New ChatGPT & Claude Break Your Prompts
Top Insights
GPT-5.5 and Claude 4.7 Just Broke Your Prompts in Opposite Directions
AI marketer Alex Prompter pulled apart the latest prompting guides OpenAI and Anthropic. The conclusion? Your old prompts probably now produce worse output than they did a month ago. And the two labs got there from opposite directions.
💡 Key Insights:
- Claude Opus 4.7 stopped covering for vague writing. It does what you type, not what it guesses you meant. Prompts that worked on 4.6 now hand back narrow, literal results because the model quit compensating for sloppy thinking.
- GPT-5.5 punishes the opposite mistake. OpenAI's guide says it outright...don't carry instructions over from older prompt stacks. Old models needed hand-holding, so legacy prompts over-specify, and that extra detail now muddies the output instead of guiding it.
- The skill quietly moved upstream. The two minutes of structured thinking BEFORE you type now decide what comes back. Both models reward clarity at the source, so the bottleneck moved from the model to the person at the keyboard.
Puma's AI Head: Let Customers Build With You
Puma's head of emerging marketing tech, Ivan Dashkov, walked Modern Retail through the brand's AI playbook during a €643.6 million loss year. The strategy hinges on one principle: hand consumers the tools and let them build the product with you.
💡 Key Insights:
- Co-creation is the spine of Dashkov's AI strategy. The AI Creator drew 54,000 fans who designed 180,000 Manchester City kits in 10 days. The winning design becomes the team's official third kit this season...the first AI-created jersey on a Premier League pitch.
- Friction nearly killed it. Initial renders took two to five minutes, so Puma worked with Google and Nvidia to cut that to 30 seconds. They filled even the 30-second wait with a Tinder-style voting feature, which produced 1.7 million ratings.
- Puma's new Vegas AI concierge pulled the strangest compliment from an introverted tester. Dylan, a 7-foot digital human that speaks 100+ languages and pings live inventory, beat the human associate option. AI in retail removes friction some shoppers actively wanted gone.
🚀 WATCH: How These AI Copy Bots Are Producing World-Class Sales Copy 50X Faster Than Even The "BEST" Copywriters On The Market…
(Plus… They Don't Get Sick, Miss Deadlines, Or Ask For Raises Either!)
Watch the full AI Copywriting Tell-All Video Here
Prompt of the Week
This Week's Feature: The Trojan Question
Prospects lie on discovery calls. Not on purpose...they just don't tell you the real budget, the real decision-maker, or how urgent this actually is.
Ask them straight up and they get cagey. Or worse, they tell you what they think you want to hear.
Trojan Questions get around that. They bury the real ask inside a forward-looking scenario, so the answer reveals what's actually true whether the prospect agrees or pushes back. This prompt builds three of them, custom to your offer and call stage.
THE FULL PROMPT:
You are an expert B2B sales strategist trained in discovery calls, MEDDIC/MEDDPICC, Sandler, SPIN selling, NEPQ, and calibrated questioning.
Your task: design Trojan Questions — single discovery-call questions that ethically surface real budget, urgency, and decision authority without sounding scripted, manipulative, or interrogative.
## USER INPUTS
Offer: [WHAT YOU SELL]
Price point / range: [PRICE]
Target buyer: [ROLE / COMPANY TYPE]
Sales call stage: [FIRST DISCOVERY / DEEP DISCOVERY / PROPOSAL / RENEWAL]
Suspected hidden issue: [BUDGET / AUTHORITY / URGENCY / ALL THREE]
Current prospect context: [WHAT THEY SAID, WANT, OR CLAIM]
Desired timeline or implementation window: [TIMELINE]
Prospect sophistication level: [LOW / MEDIUM / HIGH]
Tone preference: [CONSULTATIVE / DIRECT / WARM / EXECUTIVE / LOW-PRESSURE]
Anything to avoid: [WORDS, CLAIMS, PRESSURE TACTICS, INDUSTRY CONSTRAINTS]
## TASK
Create 3 candidate Trojan Questions.
Each question must combine:
1. A presupposition that reveals truth whether accepted or corrected.
2. Forward projection into a specific future scenario.
3. A calibrated open-ended structure using “what,” “how,” or “walk me through.”
Each question should surface:
- Budget reality
- Decision authority / buying committee
- Urgency / timeline pressure
Do not ask blunt questions like:
- “What’s your budget?”
- “Who’s the decision maker?”
- “When do you need this?”
## OUTPUT FORMAT
For each candidate, provide:
### Candidate [#]: [Question Name]
**Trojan Question:**
[One natural, call-ready question]
**What it presupposes:**
[The embedded assumption]
**What it reveals:**
- Budget:
- Authority:
- Urgency:
**Branch Tree:**
Show 4–5 likely prospect answers. For each branch, include:
- What their answer likely reveals
- The best follow-up question to ask next
**Trap-Detection Check:**
Rate from 1–10:
- Naturalness:
- Pressure level:
- Curiosity vs. manipulation:
- Deal-risk:
Then revise the question if it sounds scripted, clever, cornering, or overly tactical.
## QUALITY RULES
The final questions must sound like genuine problem-solving in real time. Use plain spoken language. Avoid jargon, salesy phrasing, interrogation energy, or obvious “gotcha” framing.
End with your recommended best question and explain why it is strongest for this situation.
Tool to Try
Chronicle is an AI presentation platform built for teams that need polished, on-brand decks for high-stakes business moments — pitches, board meetings, client proposals, and beyond. Designed by ex-McKinsey, BCG, and Apple operators, it pairs AI generation with serious design taste.
AI Tool Highlights:
📥 Start From Whatever You've Got: Paste raw notes, meeting docs, outlines, or existing decks. Chronicle pulls scattered inputs into a single, structured presentation without forcing you to start from a blank slide.
✨ One-Click Polished Slides: Go from rough ideas to professional decks instantly. Chronicle generates on-brand layouts, diagrams, and visuals that look hand-crafted, not templated — ready for executive audiences out of the box.
🎨 Freeform Canvas, Not a Cage: Unlike rigid AI generators, Chronicle is built on a freeform canvas. Instruct the AI to refine, then take full manual control over every layout and visual element.
🏷️ Brand Guidelines in a Click: Set your fonts, colors, and visual rules once. Every deck across every team stays consistent automatically — no more design drift between salespeople, marketers, or regional offices.
📊 Charts, Data, and 100s of Embeds: Visualize metrics, tables, graphs, and live data natively. Embed hundreds of tools you already use, so decks become interactive surfaces instead of static screenshots.
📤 Ship It in Any Format: Present live with audience-guiding controls, export clean to PPT or PDF, or publish as a website — one deck, every channel your stakeholders actually consume.
One More Need-to-Know News Story
X Rebuilds Its Entire Ad Platform After 20 Years
Tearing down a working revenue engine is the kind of move most companies wouldn't survive a board meeting proposing. X is doing it anyway. Twenty years of advertising infrastructure, demolished.
The new Ads Manager started rolling out in phases this month, built on AI retrieval and ranking systems that read user behavior in real time and match ads to whatever's happening on the platform that second. Monique Pintarelli, who runs global advertising at xAI, said:
'We are designing this new ad stack to enable more rapid and seamless integration of ongoing innovation. Advertisers can expect a smooth delivery of continuous improvements and a regular drop of new features as we keep pushing the platform forward.'
Mind Fodder
Thanks for reading.
Until next time!
The AI Marketers
P.S. Help shape the future of this newsletter – take a short 2-minute survey so we can deliver even better AI marketing insights, prompts, and tools.
[Take Survey Here]
