Every morning, professionals all over the country open ChatGPT or Claude and start from zero. They explain who they are, what the business does, who the client is and what the report needs to say. The next morning they do the whole thing again.
If that is you, the problem was never your prompts.
The Groundhog Day problem
When researchers at UCL interviewed UK professionals about how they use AI at work, the same frustrations kept surfacing. A bid writer described colleagues who had "popped a question in, it’s regurgitated out a piece of writing and then they’ve sent it over", output that could not read between the lines or judge what mattered. A copywriter put it more bluntly: "I can tell it’s AI generated. It’s just a little bit average."
The scale of the workaround culture is striking. Research by Glean in 2026 found that 60% of workers re-run the same prompt across multiple tools because the first answer was not good enough. That is an enormous amount of professional time spent negotiating with software that has already forgotten who you are.
It happens because these tools have no durable knowledge of your business. Each conversation starts empty. The built-in memory features were supposed to fix this, but they save fragments you did not choose and drag old assumptions into new work, which is why so many people turn them off.
Generic context produces generic output
Here is the mechanism worth understanding. When your input carries no context about your business, the model reaches for the most statistically ordinary answer available, because ordinary is the safest guess. "It sounds nothing like me" is a context problem. So is "it keeps suggesting things we would never do". You have not hired a bad assistant; you have hired a brilliant one and told it nothing.
Most people respond by writing longer prompts, and it helps a little. But a prompt is a briefing you have to deliver again every single session, and the whole point of good infrastructure is that you stop briefing and start working.
Give it your context once, properly
The fix is context you control: a small set of plain text files that describe your work, written once, improved over time, and readable by any AI model. At Aivian we call this a second brain, and it is the same system we use to run our own consultancy.
You can start this week without any special tools:
- One page on the business. What you do, who you serve, what makes you different, what you never do. Write it as if briefing a capable new hire on day one.
- One page on your voice. How you write, with two or three real examples of work you are proud of. AI is far better at matching a demonstrated voice than a described one.
- One page per live project. The client, the goal, the constraints, the current state. Update it as things move.
Attach these files at the start of a session, or load them into your tool’s project feature, and the difference is immediate: first drafts start in the right place, in something close to your voice, aware of your constraints. Keep the files on your own machine rather than inside any one platform and they become portable, so your context survives every pricing change, product pivot and shutdown the AI industry has planned for you.
What this does not fix
Honesty matters here. Context solves the generic-output problem, but a well-briefed AI can still be confidently wrong, so your own judgement still has to handle the trust problem. The professionals who get the most from these tools run both systems together: infrastructure that carries their context, and the critical thinking to know when the output should not go out the door.
If you want to feel the difference context makes in the next ten minutes, our free SEE tool builds you a decision prompt that carries your stakes and your lens. And if you want to build the full system, a second brain that lives on your machine, knows your work and writes in your voice, that is exactly what we teach at AI Unlocked.