William Demoraes

AI chat / Content design

2026 / Case study draft

AI chat entry point for Ads Manager

I helped turn static mid-flight recommendations into an AI chat entry point that explained why guidance appeared, what signals supported it, and what an advertiser could do next.

Visual mockups

The entry point had to earn the click.

Recreated product visuals showing the entry point, the answer structure, and the prompt/spec thinking behind the experience.

Mid-flight guidance

Recommendation card

Prevent budget-capped delivery

Static guidance

Current spend

$4.8K

Delivery pace

Early cap

Confidence

Medium

Signals behind the guidance

7 day view
Daily budget reached3:14 PM avg.
High-converting hours missed2.6 hrs
Audience response stabilitysteady

AI entry point

Ask why this budget change is recommended

Turns static guidance into a conversation about the signals, tradeoffs, and next step.

Response anatomy

A useful answer had to explain the recommendation.

1

Name the trigger

Which mid-flight signal caused the recommendation to appear.

2

Personalize the why

How the signal applies to this campaign, audience, and objective.

3

Give one next move

A realistic action, with caveats when confidence is conditional.

Prompt spec sample

Static guidance became reusable chat instructions.

Marketing science

Explain budget guidance through pacing, missed delivery opportunity, and campaign objective.

Conversation rule

Start with the observable signal. Use could when the recommendation is conditional.

Engineering hook

Requires recommendation type, campaign objective, spend pacing, active date range, and confidence state.

Overview

What changed

This case frames AI chat as a content-design system, not a novelty layer. The work connected entry-point language, response structure, prompt guidance, and quality review so static recommendations could become explainable product guidance.

Role

My part

  • Defined entry-point language for mid-flight recommendation cards so advertisers knew what the AI chat could explain.
  • Translated marketing science expert input into prompt guidance, example responses, and user-facing explanation patterns.
  • Partnered with product design and engineering on recommendation triggers, chat context, response quality, and implementation details.

Problem

The product moment was unclear

Ads Manager could surface mid-flight guidance while a campaign was running, but the recommendation itself was static. The copy could say what to do, but it could not open up why the recommendation mattered, which signals informed it, how it applied to that advertiser's setup, or what tradeoffs to consider before acting.

Constraints

Rules of the work

  • AI responses needed to personalize the explanation without overstating certainty or sounding like unsupported marketing advice.
  • Marketing science guidance had to be translated into language advertisers could act on while a campaign was already in flight.
  • Engineering needed prompt patterns, recommendation context, response states, and guardrails that could scale across campaign types.

Users

Who needed clarity

  • Advertisers reviewing active campaigns and deciding whether to act on a recommendation.
  • Marketing science experts whose guidance needed to survive translation into product and AI behavior.
  • Engineers and product partners connecting recommendation data to the chat prompt, response structure, and quality checks.

Before / after

Examples in context, with the reason for each change

AI chat entry point

Entry point shown on a mid-flight recommendation card.

Before

Increase budget to improve results.

After

Ask why this budget change is recommended

Chat response

AI response explaining the recommendation in context.

Before

Increasing your budget may improve results.

After

Your campaign is reaching its daily budget before its highest-converting hours. Increasing it could keep delivery active when this audience is more likely to respond.

Content decisions

The writing system underneath

Move from recommendation to explanation

The entry point could not feel like a generic assistant pasted onto the page. It had to extend the recommendation by making the reason behind it available.

Ask why this budget change is recommended

Design responses around evidence, not authority

The chat response had to show its work: what signal the system noticed, why it mattered for this campaign, and what the advertiser could do next.

Your campaign is pacing out around 3 PM, before the hours that usually drive the most conversions.

Turn expert heuristics into promptable patterns

Marketing science expertise became reusable guidance for prompts, examples, and quality review instead of sitting in meeting notes.

When recommending a budget change, name the observed signal, explain the likely tradeoff, and offer one decision path.

Process

How I got there

  • Mapped mid-flight recommendation cards by trigger, advertiser question, data context, and likely follow-up.
  • Worked with marketing science experts to capture the reasoning behind budget, delivery, audience, and creative recommendations.
  • Wrote conversational patterns for entry prompts, answer structure, follow-up suggestions, and transparent caveats.
  • Partnered with engineers to align recommendation context, prompt inputs, response states, and quality checks.

Outcome

Impact signals

  • Defined a repeatable pattern for turning static product guidance into explainable AI entry points.
  • Created response standards for signal, rationale, tradeoff, caveat, and next action.
  • Impact to add: advertiser comprehension, chat engagement, and recommendation follow-through.

Learnings

What I would carry forward

  • The best AI entry point starts with the user's next question, not the product's feature label.
  • Transparency is a writing problem and a systems problem.
  • Static guidance becomes more valuable when users can inspect the reasoning before they act.

Next project

Content design system for an enterprise app suite

Next case study

Back to all work