How to Get Good AI Prompts for Work

Good work prompts name the actual deliverable, the audience it’s for, and any real constraints — generic requests produce generic drafts that need heavy editing, while specific ones produce something closer to usable on the first attempt.

Name the actual deliverable, not the general topic

“Help with the quarterly report” is a topic. “Draft the executive summary section of the quarterly report, under 200 words, for a board audience” is a deliverable — the second version needs no follow-up because it already contains decisions the AI would otherwise guess at. This mirrors the same specificity principle covered in how to write prompts for ChatGPT, applied specifically to workplace deliverables.

Audience shapes tone and depth more than any other factor

The same content needs genuinely different register for a board audience versus a team update versus a customer-facing message. Naming the actual audience explicitly is often the single highest-leverage detail in a work prompt.

This matters even when the audience seems obvious from context — what’s obvious to you as the person with full context on the situation isn’t necessarily inferable by the AI from the prompt alone. Stating audience explicitly, even when it feels redundant, removes a genuine source of ambiguity rather than assuming it’ll be correctly inferred.

State real constraints upfront

Word count, tone, formatting requirements, things that must be included or avoided — naming these upfront saves a round of “actually, can you also…” follow-ups that a first draft without constraints usually needs.

Constraints worth stating explicitly for work content specifically include: any required elements from a template or house style, information that must not be included (confidential figures, internal-only details), and the actual format the output needs to land in (a slide, an email body, a formatted document) since this affects how the content should be structured, not just its length.

Handling recurring work with saved prompt templates

For genuinely recurring deliverables — weekly status updates, standard client communications, routine reports — building a saved template prompt with the stable elements filled in and clear placeholders for what changes each time saves considerable repeated effort compared to reconstructing an effective prompt from scratch each cycle. A good template captures your organization’s actual conventions (standard sections, typical tone, common audience) so you’re not re-explaining institutional context every single time.

Balancing AI drafting speed with genuine review

AI-drafted work content should generally go through the same review rigor you’d apply to a human-drafted first pass, not less, particularly for anything representing your organization externally or informing a real decision. Speed of drafting isn’t the same as accuracy or appropriateness, and treating AI output as a finished product rather than a strong first draft is a common source of avoidable mistakes in professional contexts.

This is especially worth attention for any content involving specific figures, claims about performance, or commitments — these deserve verification against source data before being finalized, regardless of how polished and confident the AI-generated version reads.

Common mistakes in workplace AI prompting

Providing the topic but not the actual deliverable format is the most common gap — people get comfortable stating what they want written about but forget to specify the concrete shape it needs to take. A related mistake is omitting audience, which often produces content that’s technically correct but pitched at the wrong level of detail or formality for who’s actually going to read it.

Another common mistake specific to work content: treating a single AI-generated draft as final without the verification pass that any consequential business communication deserves, especially when specific facts, figures, or commitments are involved.

Prompting for different types of professional communication

Email, presentation content, and formal documents each benefit from somewhat different emphasis within the deliverable-audience-constraint structure. Email prompts benefit most from explicit tone guidance, since email tone can read very differently than intended without clear direction. Presentation content benefits from explicit density constraints — how much text per slide, how many key points — since AI-generated slide content defaults toward more text than typically works well in an actual presentation. Formal documents benefit from explicit structural guidance, since document conventions (required sections, standard formatting) vary more by organization and context than the underlying content itself.

Recognizing which category a given request falls into, and adjusting which constraint category to emphasize accordingly, produces more immediately usable drafts than applying identical prompt structure regardless of what kind of communication is actually being created.

Prompting for internal versus external communication

Content headed to an internal audience can generally assume more shared context and tolerate a more casual register than content headed externally, where tone missteps or unclear communication carry higher real cost. Being explicit about internal-versus-external in the prompt itself — not just implied by the topic — helps calibrate formality and the amount of background explanation needed, since an AI without this signal may default to a middle-ground tone that’s not quite right for either context.

For external-facing content specifically, additional review scrutiny is worth building into your process by default, given the higher cost of an error reaching an external audience compared to an internal one.

Using AI for feedback and revision, not just initial drafting

Beyond generating first drafts, AI tools are genuinely useful for reviewing content you’ve already written — checking clarity, catching inconsistent tone, identifying places where the argument or explanation could be tightened. This is a different prompt structure than drafting from scratch: instead of task-context-format, it’s closer to “here’s my draft, review it for [specific concern],” which works best when you name the specific type of feedback you want (clarity, tone consistency, conciseness) rather than a generic “make this better.”

This review use case is often underused compared to drafting, even though it can be just as valuable — particularly for content you’ve already invested significant effort in and want a fresh, structured pass on before finalizing.

Building institutional consistency across a team using AI tools

When multiple people on a team use AI tools for similar recurring work, individually-developed prompting habits can produce inconsistent output style across the team, even when everyone is genuinely trying to follow the same conventions. Sharing effective prompt templates across a team — not just individual technique tips, but actual reusable prompt structures for common deliverable types — helps maintain consistency in a way that’s hard to achieve when everyone develops their own approach independently.

This is worth deliberate investment for teams doing high-volume, recurring AI-assisted work, since the time spent building and sharing good templates once pays back repeatedly across the team, compared to each person separately reinventing similar prompt structures with slightly different, inconsistent results.

Recognizing when AI drafting isn’t the right tool for a specific piece of work

Not every work deliverable benefits equally from AI-assisted drafting. Content requiring deep domain expertise the AI genuinely lacks, highly sensitive communications where getting the nuance exactly right matters enormously, or content where the actual value lies in it being demonstrably human-authored (a personal note, a highly individual piece of writing) may be better served by traditional drafting, with AI reserved for a supporting role like structural feedback rather than initial generation.

Recognizing this distinction — rather than defaulting to AI drafting for literally everything — is itself a mark of using these tools well, matching the tool to tasks it’s genuinely suited for rather than applying it uniformly regardless of fit.

Applying iteration to work drafts effectively

The same iteration principle that improves results across AI prompting generally applies to work content specifically — a rough first draft refined through two or three rounds of specific feedback tends to produce a more polished final result, faster, than trying to specify every requirement perfectly in the initial prompt. For work content specifically, this means treating the first AI draft as a genuine starting point for your own editorial judgment, not a finished deliverable to submit as-is.

Specific, targeted feedback on the draft (“tighten the second paragraph,” “this needs a stronger opening line for this audience”) produces better refinement than a vague “make it more professional,” for the same reason specific requests outperform vague ones throughout every stage of AI-assisted work.

Try it yourself

Draft [specific deliverable] for [specific audience]. Constraints: [word count, tone, must-include items].

For all my work requests going forward, default to [tone/style preference] unless I say otherwise.

FAQ

For broader prompting technique, see how to write prompts for ChatGPT or the AI prompt generator.