Testing LLM Models to Improve my Launch Marketing AI Agent

Right now I’m in the lab writing up my plan for my next AI agent: Launch Marketer. This agent is going to help me automatically market web directories prospected and made by my Prospector and Build agents.

I’ll publish my full plan in the next post, first I wanted to share the test I did; I put my own launch marketing playbook through the 3 big AI models and challenged them to come up with additional tactics.

The Marketing Test

The Input: My Launch Marketing Playbook

At this point I’ve sold a fair amount of software online. From these various entrepreneurial endeavours I’ve built up a playbook on how to launch a website which used to work phenomenally.

Now, after Covid pushed many into side-hustles, more engineers came online, LLMs absorbed a quadrillion GPUs, and then started pumping out slop and automations – at least half of my launch strategies don’t work anymore 🥵.

But that’s how it goes.

I took that big sheet I have and stripped all of the stuff which no longer works.

Then I took out stuff not relevant to directory websites.

Then I took out stuff which is not yet automatable.

There wasn’t much left, but it was legit. (Read the other post where I share all this.)

The Prompt: AI, Help me Launch Marketing!

So I had my stripped out playbook, but it seemed sparse. What am I missing, I thought?

I wrote a quick prompt to see if any of the garden-variety LLMs had any ideas on marketing tactics:

Please can you fully read and make sure you comprehend the attached marketing plan PDF.
I'm looking for highest return, easiest to automate, launch-marketing tactics to promote my new directory website.
Please present the top 10 tactics I'm missing for promoting an online directory website.
I'm intending to automate the execution of these tasks, so exclude anything which cannot feasibly be automated with AI/LLMs/RPA/Workflows

What follows are the LLM by LLM AI results, if you just want the summary – jump to that here.

Is Gemini Flash any Good at Marketing?

Here’s what Gemini Flash 2.5 Pro thought I should add to the marketing plan:

Good Marketing Tactics	Missed the Mark / Unhelpful Tactics
Listing Scraping & Outreach (Ongoing listing additions)	Programmatic SEO (I already had this on the list)
Structured Data Schema (Better SEO value from pages)	Data-Driven Content (same same)
Reactive PR (Automated social media replies)	Listing Enrichment & Re-engagement (same same)
Content Repurposing (Content waterfall)	Programmatic Paid Social & Search Ads
PR Outreach to Journalists & Bloggers
Partnership & Affiliate Outreach

Gemini’s ideas feel a bit dry, a bit typical, but overall some good suggestions. I’m apprehensive about PR outreach via AI with LLMs at the current level they are; still I’ll test it.

Here’s the full list from Gemini:

1. Automated “Supply-Side” Scraping & Outreach – Generate better listings / ongoing listings
2. Full-Scale Programmatic SEO (pSEO) Page Generation
3. Automated Structured Data (Schema) Implementation – When you generate your pSEO pages (Tactic #2), also generate the corresponding JSON-LD schema code automatically.
4. Automated Keyword & Mention Monitoring for “Reactive PR” – Brand24, Mention, or even Google Alerts for keywords – generate replies + get human to post
5. Automated Content Repurposing & Distribution -> waterfall
6. Automated Data-Driven Content (from your own listings)
7. Automated PR Outreach to Journalists & Bloggers
8. Programmatic Paid Social & Search Ads
9. Automated Partnership & Affiliate Outreach
10. Automated Listing Enrichment & Re-engagement – Use automation to continuously improve the quality of your listings

I rate Gemini 2.5 Flash Pro’s marketing skills as:

Screwball

---

Is ChatGPT o3 any Good at Marketing?

o3 was a much more creative thinker. I liked the way it presented the options (it gave me a nice clean table with common-sense columns). Overall I feel like OpenAI’s models just talk my language better. Perhaps this is user memory.

Here’s the breakdown for o3’s marketing ideas:

Good Marketing Tactics	Missed the Mark / Unhelpful Tactics
“As-Seen-On” Badge Generator & Outreach	Affiliate/Influencer Programme on Autopilot
AI-Personalised Invite Bot for NEW Prospects	Dynamic Google Ads / Meta Ads Feed
Weekly “Top-X” Listicles & Social Threads	Broken-Link Outreach Bot
Answer-Engine Farming	Embeddable Mini-Directory Widget
AI-Generated Video Shorts Factory
Automated “Performance Snapshot” Drip to Vendors

Additionally o3 offered a prioritisation prompt for it’s marketing tactics – which is a nice touch:

Here’s the full list from OpenAI’s o3 model:

1. “As-Seen-On” Badge Generator & Outreach – Every listing that embeds your badge becomes a permanent backlink and traffic channel.
2. AI-Personalised Invite Bot for NEW Prospects – Grows inventory rapidly while the directory is hottest
3. Affiliate/Influencer Programme on Autopilot – Influencers bring trust and reach you can’t buy with ads
4. Dynamic Google Ads / Meta Ads Feed – Paid clicks convert well during launch when CPCs are lowest
5. Weekly “Top-X” Listicles & Social Threads – Free evergreen content that ranks, plus social proof
6. Broken-Link Outreach Bot – High-authority links with near-zero cost.
7. Embeddable Mini-Directory Widget – Turns partner sites into compounding referral firehoses
8. Answer-Engine Farming (Quora / Stack sites / Reddit AMA Scheduler) – You rank on popular Q&A pages long after launch day
9. AI-Generated Video Shorts Factory – Short-form video reach dwarfs static posts and is 95 % automatable now
10. Automated “Performance Snapshot” Drip to Vendors – Keeps vendors engaged and nudges upgrades/upsells

I rate OpenAI ChatGPT o3’s marketing skills as:

Twister

---

Is Claude Sonnet 4 any Good at Marketing?

Sometimes I love Claude, sometimes it’s just a bit lacking. I found these tactic ideas overall professional, less spammy in some ways than the other models, but kind of vanilla. Here’s the split for Claude:

Good Marketing Tactics	Missed the Mark / Unhelpful Tactics
Review Aggregation & Syndication	Automated Business Outreach for Free Listings (already covered)
Schema Markup Automation	Location/Category Landing Pages (already covered by pSEO)
Competitor Listing Monitoring & Outreach	Automated Press Release Distribution (I feel like PR spam is dead)
Google My Business Integration (This might work for more local-focused directories, but not sure I’ll use it)	XML Sitemap & Feed Generation (long since dealt with)
Industry News Content Automation (Kind of falls under pSEO, but I like the spin)	Local Citation Building (again, already in the Marketing plan I passed it)

… and here’s the full list for Marketing ideas Claude managed:

1. Automated Business Outreach for Free Listings
2. AI-Generated Location/Category Landing Pages – Mass-produce “Best [Business Type] in [City]” page
3. Review Aggregation & Syndication – Pull reviews from Google, Yelp, Facebook for listed businesses & Auto-generate review summaries using AI
4. Automated Press Release Distribution – AI-written press releases about directory milestones
5. Schema Markup Automation – Auto-generate LocalBusiness, Organization schema for all listings
6. Competitor Listing Monitoring & Outreach – Monitor competitors’ new business additions & Auto-reach out to those businesses for your directory
7. Google My Business Integration – Auto-sync business data between GMB and your directory
8. Industry News Content Automation – Auto-curate and rewrite industry news relevant to your directory sectors
9. XML Sitemap & Feed Generation – Auto-generate category/location-specific sitemaps
10. Local Citation Building – Auto-submit your directory to other business directories

I rate Claude Sonnet 4’s marketing skills as:

Vanilla Icecream

---

AI Launch Marketing: Which Model? (Summary)

At this point in the LLM race each of these models are actually pretty good. There were several crossover marketing tactics that all 2/3 agreed on, and there were good, if somewhat trite, ideas all round.

Given that all AI models did a fair job, there’s not much in it.

I’d probably choose o3 if I had to prioritise one AI model, but much of that is subtle preference in style.

Gemini Flash 2.5 Pro Preview	OpenAI ChatGPT o3	Claude Sonnet 4
6 tactics I’d use, 4 I wouldn’t	6 tactics I’d use, 4 I wouldn’t	5 tactics I’d use, 5 I wouldn’t
Best tactic: Structured Data Schema	Best tactic: Answer-Engine Farming	Best tactic: Competitor Listing Monitoring & Outreach
Style: Technical, with a focus on search optimisation	Style: Hacky Founder (may be reflection of user memory)	Style: Professional, straightforward

TL;DR; I guess, is that all of the current models are good marketing assistants, but don’t expect novel approaches out of any of them, (to be expected.)

How I’d Test AI Models Marketing Skills in Future

This was a quick test that I thought would produce more conclusive results 😅. I didn’t expect such samey marketing tactics from different models, (though it probably makes sense given the data they’re likely all training on). Ultimately it was stylistic differences which emerged, more than any objective winner.

In future when I run tests like this I will:

• Try more models – to get a better range of outputs
• Define parameters better – I forget that my whole business approach manifesto isn’t naturally included
• Use fresh accounts – Using a ‘daily driver’ AI biases these tests by providing user memory
• Allow iterations – How much better would the results be with one firm response volley?

See how I’m Marketing with AI Agents

This test, and others like it, are informing my overall plan for automating marketing (making a AI Launch Marketing agent). You can read all about my AI Launch Marketing agent here.