A founder I worked with last quarter had typed “write me a sales email to a marketing director” into ChatGPT, read the output, and told me AI was overhyped. The email was technically competent. It was also indistinguishable from every other sales email a marketing director had received that month. She closed the tab and went back to writing them by hand, an hour per email, three or four a week. The AI had not failed her. The brief had.
This is the most common AI failure pattern I see in owner-operated firms. The complaint is “the output is generic”, but the underlying issue is upstream of the model. Anyone who has hired a freelancer on Upwork knows what happens when you instruct them with “write something good for our website”. You get mediocrity. The fix in both cases is the same artefact, a proper brief.
What is a contractor-style brief for AI?
A contractor-style brief is a short structured instruction containing the six pieces of information a remote worker needs to produce usable output without follow-up questions. Context, audience, goal, constraints, format, tone. The shape comes from professional services briefing practice, where the cost of rework drove firms to treat the brief as insurance. Anthropic’s prompt engineering guidance reaches the same conclusion under different labels.
The structure is not theoretical. Upwork’s guidance on onboarding a new client lists the same artefacts that show up in any decent SOW, scope, deliverables, expectations, and the prior context that lets the contractor work without supervision. Smartsheet’s SOW reference builds on this with acceptance criteria and standards. AI sits in exactly this position. It works in isolation, with only your words to guide it, against a brief that either specifies enough or does not.
Why does this matter for your business?
Because the alternative is the four-line prompt that wastes the model’s biggest advantage. AI produces a usable first draft in seconds, but only if the brief carries enough specification to anchor the output. Without context, the model defaults to the average of its training data, which is by definition generic. With a proper brief, the same model produces work that sounds like someone in your firm wrote it.
For a services firm in the £1m to £10m turnover band, where the founder’s writing time is the binding constraint, the maths is direct. A weekly sales email that took ninety minutes by hand becomes fifteen minutes briefed, generated, and edited. Smartsheet’s SOW guidance points at the same insight in the contractor world, briefs that specify scope, objectives, deliverables, and acceptance criteria produce fewer revision cycles. The OpenAI prompt engineering reference recommends specificity on context, outcome, length, format, and style, the same six elements under different labels. The two disciplines are the same skill on different surfaces, and the firms that internalise this stop complaining about generic AI output within a fortnight.
What goes in the six-element brief?
Each element answers one question the model cannot infer from a four-line prompt. Context tells the model what surrounds the work, what your firm does, who you serve, what is at stake. Audience names the person who will read the output, not the person paying. Goal is the shift you want the reader to make. Constraints are the rules. Format is the shape. Tone is the voice, best taught by example.
The single highest-yield move inside the six is showing rather than describing tone. The few-shot prompting research from the GPT-3 paper, and the DAIR.AI prompting guide that built on it, both show that worked examples outperform adjective lists. Paste a paragraph of your own writing, or a competitor sample you admire, and say “in this voice”. The model pattern-matches on a sample faster than it can interpret “warm but professional, technical but accessible”.
A worked example sharpens the contrast. Briefed badly, “write a sales email about our finance services, make it compelling”, you get the standard cold-email template, generic subject line, no specific hook, a closing ask the reader has read a hundred times. Briefed well, with a paragraph of context on what you do and who you serve, an audience line (“the founder of a £2m consulting practice who has never had formal financial management”), a specific goal (“get them to book a fifteen-minute call, overcoming the worry that we’re too expensive for a small practice”), constraints (“under 150 words, no jargon, no price reference”), a format (“short paragraphs for mobile reading, single CTA”), and a tone sample (“like a fellow founder texting, not a sales department”), you get an email that opens with the cash-timing problem services firms know cold and reads like the firm wrote it. Subject lines stop being “Optimise your financial future” and start being “The cash gap founders never sort”. The model has not changed. The brief has.
When does it pay back, and when do you skip the brief?
The brief pays back on recurring or high-stakes work. Anything you do more than three times a month, anything that has to land in a specific shape for a specific reader, anything where a generic draft would damage trust. For exploratory questions or open-ended thinking, the brief is overkill and you should type freely. The signal you have crossed the threshold is rebuilding the same prompt from scratch each week.
Ethan Mollick at Wharton has written about working with AI iteratively rather than crafting a perfect single prompt. His advice and the brief discipline pull in the same direction, the brief gets you to a good first draft faster, the iteration tightens it. Treat the AI’s first output as a contractor’s first draft. Read it, give specific feedback (“too formal, use this paragraph as the tone reference”), get a second pass, then edit yourself. That cycle takes ten minutes for work that used to take an hour. The brief is the upstream investment that makes the iteration short.
What related practices keep the brief sharp?
Three habits compound the brief over time. The first is saving the briefs that work. Once a brief produces good output twice, it earns a slot in your standing prompt library, called by name rather than rebuilt. The second is refining the tone sample as your writing evolves, what you pasted in February may not match how you sound in November. The third is learning the editor’s eye.
The wider context sits in the pillar piece on this cluster, AI for your own work, not just your business, and in the EAD-Do framework recast for AI, which places one-shot delegation in the Delegate quadrant alongside drafting first passes with AI. For founders working with personal data inside a brief, the ICO’s UK GDPR guidance on AI is the reference point, anything you paste in to set context counts as processing.
The brief is the foundational skill that decides whether AI sits inside your firm as infrastructure or stays a novelty you abandoned by March. Master the contractor brief and you have mastered AI. The two skills are the same shape, applied to different workers.
