rules (tone, length, format, standards).
Before sending : request a clear output format and, if necessary, sources or a quick test.
After receipt : Check that it is clear, accurate, and directly usable. If necessary, request a simpler, shorter version, or one with examples.
Biases & Limitations : Remember that AI can make mistakes or think with biases. Phrase your questions neutrally and ask for reservations if the answer is uncertain.
Sources & Dating : Ask the AI to cite its sources and provide a verification date.
Example :
“You're a B2B copywriter. Goal: 5-point newsletter plan for SMEs. Clear tone, 150–180 words. Add 1 CTA and 2 alternative headlines. Cite your sources and specify the date of verification.”
Mini-checklist :
Context
Result
Constraints
Output format
Sources/Tests
2) Start with the conversational assistant
Before using image, video, code, audio, or translation AI, first use a conversational assistant. It structures your ideas, clarifies the request, and produces clear instructions. You can then copy and paste these instructions into the specialized tool: greater quality, consistency, and speed.
At each small step, do a quick check: “Does this match the objective?” If not, adjust (shorter, different tone, different format) and relaunch.
Bias & Limitations : Ask the assistant to indicate areas of uncertainty and sensitive points.
4-step method :
Frame : context, audience, objective.
Specify : expected format (e.g. “16:9 poster”, “150-word text”, “5 KPI table”), tone, constraints (style, standards, technology).
Validate : the assistant reformulates the instruction, fills in the gaps, reports biases/limitations, adds sources + date.
Copy and paste : Paste the instruction into the targeted AI tool (image/video/code/translation), then adjust.
3) Check before publishing
It's crucial to double-check before publishing to avoid errors, maintain your credibility, and comply with legal requirements (data, rights, compliance). A quick review, sourced facts, and technical validations (links, media, layout) make all the difference. Here are the essential checks to perform before publishing anything online.
Goal : to avoid mistakes and strengthen credibility.
Check the facts (dates, numbers, names) on a reliable source (official website, documentation, recognized publication). For code, test a small example. For numbers, redo a simple calculation.
Bias & Limitations : Look for generalizations, stereotypes, or unsourced statements; add a statement of uncertainty if necessary.
Sources & Dating : Keep source links and verification date to facilitate updates.
Mini-cycle : first draft → proofreading (you, colleague or other AI) → final version.
Useful request to add :
“Point out what is uncertain, suggest two sources to verify it, and explain in two sentences how you obtained the result, including the date of verification.”
4) Reduce errors by crossing multiple AIs
Cross-referencing multiple AIs is a good idea to reduce errors, limit bias, and improve reliability. By having two or three models respond and then comparing their outputs (facts, figures, sources, consistency) before a synthesis or consensus vote, we identify discrepancies and consolidate the best answer—with a final human check on sensitive points if necessary. This is the essential approach for effectively cross-referencing AIs.
Purpose : to secure sensitive or ambiguous subjects.
Interview two different AIs and compare. If there are any discrepancies, ask one to review/correct the other's answer, then confirm key points with a trusted source.
Bias & limitations : Use cross-criticism to detect bias; document remaining uncertainties.
Sources & dating : require references and a verification date for each version in order to trace developments.
Role division : One AI writes, one fact-checks, one formats. If information remains uncertain, state it clearly rather than making random assertions.