All of my team members and most of my clients are concerned about the darkside of AI. In our corner of the internet, we've been slow to adopt, fast to raise our standards, and are diving deep into the work of ethical governance. We use it with care, for strategic leverage, and spend time engineering powerful outcomes only. We build tools that help the small biz community collectively reduce their footprint while getting greater business outcomes.

The two primary lanes we use it in are marketing (testing for new efficiencies in strategic and creative assistance) and client success (designing tools that create better outcomes for clients, with reduced carbon footprint and zero AI learning curve).

I made this list for my team around how to increase efficiency in the most common use cases for marketing and founder/CEO usage. It maps to Claude's lineup as of right now. Since many of you use ChatGPT day-to-day, I've added the closest equivalent there too. But OpenAI's naming has changed twice in the last few months alone (most recently to a Sol / Terra / Luna tier system in July 2026). Model names can shift every few months, but the timeless principle is about matching compute to consequence.

These tips fall into two categories: Tips 1 through 4 are about choosing the right-sized model, and Tips 5 through 7 are about using whatever model you've chosen more efficiently. Model choice makes a bigger difference, but prompt discipline is an opportunity to bake efficiency into an everyday best practice.

The Rule of Thumb

Think of the models like people you'd hire:

• Haiku is the intern with perfect grammar (ChatGPT equivalent: Luna)

• Sonnet is your sharp senior colleague (ChatGPT equivalent: Terra)

• Opus is the expensive consultant (ChatGPT equivalent: Sol)

• Fable is the specialist firm you bring in for the bet-the-quarter project (ChatGPT former 4th “Pro” tier has been combined into Sol. ChatGPT works off 3 models to Claude’s 4.)

If a smart intern could do it, use the small model. Default down for mechanical work, start high for complex work, and never pick the middle out of indecision.

First gut check baked into this rule: the goal isn't just the smallest model, it's the smallest model that can do the job well. A cheap attempt that fails and gets redone up a tier costs more than starting in the right place the first time.

Second gut check: before you reach for any model at all: is this a transformation or a retrieval? If you want a fact, a link, or a source, that's retrieval, use search. AI earns its keep on transformation, not fetching. Asking AI to be Google is both less accurate and less efficient, no matter how small the model.

Tip 1. Haiku: Transform What Exists

*In ChatGPT use Luna, or the mini/nano tier if you're on the API.

Mechanical work. The thinking is done, you just need hands.

• Proofreading for typos, grammar, house style

• Reformatting: blog post into bullets, email copy into SMS length

• Resizing approved copy across platforms (caption to tweet to LinkedIn)

• Summarizing a meeting transcript or a competitor's email

• Alt text, meta descriptions, UTM naming

• Pulling data from a doc into a table

• Subject line variations on a concept you've already approved

• Cleaning a spreadsheet column

Tip 2. Sonnet: Draft and Decide Day-to-Day

*In ChatGPT use Terra, OpenAI's own "everyday, balanced" tier.

Judgment plus execution. Most of your week lives here.

• Substantive editing: restructuring an argument, tightening a draft without flattening the voice

• First drafts of emails, captions, or landing pages from a solid brief

• Analyzing campaign results and suggesting the next test

• Turning a strategy doc into a content calendar

• Adapting copy across personas

• Everyday coding: a script to dedupe a CSV, a small automation, a dashboard tweak

• A competitive teardown of one brand

Tip 3. Opus: Think Hard About Hard Problems

* In ChatGPT use Sol, its deeper reasoning mode.
Genuinely difficult reasoning, where a subtle error is expensive.

• Positioning and messaging strategy from messy inputs (survey data plus brand voice plus market context)

• Long-form thought leadership where voice fidelity actually matters

• Multi-document synthesis, like twenty client calls into one insight report

• Pricing strategy, offer architecture, anything where being slightly wrong costs real money

Tip 4. Fable: Build Things People Will Use

* In ChatGPT also Sol for now. This is the ceiling.

Frontier work where the output becomes an asset.

• Prototyping internal tools your team will actually use: a video editing prototype, a design review tool, a content QA app

• Client-facing product work, meaning anything a client will touch or judge

• Long agentic projects where the model plans, executes, and self-corrects across many steps ("audit our entire email archive, categorize it, flag voice drift, propose the fix")

• The rare strategy work where the stakes justify the ceiling

Here's the part that surprises people: for prototyping, the biggest model is often the most efficient path. One working prototype in one session beats three failed attempts on a smaller model plus a developer's week untangling the mess. The waste isn't in using the big model. The waste is using it to write subject lines.

Tip 5. Treat Editing as Three Different Jobs

"Editing" hides three tasks that belong to three tiers:

• Proofreading goes to Haiku (or ChatGPT’s Luna)

• Line editing (rhythm, clarity, voice) goes to Sonnet. (or ChatGPT’s Terra)

• Developmental editing (is this even the right argument?) goes to Opus (ChatGPT’s Sol), or a human

Tip 6. Brainstorming and Research: Cap the Output

This one habit alone will cut your compute (and your reading time) more than any model choice. When you're brainstorming or researching, AI defaults to giving you everything, and everything is expensive to generate and expensive to wade through.

• Set strict word count parameters. "Give me 10 angles, 15 words each" beats "give me some ideas" every time. Shorter outputs, sharper thinking, less waste.

• Tell it to check with you before running off. If you're in exploratory mode, say so: "check with me before doing anything beyond answering the question." Otherwise you ask one thing and get a research report, a table, and three follow-ups you never wanted.

• Expand only what earns it. Get the short list first, then say "go deeper on number 3." You spend computing on the one idea that matters instead of ten that don't.

In practice: literally every single time I'm having it help me think something through, I tell it to cap its responses to 150 to 300 words, depending on the task. That one constraint alone forces sharper thinking on both ends.

Tip 7. Ask: Have I Given It What It Needs to Get This Right in One Pass?

Every vague prompt that needs three rounds of "no, more like this" is four times the compute of one specific prompt. Good prompting is efficiency.

That said, you shouldn't have to become an AI prompt engineer to grow your business. The least efficient thing of all is amateur marketers tinkering with AI in an endless loop, trying to build marketing operating systems that may or may not work. We're charting a better path forward for everyone. Get on the waitlist here to hear about it.

The Bigger Picture

The biggest lever of all is your opportunity to think about what companies you're aligning with, and how to use AI competitively while helping shape a better future. We've been analyzing comparisons between 500 business owners running their own agentic DIY AI setups against a world where we're all running a single source, The S3 System ™, and the gap is exponential.

This is what's behind our product development. Get on the list to be the first to know.

Katie Wight is the founder and CEO of Strong Brand Social, a social media strategy agency she built to $4M in five years. With 15+ years in the industry — including global social programs at Burton Snowboards and Tata Harper — she created The S3 System™ to give founder-marketers a strategy framework that’s actually built to drive revenue. She works with bootstrapping brand founders, family businesses, and marketing leads who are done guessing.

© 2026 Strong Brand Social. All rights reserved. The S3 System™ and Customer Acquisition Machine™ are proprietary frameworks developed by Strong Brand Social. No part of this content may be reproduced, redistributed, or used to develop derivative works — including courses, workshops, or training programs — without express written permission. To request permission or report unauthorized use: hello@strongbrandsocial.com