AI Prompt Cookbook

🎯

Anatomy of a good prompt

Every strong prompt has four elements. You don't need all four every time, but the more context you give, the better the output.

Element	What it means	Example
Role / Context	Who you are and what you're working on	"I'm an Insights Manager preparing a debrief for an FMCG client in beverages"
Task	What you need done, specifically	"Write an executive summary of the key eye-tracking findings"
Format	How you want the output structured	"3 bullets max, business language, no methodology jargon"
Constraints	What to avoid or prioritize	"Client is risk-averse — frame findings as opportunities, not problems"

Template — copy and fill in:

"I'm [role] working on [project/task] for [client/purpose]. I need you to [specific action]. Please format the output as [format]. Focus on [specific aspects] and keep it [tone/length]."

🧠

The context window — what it is and why it matters

Think of the context window as the AI's working memory for your current session. Everything in this chat — your messages, its replies, uploaded files — lives inside a finite space. When it fills up, the oldest content starts falling out.

This is not long-term memory. It does not remember your conversation from yesterday. Every session starts fresh.

Situation	What to do
Switching topics entirely	Start a new chat — don't continue the old one
Long conversations	Re-state key facts mid-way: "As a reminder, the client is X and we're testing Y"
Sharing long documents	Upload the file — don't paste walls of raw text into the chat
Multiple projects in one chat	Number your sections: "New question, completely different project:"
Something discussed days ago	It won't remember. Paste the relevant part again in the new session.
Long chat you want to continue later	Ask the bot: "Summarize our full conversation and save it as an .md file." Paste that file at the start of your next session to pick up exactly where you left off.

💡

Tips & tricks — do's and don'ts

✅ Do

Be specific about your audience — client, C-suite, internal pod, Joris
Upload files when working with data, reports, or briefs
Ask for multiple versions: "formal" and "casual Slack version"
Iterate — "Make this shorter" / "Rewrite paragraph 2" / "Add a real example"
Reference EyeSee context: methodology, category, client type
Ask for structure first: "Outline the approach before writing anything"
Tell it the tone: consultative / direct / warm / executive
Use follow-up prompts to refine — version 1 is a draft, not a final

❌ Don't

Assume it knows internal project codes or client nicknames
Upload sensitive client data without checking your data policy
Ask it to guess — give it the context it needs to be useful
Stack multiple unrelated asks in one message
Stop at version 1 — always refine with a follow-up message
Expect it to remember details from a previous or older session
Accept vague output — push back and ask for something better

Quick prompt fixes:

❌ Weak	✅ Strong
"Help me with this project"	"I'm working on a pack test for a beverages client. Help me structure the methodology section of my proposal."
"Analyze this data" [uploads file]	"This is eye-tracking data from a shelf test with 6 SKUs. Identify which pack had the highest visibility and fastest first fixation."
"Update the Johnson report"	"I'm attaching the draft report for Johnson & Co's retail media study. Revise the exec summary — 30% shorter, lead with the business implication."
"Write an email + analyze data + make a deck"	"Let's do these separately. First: help me write the email. We'll do the analysis after."

⚡

Built-in skills — Agentic AI bot on Collab

The Agentic AI bot on Collab has specialized modes purpose-built for EyeSee. Activate them by naming what you want — they understand EyeSee's methodology, products, and context out of the box.

General Assistant

Default — just start typing

Drafting, writing, summarizing, planning, analysis. Works for everything not covered by a specific skill.

Translation

"Translate this to [language]"

Client reports, emails, survey content. Understands EyeSee terminology in context.

Pack Analysis

"Analyze this pack" + upload image

Upload a pack image and the bot evaluates it across universal clusters (quality, premium, modern, natural, effective…) and relevant conditional clusters (skin care, taste, eco…). Returns scores, strengths, gaps and actionable recommendations.

Market Research

"I need a market research summary on [topic]"

Competitive intel, category context, trend summaries for pitch preparation.

Copilot doesn't have skill modes — it works well as a general assistant and is particularly strong for document drafting, meeting summaries, and tasks within the Microsoft 365 ecosystem.

🔬

Advanced prompting techniques

Iterative refinement — don't stop at version 1

Follow up to improve. Each iteration gets closer to what you actually need.

"Make it more concise" / "Less corporate, more direct" / "Add a specific example from the beverages category"

Role-play — for stress-testing and perspective

Ask it to take a specific role. Great for anticipating client questions before the meeting.

"You are a skeptical client who just saw these results and doesn't trust the sample size. What questions would you ask?"

Chain-of-thought — for complex problems

Ask it to reason step by step before giving the final answer. More reliable on hard analytical tasks.

"Walk me through your logic step by step before giving me the final recommendation."

Perspective-taking — multi-stakeholder situations

Get three different viewpoints in one prompt. Useful before decisions that affect multiple teams.

"Analyze this from three angles: what the client cares about, what Joris would flag, and what legal would push back on."

Red-teaming — for important decisions

Ask it to argue against your plan. Surfaces blind spots before you commit.

"I'm about to send this proposal. Act as a harsh critic — what's weak about it and what am I missing?"

Template extraction — for repeatable work

Once you have a good output, extract it as a reusable template to save time on future similar tasks.

"Turn this into a reusable template I can fill in for future clients in the same category."

Constraint-based creativity — when options are limited

Use limitations to drive creative thinking rather than letting them block you.

"We need to [goal] but we can't [constraint]. Given this limitation, what are 5 creative alternative approaches?"

Negative prompting — tell it what NOT to do

Explicitly rule out unwanted patterns, words, or formats. More reliable than hoping the AI avoids them on its own — especially useful for tone, structure, and style control.

"Write this executive summary. Do NOT use bullet points, do NOT use the word 'leverage', and do NOT open with a question."