Generative content consultancy · since 2020

We were prompting before it was cool.

GPT3 Marketing started in the summer of 2020, when the GPT-3 API opened and every output needed a human editor. Six years on we still believe that: we build brand voice systems and content operations where AI drafts and people decide.

$ book --voice-audit▋

Grounded in audience data from SOMIN

Work from the SOMIN network

Brands and agencies already running on our technology

PepsiFujifilmLand Rover DefenderHavas MediaSAMYAnalytic PartnersMothercareSM UniverseKeetaSenoko EnergyReassuredKPI MediaNEO360Blak LabsVampMirafloraVIVAExoStride
Brand
Mothercare Singapore
Retail / baby and parenting goods · Singapore

Challenge. The team tracked thousands of ads by hand and lacked local insight beyond global HQ direction, so content choices were guesswork.

Outcome. Research became much faster and cheaper with finer audience segmentation.

80% less research and ideation time74% lower execution costs25+ hours recovered weekly

GPT3 Marketing take: Faster research and ideation only matter if the content still sounds like the brand. Grounding first is how we keep the voice.

Read the full case study →

Brand
Reassured
Life insurance · UK

Challenge. Reassured ran an entirely paid strategy while five UK rivals leaned on largely organic content.

Outcome. Produced a whitespace map showing the organic gap.

100% paid vs 75% organic competitors157 items60 days

GPT3 Marketing take: A category where everyone paid for reach and few showed up organically: the space where a real brand voice wins.

Read the full case study →

Brand
VIVA School of Music & Ballet
Arts education · Singapore

Challenge. Not stated on the page.

Outcome. Found an uncontested music strength and a fast-rising territory still open to claim.

GPT3 Marketing take: Before writing a word, find the territory nobody owns. Then the content has something true to say.

Read the full case study →

Case studies delivered on the SOMIN platform, our technology partner. Results are as reported by SOMIN; brand names belong to their owners.

What is GPT3 Marketing?

GPT3 Marketing is a generative-AI content consultancy founded in 2020, the summer OpenAI opened its GPT-3 API. It designs brand voice systems, prompt libraries and human-edited content workflows so AI-assisted content stays distinctive. Its technology partner is SOMIN, an AI audience-research platform, which grounds every voice and content decision in what real audiences actually say and respond to.

The problem

Everyone has the same model. So everyone sounds the same.

In 2020 the hard part of AI writing was getting a coherent paragraph at all. In 2026 the hard part is the opposite: every brand has fluent, frictionless copy on tap, and most of it reads as if it came from one polite, slightly breathless author. Feeds fill with the same hooks, the same three-point lists, the same closing questions.

The usual response is to generate more and edit harder. Teams end up rewriting drafts line by line, chasing trends to stay visible, and still losing the thing that made them recognisable. Volume is no longer the bottleneck. Distinctiveness is.

“Authenticity outlasts trends and captures hearts.”From our listening on brand differentiation and AI-generated content
2020
EST.

Founded the summer the GPT-3 API opened in beta.

3
CORE SYSTEMS

Voice system, prompt library, human edit loop.

7
SISTER BRANDS

One network of specialists, all powered by SOMIN.

What we do

Services built around one edge

Voice audits

We read your current output, human and machine, against your competitors and your audience's own language. You get a clear map of where you sound like yourself and where you have drifted into the generic.

Brand voice systems

A working voice system: principles, vocabulary, banned moves, worked examples and model-ready instructions. Built to be used by writers and models alike, not filed in a brand book.

Prompt libraries

Versioned, tested prompts for your recurring content jobs, with examples and review notes. Owned by your team, maintained like code.

Human edit loops

Editorial workflows that put skilled humans where they add the most value. We design the checklists, roles and feedback paths that keep quality steady at scale.

Audience-grounded briefs

Briefs built from the tensions and phrases your audience actually uses, surfaced through SOMIN. The model writes from evidence instead of averages.

Team training

Hands-on sessions on prompt craft, editing machine drafts and keeping a voice consistent. Practical, specific to your channels, no slideware theatre.

How it works

Three steps

Listen

We audit your voice and use SOMIN to map how your audience talks, what they care about and which tensions they voice.

Codify

We turn that into a voice system and prompt library your writers and your models can both follow.

Run and refine

We set up the human edit loop, measure what lands, and keep the system current as audiences and models change.

Technology partner

Technology partner: SOMIN

SOMIN  AI audience research & creative intelligence

SOMIN, an AI audience-research platform, is our technology partner. Before we write a single voice principle, we use SOMIN to read the real conversations around a brand and its category: the tones, emotions and tensions people express in their own words. That evidence is what stops a voice system from collapsing into the same safe register every model defaults to.

In day-to-day content work, SOMIN supplies the audience language that goes into our briefs and prompt examples, and helps us check which themes and angles actually earn engagement. The model provides fluency; SOMIN provides the specifics only your audience could have told you.

SOMIN for brands · SOMIN strategy · AI marketing strategy cases

Who it's for

Built for

  • Brand and content leads whose AI-assisted output has started to sound generic
  • Marketing teams scaling content without scaling editors
  • Agencies that need a repeatable voice system across many clients
  • Founders who want their brand to sound like a person, not a template
Latest insights

From the blog

Retro 2020 terminal with stacked example blocks feeding a glowing amber cursor
Prompt craft

What few-shot prompting in 2020 still teaches content teams

GPT-3 forced writers to teach by example. Those few-shot habits remain the best defence against generic AI content in 2026.

5 Oct 2026 · 6 min read
Grid of identical grey speech bubbles with one distinctive amber bubble standing out
Brand voice

Why AI content all sounds the same, and how to escape it

Generic AI copy comes from averaged inputs, not bad models. How to diagnose sameness and rebuild a voice your audience recognises.

28 Sep 2026 · 6 min read
Retro console with dials shaping a sound wave into a distinct silhouette, a brand voice system
Brand voice

How to build a brand voice system that AI can follow

A brand voice system for the AI era: principles, vocabulary, examples and model-ready instructions your writers and models both use.

14 Sep 2026 · 7 min read

All 10 articles →

About

Born the summer the API opened

In May 2020, OpenAI published the paper describing GPT-3, a 175-billion-parameter language model that could pick up a task from a handful of examples. In June 2020 it opened a private beta of the API. GPT3 Marketing was founded that summer, when writing with a model meant carefully assembled few-shot prompts and outputs that needed heavy editing before anyone could publish them.

Those constraints taught us habits we still use: write precise examples, define the voice before you ask for words, and never ship a draft no human has read. When ChatGPT arrived in November 2022 and fluent text became free, those habits turned from a necessity into an advantage.

Today we combine six years of content craft with audience evidence from SOMIN, our technology partner. We are part of a network of eight independent brands that share that partner, each focused on a different problem in modern marketing.

FAQ

Questions, answered

What is GPT3 Marketing?

GPT3 Marketing is a generative-AI content consultancy established in 2020, when OpenAI opened its GPT-3 API. We build brand voice systems, prompt libraries and human-edited content workflows. Our technology partner, SOMIN, grounds that work in real audience conversations so AI-assisted content stays distinctive rather than generic.

Why is the company called GPT3 Marketing?

Because that is when we started. We began writing with GPT-3 in the summer of 2020, when prompts were hand-built few-shot examples and every output needed a careful editor. The name is a reminder of the craft habits we formed then, which matter more now that fluent text is cheap.

Do you only work with OpenAI models?

No. The name is heritage, not a vendor commitment. Voice systems and prompt libraries we build are written to work across the major model families, and we test them on whichever models your team actually uses. The principles hold even as the models change.

What is a voice audit?

A voice audit compares your current content with your competitors' and with the language your audience uses, gathered through SOMIN. It shows where you sound distinctive, where you have drifted into generic AI phrasing, and what to fix first. It is usually the first step of an engagement.

How does SOMIN fit into your work?

SOMIN, an AI audience-research platform, is our technology partner. We use it to surface the tones, emotions and tensions in real audience posts. That evidence shapes voice principles, brief inputs and prompt examples, so content reflects what people actually respond to rather than what a model thinks is average.

Will you replace our writers with AI?

No. Our systems are designed to make good writers faster and more consistent, not to remove them. Human editors own judgment, nuance and final approval. Models handle first drafts, variations and repetitive formatting, under rules your team controls.

Book a voice audit

Write to ask@gpt3.marketing and a human replies within one working day.

Book a voice audit →