AI DUNN Right Newsletter #16

What Matters Most

Marketing to machines just became your next job

Here's the shift nobody's talking about...

When someone asks ChatGPT "what's the best CRM for startups," they don't get ten blue links anymore. They get one recommendation.

If your brand ain't that recommendation, you don't exist.

This week, a Y Combinator startup called Unusual coined the term for what's happening...

AI Relations. It's public relations, but for AI models instead of journalists.

Here's what you need to understand...

AI models don't rank websites like Google does. They form opinions. If an AI thinks your brand is "niche" instead of "enterprise-ready," that perception shapes recommendations for millions of users.

Unusual helped one company shift their enterprise readiness score from 18 to 54 out of 100 in one month. Another platform identified gaps in what Perplexity knew about their compliance. They published targeted content. Within a month, those pages became top-3 citations.

Real infrastructure is being built right now:

• Adobe created LLM Optimizer to track how brands appear across ChatGPT, Google AI, and Perplexity

• Semrush built a database of 130 million prompts to monitor brand mentions

• Profound raised $20 million for AI perception analytics

• Harvard Business Review said, "Forget What You Know About Search. Optimize Your Brand for LLMs"

  
  

The new game tracks different metrics:

• How often your brand appears versus competitors in AI responses

• Whether models recommend you without prompting

• How AI positions your brand over time

• Which pages AI crawlers visit

Here's what matters for your business...

You can't buy your way into AI recommendations like you do with PPC ads. The AI forms opinions from sources you influence but don't control... Wikipedia, Reddit, G2 reviews, analyst reports, news coverage.

Start tracking what AI models say about your brand. Get covered in publications AI trusts. Build presence where AI learns. Monitor your Wikipedia entry. Create content that positions your brand the way you want AI to see it.

By 2027, AI Relations will be as standard as PR and SEO. Companies will run dedicated teams managing AI perception.

The SEO playbook is getting rewritten. And businesses figuring this out first will own the next decade.

---

What's New This Week

Gmail launches AI Inbox

Google rolled out AI Inbox for Gmail this week.

Two main features:

• "Suggested to-dos" shows priority emails needing action ("You have a bill due tomorrow")

• "Topics to catch up on" groups updates by category (purchases, finances, travel)
  

They also launched AI Overviews in search. Ask natural language questions like "who was the plumber that quoted my bathroom renovation" and get instant answers pulled from your emails.

Plus a Grammarly-style Proofread feature built directly into Gmail.

Rolling out to Google AI Pro and Ultra subscribers first.

Why this matters... Every major platform is building AI assistants that manage your workflow. Gmail processes billions of emails daily. If AI Inbox works, it becomes the default way people interact with email. The shift from reactive (checking what came in) to proactive (AI tells you what matters).

26 predictions that'll define AI in 2026

Peter Gostev published evidence-based predictions for where AI is headed this year.

Key takeaways:

• China isn't catching up anymore - They're competing head-to-head, faster and cheaper. DeepSeek trained GPT-4 performance for $5.5 million versus Western labs spending $100M+

• Multimodal convergence is coming - Text, video, audio, music, and speech merging into single models, but only where users actually pay today

• Agents are improving but autonomy is stalling - Enterprise buyers remain cautious due to reliability concerns and unclear ROI. 2025 was supposed to be the year of agents. It wasn't.
  

The reality check... AI agents still stuck in pilot mode. Patchy adoption, weak integration, and missing returns keep them from production deployment. Most enterprises won't see mainstream agent adoption until clear ROI shows up.

OpenAI launches ChatGPT Health

OpenAI announced ChatGPT Health. A dedicated space for health conversations, separate from regular chats.

Over 230 million people ask ChatGPT health questions every week. Now those conversations are siloed so your health context doesn't leak.

It integrates with Apple Health, Function, and MyFitnessPal. OpenAI promises they won't use Health conversations to train models.

The catch... OpenAI's own terms say it's "not intended for diagnosis or treatment." So they're launching a product for health conversations while legally disclaiming it shouldn't be used for medical advice.

What this actually is... useful for general health research ("What are the symptoms of X?" or "What questions should I ask my doctor?"). Not a replacement for medical advice. Starting point for research.

The danger... people will treat it as medical advice anyway.

---

Tool of the Week

AI-powered onboarding optimization workflow

Most SaaS teams watch new signups vanish before hitting their first milestone. This week's focus... using AI to spot where users drop off and redesign activation paths.

Here's the workflow:

Step 1: Pull your onboarding funnel metrics

• Activation rates by step (signup → first action → key milestone)

• Drop-off points (where users exit)

• Support ticket themes tied to setup

• Time-to-value averages
  

Export from Amplitude or Mixpanel as CSV, feed into ChatGPT or Claude.

Step 2: Ask AI to identify friction points

Sample prompt:

You are a product growth strategist analyzing onboarding data.

Using these funnel metrics:
- Step 1: Signup → 100% completion
- Step 2: Profile setup → 68% completion  
- Step 3: Connect integration → 41% completion
- Step 4: First action completed → 22% completion

Top support tickets during onboarding:
- "How do I connect my account?"
- "What happens after I click 'Next'?"
- "I don't see the button to proceed"

Identify:
1. The highest-friction step in this flow
2. Likely causes of drop-off at each stage
3. Recommendations to reduce friction

AI flags patterns you'd miss manually. Maybe Step 3 asks for credentials users don't have handy. Maybe Step 4 lacks clear feedback.

Step 3: Generate redesign alternatives

Ask AI to brainstorm ways to fix the friction:

• Three alternative ways to present stuck steps

• Tooltip or help text that reduces confusion

• Deferred setup option (let users skip and return later)

  
  

Step 4: Test persona-specific messaging

Different users need different nudges. Technical founders want speed. Marketing managers want templates. Sales teams want quick wins.

Use AI to write targeted messaging for each persona without building separate onboarding tracks.

Why this works:

• Spots friction faster than manual analysis

• Generates multiple redesign options in minutes

• Tailors messaging without added complexity

• Lets you test without waiting on engineering
  

The result... more users reach activation, fewer support tickets, better retention from day one.

---

Quick Hits Worth Your Time

→ UK Prime Minister says government "will take action" on Grok deepfakes. Governments grappling with flood of non-consensual AI-generated content on X.

→ Anthropic adds Allianz to growing enterprise wins. More Fortune 500 companies deploying Claude at scale.

→ OpenAI acquiring Convogo team. The executive coaching AI tool team joins OpenAI as talent acquisition continues.

→ 1 GW AI training clusters under construction. Meta's Prometheus and OpenAI's partnership with NVIDIA targeting late 2026 deployment. Models 5-10x larger than GPT-5 coming.

→ Computer-use agents still not mainstream. Despite Anthropic releasing Claude computer-use capabilities, adoption remains limited to technical users and pilot programs. Enterprise buyers cautious about reliability.

---

Prompt of the Week

AI Prompt Library Builder

Stop reinventing prompts every time. Build your library once, use it forever.

Act as an AI workflow specialist. Create one organized, reusable prompt library for [MARKETING FUNCTION] that eliminates repetitive prompting and saves hours daily.

Essential Details:
- Marketing Function: [CONTENT/EMAILS/ADS/SOCIAL/SALES/SUPPORT]
- AI Platform: [ChatGPT/Claude/Gemini/Multiple]
- Use Frequency: [DAILY/WEEKLY TASKS]
- Team Size: [WHO USES IT + SKILL LEVELS]
- Quality Standard: [OUTPUT REQUIREMENTS + BRAND VOICE]
- Storage Method: [Notion/Google Docs/Airtable]
- Current Pain Point: [WHAT'S BROKEN WITHOUT THIS]

Create one complete prompt library including:
1. Category organization structure (logical grouping by function)
2. 25 core prompt templates (covering all common marketing tasks)
3. Variable placeholder system ([BRAND], [PRODUCT], [AUDIENCE], [TONE])
4. Output quality checklist (what makes a prompt "good enough")
5. Version control method (track what works, retire what doesn't)
6. Team sharing protocol (who accesses what, how to contribute)
7. Quick-start guide (how new team members use the library)
8. Performance tracking (which prompts save the most time)

Output as ready-to-use prompt library with templates organized by category, accessible to entire team.

Why this works... Most people spend 30-40% of AI conversations just setting context. With a prompt library, that drops to near zero. You start every conversation already in context.

Best use case... Marketing teams running similar campaigns repeatedly, agencies serving multiple clients with similar needs, content teams publishing on consistent schedules.

---

My Take

This week showed me something that's been building quietly...

AI is shifting from "nice to have" to "you better figure this out or your competitors will."

The AI Relations concept ain't hype. It's infrastructure. When Adobe, Semrush, and Harvard Business Review all move in the same direction, that's signal, not noise.

For business owners, here's what matters:

The barrier to entry for AI just got lower (Gmail AI features going free, cheaper training costs from China). But the competitive advantage goes to people who understand the new rules.

Old game... rank on Google.New game... become the answer AI gives when asked.

Old game... pay for ads.New game... earn AI's trust through quality content in places AI learns.

My advice...

Start tracking what AI says about your brand this week. Ask ChatGPT, Claude, and Perplexity about your category. See who gets recommended. See if you're mentioned.

If you're not showing up, you've got work to do. Get coverage. Build presence. Create content AI can learn from.

The businesses that figure out AI Relations in 2026 will have distribution advantages for the next decade.

Everyone else will be paying premium prices to reach audiences that already moved on.

That's it for this week.

What are you doing to make sure AI recommends your business?

Hit reply and let me know.

Jackie