We’re being seriously unfair to our AI ‘team mates’.
Imagine dropping a new intern into your business and asking them to deliver the same results as your most experienced employees, on day one, with no training, no access to internal systems and no idea how your company actually works.
That’s how most people are using AI. And they blame the tool, rather than the way they are using it, when it comes up short.
We tend to think of AI tools as omniscient. Tools are trained on oceans of data, can synthesise information in seconds and can generate text, code and strategy on command. But we vastly underestimate the importance of context. AI cannot read minds.
Take image generation as a simple example. You might picture an elephant riding a bicycle in your mind, but if you ask an AI to generate that image, the odds of it matching what you imagined are tiny. There are millions of variables at play: is it a cartoon or photorealistic image? Is the bike a children’s tricycle or a penny-farthing? Is the setting a circus tent or Brighton Pier? Is the mood dark and dramatic or bright and playful?
The same principle applies to any AI task, whether it’s generating an image, drafting a strategy document, or writing code. Without clear context, the AI is forced to guess from millions of possible interpretations. The more detail and direction you provide, the closer the output will align with what you actually had in mind.
We frequently see scenarios where people have felt they ‘hit the limits of what AI can do’ and haven’t been happy with the output, often relying on human staff to complete the task instead. But let’s look at the difference in context that an AI tool has vs even a junior employee.
Junior employee
If they’ve worked at the company for two years they might have:
- 3,000-4,000 hours of real world, on the job experience
- Thousands of feedback loops based on the work they’ve done
- Seen thousands of interactions between colleagues, clients and suppliers
- Had tens of hours of training
They’ve sat in meetings, learned from mistakes, watched the company evolve and internalised the subtle ways decisions get made. This adds up to a lot of context about what the right thing to do is.
Note: The public vs. the private reality
AI has a broad, surface-level understanding of how businesses work in general. It knows what an HR policy looks like, what a marketing plan might contain and how most companies think about customer service.
But don’t underestimate the gap between the publicly available version of “how businesses operate” and the lived experience of working inside a specific company. The real secret sauce is likely hidden away.
An LLM can give you a generic answer about running a product launch. But if you want it to mirror your team’s actual process, with the quirks, shortcuts and priorities that make you unique, you have to feed it that context.
There’s not much you can do about the first two, but there is a lot you can do about the third. And this is where you can unlock huge gains in the usefulness of your AI systems. For example:
- OpenAI reports that connecting an LLM to your data (via RAG) can cut hallucinations by 60–80%, dramatically improving trustworthiness
- Anthropic’s Claude research shows that combining context documents with “chain-of-thought” prompting boosts factual accuracy by 50%+
- Microsoft Copilot customers report 2× productivity gains once they integrate internal knowledge bases, compared to using it as a generic chatbot
The context an AI tool, e.g. an LLM, has
An LLM has three main sources of “context”:
- Its training data – a vast but static snapshot of the public internet, books and other data sources up to a certain point in time.
- Whatever it can access live – if it’s connected to the internet, it can pull in fresh data (but still doesn’t know which data matters most to your situation).
- What you tell it – your prompts, documents, knowledge bases and connected systems.
Improving your AI with context
If you want better output from AI, you have to give it as much context as possible.
There are three primary ways to do this:
1. Better prompts
Be explicit, detailed and specific.
Using “prompt stacks” is a powerful way to improve results. In each prompt, clearly include four elements: the goal, the desired format and constraints, any warnings or pitfalls to avoid and the relevant context or background.
Then iterate and refine. Don’t expect the first attempt to be perfect, provide feedback, narrow in on the right answer and keep improving. Encourage the system to ask clarifying questions so it can deliver the best possible output.
2. RAG (Retrieval-Augmented Generation)
Connect your AI to a knowledge base containing your own documents. The more context you provide, the better the results. SOPs, company positioning documents, sample outputs to show the AI what “good” looks like, call transcripts, project files, all of these help the AI refine its responses and produce outputs that have a higher probability of being what you specifically want.
3. Integrations
Connect your AI to the systems your team already uses, SharePoint, Google Drive, Notion, Slack, CRM tools, so it can pull in the same resources your employees rely on. This turns your AI from a generic assistant into one that understands your documents, processes and workflows.
The bottom line
AI is not a mind-reader. It’s a tool that’s only as good as the context you give it.
If you want to make your AI tools more powerful, feed them the details, examples and background they need to think like someone on your team.
Most companies are leaving 50% to 80% of an LLM’s potential power untapped by failing to provide proper context.