How to Talk to AI: Think Colleague, Not Search Bar

The Same AI, Different Results

We recently interviewed 34 Wissly users. A clear pattern emerged.

"Q1 marketing strategy" "ESG policy updates" "New hire onboarding"

Most people were typing queries like this—short keyword phrases, exactly how you'd search on Google.

The feedback? "I'm not getting what I need." "The search results aren't great."

Here's the thing: it's not an AI problem. It's a communication problem.

Keywords vs. Conversation

Imagine you're asking a new team member to help with something.

❌ You wouldn't say:

"Q1 marketing strategy"

✅ You'd say:

"I have a leadership presentation next week and need to pull together our Q1 marketing performance. Can you summarize the results from our three main campaigns—focus on the key metrics and what we could improve—and keep it to one page?"

See the difference?

AI works the same way. Toss it a keyword, and it's left guessing what you actually want. Give it context, purpose, and a clear picture of the output, and you'll get something far more useful.

Four Principles for Better Prompts

Drawing from recent prompt engineering research out of Imperial College London (Romanov & Niederer, 2025), here are four principles that anyone can apply right away.

1. Lead with Context

AI doesn't know your situation. Tell it what team you're on, why you need this, and who's going to see it.

2. Describe What You Want Back

"Summarize this" is vague. "Give me three key takeaways" is specific. "Analyze this" is vague. "Compare these in a table" is specific.

3. Show, Don't Just Tell (Few-shot Prompting)

Research shows that giving just 2-3 examples dramatically improves output quality.

4. Break Complex Tasks into Steps

Don't ask for everything at once. Studies show that going back and forth with follow-up requests actually decreases accuracy. Instead, lay out the steps upfront.

Real-World Examples: Before & After

Wissly automatically learns from all the documents in your connected folders—no need to attach files every time. You can also tag specific documents or folders to narrow the search scope.

Here's how to make the most of it.

Meeting Notes

Before:

"Meeting notes summary"

After:

"@Product Team Meetings I need to share today's sprint meeting notes in our dev Slack channel. Break it down into:

• What we completed this sprint

• What's planned for next sprint

• Any blockers Keep the technical terms as-is and include names for each action item."

Data Analysis

Before:

"Sales update"

After:

"I'm on the sales planning team. Using @Q1 2024 Sales Report, help me prep for next week's strategy meeting.

1. Top 5 products by YoY growth

2. Products with declining sales but high inventory

3. Products with the biggest regional variance Round everything to the nearest million."

Drafting Documents

Before:

"Write a proposal"

After:

"Draft a service proposal for a B2B prospect.

Background: They're a 500-person manufacturing company struggling with document management. Pull from @Service Overview Folder for our key selling points.

Structure:

1. Brief greeting (2-3 sentences)

2. Acknowledge their pain points (one paragraph)

3. Our solution (three core features)

4. Expected outcomes

5. Suggested next steps

Tone: Professional but warm. About two pages."

Connecting the Dots Across Documents

Before:

"Project status"

After:

"Look through everything in @Project Alpha Folder and summarize where we stand. Specifically, compare the milestones in @Project Plan.docx against what's in the last three entries in @Weekly Updates to show actual progress vs. plan."

Common Pitfalls to Avoid

⚠️ Say What You Want, Not What You Don't

Research shows AI follows positive instructions ("do this") better than negative ones ("don't do that").

⚠️ When Conversations Drag On, Start Fresh

The longer a conversation goes, the more AI tends to "forget" earlier context. Researchers call this "multi-turn performance degradation."

If you're five or six exchanges in and the responses are getting worse, start a new chat and re-establish the context from scratch.

⚠️ Don't Take AI Output at Face Value

AI sometimes generates confident-sounding but incorrect information (hallucination). Watch out especially for:

• Specific numbers or dates

• Quotes and citations

• Recent events or changes

Always double-check these.

⚠️ Narrow the Scope

Since Wissly learns from all your connected documents, vague questions might pull answers from irrelevant files. When asking about a specific project or time period, tag the relevant folder or document to focus the search.

The Mindset Shift: AI as a Teammate

A search engine is a tool that finds information. AI is more like a colleague you work with.

When you delegate to a colleague, you:

• Explain why this task matters

• Describe what you need back

• Point them to where to find what they need

The better AI understands your intent, the better results you'll get.

Final Thoughts: The Rise of Vibe Working

You may have heard of "vibe coding"—developers building software through natural conversation with AI.

We believe we're entering the era of Vibe Working: collaborating with AI as naturally as you would with a coworker, offloading repetitive tasks so you can focus on the work that actually matters.

Writing good prompts isn't some technical skill reserved for engineers. Just ask yourself: "How would I explain this to a smart new hire sitting next to me?"

That's all it takes.

The Wissly team is building your AI-powered work partner. Questions? Get in touch.