Who builds the content that AI engines trust?

I’m Daniel Spence Agnew, an AI prompt designer and UX content strategist based in London. I’ve designed prompts for Google Gemini, built GEO playbooks for Namecheap and Spaceship, and created persona research decks and Claude Projects that content teams now run on.

That’s how your brand gets cited by ChatGPT, Gemini, Perplexity, Claude, Copilot, and surfaced in Google AI Overviews.

You’re welcome to review the work below.

Table of contents

1. Google: Text prompts for business and education
2. Google: Nano Banana Pro image prompts
3. Google: Search trend infographics
4. Namecheap and Spaceship: GEO playbook
5. Namecheap and Spaceship: AI platforms comparison table
6. Namecheap and Spaceship: FastVPN personas and canonical prompts
7. Namecheap and Spaceship: Claude Projects (Nib and Volt)
8. Namecheap and Spaceship: UX landing pages and blogs
9. JustAnswer: Conversational analysis with Whisper
10. Let’s work together

What does prompt design for Google Gemini look like?

Google hired me through YunoJuno to design prompts for Gemini, targeting Gen Z and EMEA audiences across multiple verticals and showcasing the platform’s creative power to journalists, PR teams, and retail users

Every prompt had to give PR teams something quotable on first read.

How did I design prompts for Gemini’s business and education verticals?

Following Google’s persona, task, context, and format methodology, I designed prompt packs for UK-based businesses across coffee, tea, food-waste apps, and artisan bakeries.

I also designed education prompts for a local English school, helping Year 5 teachers explain physics, geography, and orienteering through lesson plans and field trips.

💬I’m a Year 5 teacher. Summarise how the four seasons affect change in Nonsuch Park for my 10-year-old pupils in a detailed comparison table sheet and a 10-question quiz about seasonal changes in the park. Identify and describe the key visual changes in flora (including specific tree species and flowering plants), fauna (focusing on birds, insects, and small mammals), and overall landscape appearance. Include an interactive month-by-month timeline of the year in Nonsuch Park. 

By late 2025, Google brought me back. This time for image prompts on Nano Banana Pro.

What is Nano Banana Pro, and how did I design image prompts for it?

Nano Banana Pro is a studio-level image generation and editing model on Gemini. With this tool, I art-directed prompts across pop culture, sport, food, and travel, specifying camera angles, lighting, and strict ‘no logos or readable text’ guardrails.

The Year in Search project celebrated Google Search trends from 2025 for a Gen Z EMEA audience.

Nano Banana requires scene-setting and rigorous iteration to get right. These prompts often involved transforming a reference portrait into AI-generated scenes. 

The example below is an adapted version of the original using a licensed stock image (SolStock) as the portrait source, not myself.

💬 Create a messy, photo-realistic image of me in Lisbon holding a sign. Keep my face realistic. I’m standing at the famous miradouro viewpoint in Graca, which lies above the city, with the castelo, Tejo river, and terracotta rooftops in the distance. I’m holding a large cardboard sign at chest height facing the camera, with a weather worn card saying ‘Year in Search 2025’ in Google Sans font.

Around me, show classic Lisbon details that aren’t perfect: patterned tiles underfoot, a slightly scuffed stone wall, a takeaway coffee cup on the ground, a tote bag slumped by my feet and a few other tourists in the background leaning on the railings, slightly blurred. Use warm, golden hour light with a bit of haze, soft shadows and fine grain so it feels like a real 2025 travel snap, not a postcard. No logos anywhere in the image, including on the sign.

Note: Nano Banana Pro image uses a base UK male image (SolStock / ID: 1328506141) licensed from SolStock for illustrative purposes.

Building infographic posters using local languages and search data

I’ve also built customised infographics on Nano Banana Pro using Google Search trends. Each template was localised per market, with placeholders for language, location, and cultural detail. Below is an adapted version set in Victoria Park, London.

💬 Make a social-first micro-guide poster for the search result “How to get good at pull-ups fast”. Set it at a neighbourhood sports area in Victoria Park, London on a fresh spring evening with soft light bursting through new leaves. Keep it believable, lived-in, and quietly aspirational.

Use a 4:5 bento layout with EXACTLY 8 tiles, phone-readable, short labels only (no paragraphs): 

Title, Start here, Step 1–4, Form cues, 4-week tracker.

Beginner targets: Dead hangs 3×20–30 seconds. Scapular pulls 3×8–10. Band-assisted pull-ups 3×4–8. Slow negatives 3×3 with a 5–10 second lower. Tracker line: train 2–3 days a week and leave 48–72 hours between sessions.

Ground the scene in everyday local realism: familiar public-space details, typical materials underfoot and around the frame, and seasonal light and weather that match the place. Use English for all on-image text.

Style: Clean bento training dashboard with subtle tactile texture: crisp grid, bold condensed headings, simple icons, light paper grain, minimal tape/marker arrows, mostly neutral palette with one punchy accent.

Avoid logos, watermarks, long text, gibberish type, distorted bodies/hands, and any non-local spelling or measurements.

How do you optimise content for AI platforms for two global brands?

As a senior copywriter for Namecheap and Spaceship, I build content systems and guides for AI-powered search, including GEO playbooks, persona and canonical prompt libraries, and Claude Projects that augment brand guidelines with data-driven research, analytics, and interactive tools.

What is the Namecheap and Spaceship GEO playbook?

Namecheap and Spaceship are popular domain registrars with over 11 million customers.

Generative-Engine Optimisation (GEO) involves shaping every content headline, sentence and statistic so AI platforms can parse it, credit Namecheap or Spaceship as the source, and influence customer behaviour before they arrive.

I wrote the GEO playbook for Namecheap and Spaceship. It explains how AI platforms choose which content to cite and gives writers a reliable method for earning those citations.

Unlike traditional search, AI platforms reason and provide people with personalised, multi-source answers by parsing and synthesising content from authoritative brands and news sources.

I distilled GEO into four repeatable cues with the following recommendations:

1. Lead landing pages with a clear 2-to-3-sentence “quick answer” that resolves the H1 question (~35–70 words)
2. Craft H1 headers as a precise, searchable question
3. Prove every blog-based claim you make with links to authoritative sources after each quotable statistic
4. Flag your quick answer with JSON speakable markup so voice assistants can read it aloud

Using these cues, I outlined in detail why optimising for AI platforms matters, the type of content they favour, and a “quick win” for writers.  

AI platforms comparison table (April 2026)

Note: Grok (xAI) is the fastest-growing AI platform by market share but operates primarily within X’s ecosystem. As its search and citation behaviours mature, it may warrant dedicated GEO attention.

The GEO playbook covers how to structure content for AI platforms. The next question was who we’re writing it for.

How did I build evidence-based personas for FastVPN?

FastVPN is a budget-friendly virtual private network (VPN) service, available to download on Namecheap and Spaceship. 

Most people accept that FastVPN users are motivated by price. I wanted to go further and build evidence-based personas that pinpoint who our customers are, where they live, and what frustrates them. So every content decision is backed by research, not opinions.

When someone challenges a content decision, you don’t defend a feeling. You point to your research.

Identifying FastVPN’s core users

Namecheap’s audience is 67% male, with 25-34 as the dominant age bracket. Leading traffic countries are the US, India, and Pakistan. (Similarweb, December 2025)

• The global VPN market tells a similar story with 54% male.
• Nearly 40% of Americans aged 18-29 use a VPN
• Over half of VPN users globally use free services only. Price isn’t just a factor, it’s the deciding factor

With this information, I used ChatGPT’s deep research feature and Perplexity Pro to trawl forums, reviews, and community posts, surfacing user language, pain points, and behavioural patterns for “FastVPN” usage. 

It revealed that FastVPN users are:

• Mostly men in their 20s and 30s
• They are not wealthy. Lots of students, junior office workers, call centre staff, and low-wage agency workers
• A smaller group of remote workers, who could pay more for Nord or ExpressVPN, but choose FastVPN because it’s “good enough”

Creating FastVPN personas with Claude

I created four Google Sheet templates covering a 90-day window (Sept to Dec ’25) and populated them with customer service data:

• Google Search Console (GSC) queries
• Namecheap live chat transcripts
• Social media and app store reviews
• Internal search queries and FAQs

Sharing this information and the deep research results, I prompted Claude to generate five customer personas for FastVPN, with a percentage weighting and demographic profile. 

FastVPN persona example: Jordan Miller

US budget buyer. ~30% segment of FastVPN’s personas

• Male, 29, Ohio. Blue collar field technician for a cable ISP. Lower-middle class.
• Owns an Android phone, cheap Windows laptop, and a shared smart TV.
• He wants one cheap VPN for the whole household.
• Streams Netflix shows and live sports. 
• Doesn’t care about protocols or features. Cares about price, whether it works, and the renewal costs.
• Speaks directly, swears online, and spells poorly. He wants simple deals.

Building a canonical prompt library

While personas define who your customers are, canonical prompts reveal what they ask.

Using the personas and Google Sheet datasets, I asked Claude to generate fictional but evidence-based queries, anticipating the problems each persona would ask an AI platform to solve, written as they would type them in real life. 

This project became the FastVPN canonical prompt library, covering three prompt types:

• Generic (no brand mentioned)
• Branded (FastVPN by name)
• Troubleshooting (something’s broken) 

What Jordan Miller types into ChatGPT

Once you know who Jordan is (age, income, devices, frustrations, and how he talks) you can build the prompts he’d write.

• Generic: “whats the cheapest vpn that works for streaming”
• Branded: “is fastvpn any good or is it one of those ones that jacks the price up after year one”
• Troubleshooting: “fastvpn keeps buffering on my fire stick. is it the server or do i need to change something”

None of these would show up in keyword research. With a persona and canonical prompt library, I understand who I’m writing for and what they’re asking for online. Together, they provide me with editorial authority backed by evidence.

What are Claude Projects, and how do they improve your content standards?

Claude Projects are specialist workspaces with chat histories that let you host brand guidelines, documents, and customised instructions in one place. Every chat that takes place inside a Project inherits this context. 

I’ve built two Claude Projects for Namecheap and Spaceship to enhance the quality and depth of our UX and social content. 

Nib is a UX/UI microcopy specialist with a structured knowledge base and customised instructions, covering everything from form-field labels and mobile push notifications to CTA buttons and in-app navigation. Every piece of microcopy it produces or checks is measured against the same high standards.

Volt handles social media creation and auditing for Spaceship. It produces platform-specific copy that adapts to each channel’s constraints, from Instagram’s character limits and carousel formats to X’s threading mechanics and YouTube’s metadata fields. When platforms update their specs or cultural trends shift, the knowledge base will be updated accordingly.

With Nib and Volt, you get consistent, brand-approved work done without re-explaining the brief every time. Both of them are becoming more sophisticated as their knowledge bases expand.

What UX and marketing copy do I produce day to day?

For Namecheap, Spaceship, and FastVPN, I produce copy that appeals to both new and existing customers. 

I also write and edit UX landing pages, CRM emails, newsletters, and blogs across all three brands.

You can review the landing pages I’ve produced:

• FastVPN for gaming (Namecheap)

• FastVPN for Spaceship

• FastVPN for Windows (Spaceship)

• IP Lookup Tool (Namecheap)

Alongside desktop and app content, newsletters, and CRM emails, I’ve also written blogs for Namecheap covering attention economics, internet freedom, Web3, and digital wills.

• The attention economy (Namecheap blog)

• Web 3.0 and the metaverse (Namecheap blog)

• Inflection point: protect your internet freedom (Namecheap blog)

• What happens to your websites when you die? (Namecheap blog)

How can deep learning speech tools improve customer service conversations?

At JustAnswer, the world’s largest Q&A expert marketplace, I led a conversational analysis experiment using OpenAI’s Whisper and ChatGPT to analyse expert-customer phone calls. 

Using Professor Elizabeth Stokoe’s “conversational racetrack” metaphor, I studied the emotional hurdles within these calls. Stokoe framework treats conversations as structured paths, similar to a racetrack, where friction and misunderstandings can be identified at specific points. These typically involve:.

• Curves (unexpected challenges)
• Straightaways (smooth, direct exchanges)
• Checkpoints (crucial information points)
• Pit stops (pauses for clarification or detours)

The analysis pinpointed where experts excelled at emotional intelligence and where adaptability under pressure needed improvement, providing evidence-based coaching insights that could scale across the entire expert network.

Let’s work together

Every piece of work on this page started with the same question. What does the audience actually need, and how do we prove it? If you need someone who can research audiences, build content systems, and write the copy that earns AI citations, I’d welcome the conversation.