The default trap
Generic AI that does not know your context
Most professionals at UK consultancies use ChatGPT the same way: open a blank chat, type a question, get a generic answer, move on. The tool works, but only at a fraction of its capability. The problem is not AI — it is the setup. Without configuration, ChatGPT has no idea who you are, what your firm does, or how you need information delivered. Every session starts from scratch, and every session costs time that should go into client work.
There is a version of AI adoption doing the rounds in professional services right now: everyone uses the same blank chat window with the same defaults, gets results that need heavy editing, and quietly concludes that AI is not quite there yet for real work. That conclusion is understandable, but wrong. Without Custom Instructions, Memory, and Projects, you are asking ChatGPT to behave like a specialist while treating it like a temp who has never met you. The capability is there. The firm has not activated it. Three features fix most of that friction in under 30 minutes.
Set once, applies always
Custom Instructions give ChatGPT permanent context
Every time you open a new session without Custom Instructions, you start over. You correct the spelling to British English. You explain your sector. You ask it again not to write five-paragraph essays when you need two lines. Then you do it all again tomorrow. Custom Instructions eliminate that loop: set them once and ChatGPT carries the context into every subsequent conversation.
To set them up, go to your account, click the profile icon in the bottom left, then Settings, then Personalisation, then Custom Instructions, and make sure the feature is enabled. You get two fields. The first asks what ChatGPT should know about you — use it to describe your role, your firm, your sector, and the work you do day to day. The second asks how it should respond, and that is the more valuable field. Worth including there:
- British English throughout. If your default output is American spellings, flag it explicitly. One line eliminates a correction you would otherwise make on every response.
- No AI filler. If you do not want every answer prefaced with “Certainly!” or reminders that it is an AI, say so.
- Tone and format preferences. Concise bullets, structured paragraphs, or plain prose — set the expectation here.
- Relevant links and context. If you regularly ask it to draft emails with your meeting link, or reference your firm’s services in outreach, add those so they are available without prompting.
The saving is not dramatic on any single interaction. Across a working week it is material. You stop re-explaining yourself, and the output starts closer to usable — which means less editing before it reaches a client.
Context that compounds
Memory builds a running record of your work
Custom Instructions give ChatGPT a fixed profile of who you are. Memory gives it a running record of what you have been working on. Without it, every chat is independent: ChatGPT knows your role and preferences, but not that you spent last week building a proposal template for a client, or that you are mid-way through rolling out a new reporting process. With Memory enabled, it saves relevant information from your conversations and references it going forward.
In the same Personalisation screen, check that Save Memories and Reference Chat History are both switched on, then click Manage to see exactly what it has retained. Memory works quietly in the background — over time, ChatGPT starts to understand your recent projects, the tools you use, and the output you produce most often. Response quality improves not because the model changed, but because the context deepened.
One important control is Temporary Chat, found in the top right. Not everything you discuss is relevant to your professional work. If you are helping a family member with homework or exploring a personal interest, you do not want that bleeding into your professional memory. Temporary Chat creates a session that leaves no trace: it does not appear in history and generates no new memories. Keeping your professional memory clean means ChatGPT stays relevant to client work rather than drifting toward a mix that serves neither purpose well.
Walls between workstreams
Projects keep separate work in separate contexts
Custom Instructions and Memory work at the account level — they give ChatGPT a picture of you as a professional. But if you run multiple distinct workstreams, such as a business development push, a specific client engagement, and an internal initiative with different stakeholders and files, keeping them in one shared context creates noise. OpenAI Projects solve that. Each project is an isolated environment with its own custom instructions, its own memory, and its own file library, so you can run completely different contexts in parallel without interference.
Click New Project in the left-hand navigation, give it a name and, if you want, an icon. The key decision at setup is memory scope: you can let the project draw on your general account memories, or restrict it to project-only memory. This setting cannot be changed after creation, so decide carefully. Once created, upload files ChatGPT can reference throughout the project — brand guidelines, client briefs, frameworks you use regularly, previous deliverables. Files added to a project can be referenced but not downloaded back out, so treat it as a working environment rather than a document store. You then add a dedicated set of custom instructions specific to that project, sitting alongside your account-level ones.
The clearest demonstration of why Projects matter is what happens without them. Take an identical prompt — “Draft a social post for our upcoming event” — and run it in a project and in a fresh general chat. The general chat draws on your account-level memory and produces something broadly professional but context-light. The project produces something grounded in the specific brief, file context, and instructions you set for that workstream.
That gap is the difference between output you need to heavily edit and output you can work with directly.
Where it pays off
What this means for your firm
These three features sit at the Knowledge layer of the KWA framework: the foundation that has to be in place before workflow automation and AI agents can deliver consistent results. In the 5 Steps for AI Leadership framework, this is where Activate happens — where people actually start using AI tools in their daily work, not just experimenting once and drifting back to old habits.
This is not the final destination, but it is where individual AI capability starts to compound. Your senior team stops losing time to repetitive re-prompting. Output quality improves without extra effort. Context that used to live in someone’s head or a separate brief starts to travel with the tool itself. Across a firm of 20 or 50 people, that matters: one partner recovering two hours a week from reduced editing and re-prompting is one thing; ten fee earners doing the same is a different conversation. The firms pulling ahead on AI are not using fundamentally different tools — they are using the same tools better, starting with the configuration most teams have not touched.