Key takeaway: Custom instructions and memory move brand consistency from human review to system design. That's the real shift, not the time saved on retyping prompts.
Why it matters: Most teams still treat AI output as something to fix after the fact. On-brand by construction beats on-brand by review — every time, at scale.
What happened
Every major assistant now ships native customisation. CNET's walkthrough of customising ChatGPT, Gemini and Claude lays out the settings: tell the tool who you are, how you want it to respond, and let it remember across sessions.
OpenAI's rollout added structure to this. TechRadar reported in January 2025 that ChatGPT introduced options to set traits, tone and rules that persist across conversations, so you stop restating context every time.
Source: TechRadar, 2025
The pitch is simple: less prompt hacking, more consistent output. The reality is more interesting, and it's where most of the commentary will stop short.
Most people will read this as a time-saving tip
The consensus take writes itself. Customisation means you type less, so you save time — a few minutes of setup reclaims hours of copy-pasting the same instructions each morning. Fair enough, and true as far as it goes.
But that framing quietly assumes the bottleneck is typing speed. It isn't. The bottleneck is that hardly anyone knows how to instruct these tools well in the first place. Forrester research reported by HR Dive found the share of employees who understood prompt engineering rose only from 22% in 2024 to 26% in 2025 — four points in a year.
Source: HR Dive / Forrester, 2026
We think AI customisation is brand infrastructure, not a settings menu
Here's the reframe. A custom instruction that fixes tone, format and forbidden phrases isn't a personal convenience — it's a brand rule, encoded once, applied on every response without anyone checking. That's creative as infrastructure: on-brand by construction, not by review.
In our experience building agents, the expensive part of AI content is never generation. It's the review loop — the marketer who rewrites the first draft, softens the American spelling, strips the banned words, and re-adds the disclaimer legal insists on. Do that fifty times a week and you've rebuilt manual work on top of an automated tool.
Customisation settings are the consumer-grade version of the right idea. The industrial version is a brand system that agents can call directly — your voice, your rules, your do-not-say list, exposed as something a machine reads before it writes, not after. When the guardrails live in the system, consistency stops depending on whoever happens to be reviewing that day.
The catch: a settings box in one person's account doesn't scale to a team. Forty marketers each hand-tuning their own ChatGPT personality is forty different brand voices wearing the same logo. The consistency you want is the consistency you build once, centrally, and make callable — which is exactly how we approach AI agents for marketing.
And it explains the numbers. McKinsey's 2025 State of AI survey found most organisations report regular gen-AI use, yet just 39% see EBIT impact at the enterprise level. Adoption is easy. Value shows up when the workflow — not the settings toggle — is redesigned around it.
Source: McKinsey, 2025
So by all means fill in the custom instructions box. Just don't mistake a personal preference file for a brand system. One saves you a minute. The other is why a content agent produces something you'd actually publish.
What this means for marketing teams
- Write your brand voice as constraints, not adjectives. "Under 200 words, UK English, no jargon list attached" beats "professional yet friendly" — do it this week and audit outputs against it.
- Centralise the rules. One shared, version-controlled voice spec beats 40 individual custom-instruction boxes that drift within a month.
- Measure review time, not generation time. Track the minutes spent editing AI drafts before and after; that number is your real ROI, and it should fall.
- Encode the do-not-say list. Banned words, legal disclaimers and spelling conventions belong in the system, applied on every response — not caught in a final read-through.
- If you're standardising voice across a team, talk to us about brand systems agents can call rather than settings each person tunes alone.
Frequently asked questions
What are ChatGPT custom instructions?
Custom instructions are settings that tell ChatGPT your role, preferred tone and response format so it applies them automatically to every new conversation, without you restating context each time.
Does AI customisation keep my brand voice consistent across a team?
Not on its own. Personal custom instructions only affect one account. Team-wide consistency needs a centralised, shared voice specification that every tool or agent references, rather than settings each person tunes individually.
Is customising AI tools worth the time?
Yes, but the real payoff isn't typing less. It's reducing the time spent editing off-brand drafts. Measure review time before and after — that fall is where the value actually shows up.
