Why does AI often produce generic output in HR?

Because it's treated like a search engine: a short question, no examples, no context. Output becomes usable through a fixed prompt kit with a role frame, real material, a boundary sentence, a format spec, a ban list, and an honesty anchor.

What's the single most important building block of a good HR prompt?

The boundary sentence: “Use only the information below, don't invent anything, ask if information is missing.” It prevents invented details and turns the AI into a conversation partner instead of a guesser.

How do you improve a weak AI result?

Iterate instead of starting over: name specifically what's bothering you, have it condense, ask for the opposing view, or have it rewritten for a different audience. The context stays intact.

Nick’s Advisory

AI & HR

The Prompt Kit for HR: Six Building Blocks Against Generic AI Answers

August 11, 2026 · 6 min read · by Nick, former Head of People · Last updated on August 25, 2026

“The AI only writes buzzwords for me.” I hear that a lot from HR teams, and it's often true. But the reason is rarely the tool. It's that AI gets treated like a search engine: a short question in, a finished answer expected.

A better mental model: treat the AI like a new colleague in her first week on the job. Competent, but with zero knowledge of your company. Nobody would tell her “just write something professional” and expect a usable result. You give her context, examples, and boundaries.

That's exactly what the prompt kit does: six text building blocks, set up once in about 30 minutes, then combined for every writing task afterward. No prompting course, no jargon. A system you never have to look up again once it's set up.

The setup: six building blocks, once, in 30 minutes

The prompt kit consists of six text building blocks that you set up once and then combine for every writing task. Create a document called “Prompt Kit” for this, or save the blocks as snippets in a text-expander tool.

1
Role frame
One sentence stating who's writing and who's reading: “You are writing as an HR representative at [company]. Audience for this text: [e.g. hourly staff without office access].” The audience line is the underrated part, without it you get the one-size-fits-all tone.
2
Context attachment
A fixed block at the end of the prompt for real material: the policy in question, the role profile, two old texts that worked well. One real example beats three adjectives.
3
Boundary sentence
Word for word: “Use only the information below. Don't invent anything. If information is missing, ask me instead.” The permission to ask questions turns the AI from a guesser into a conversation partner.
4
Format spec
Length, structure, form, for example “max. 200 words, three subheadings, informal tone.” Without a format spec, the AI decides, and it almost always decides: too long.
5
Ban list
Words and buzzwords that should never show up again, as a growing list. Every new buzzword in an output goes on the list, that's how the kit gets better with every use.
6
Honesty anchor
Word for word: “Flag every spot where you're unsure or making assumptions with [check].” The [check] spots become your edit checklist.

The prompt kit: six building blocks, set up once

Prompt kit

Role frame

Who's writing, and above all: who's reading.

Context attachment

Real material instead of adjectives.

Boundary sentence

Don't invent anything. Ask instead.

Format spec

Length, structure, form.

Ban list

Buzzwords that never show up again.

Honesty anchor

Flag uncertain spots instead of smoothing them over.

No more buzzword filler

Set up once in 30 minutes, then combined with every prompt.

The finished prompt, ready to copy

All six building blocks in one prompt. Replace the bracketed placeholders with your own details.

You are writing as [role] at [company, one sentence of context]. Audience for this text: [audience, e.g. hourly staff without office access].

Use only the information below. Don't invent anything. If information is missing, ask me instead.

Format: [e.g. max. 200 words, three subheadings, informal tone].

Don't use any of these words: [ban list, e.g. exciting, dynamic, we're thrilled].

Flag every spot where you're unsure or making assumptions with [check].

Here is the material to base this on:
[context attachment: policy, role profile, or sample texts]

The application: iterate instead of starting over

The workflow is the same for every writing task: lead with the role frame, boundary sentence, format spec and honesty anchor, attach the ban list, put real material in the context attachment, send. A mediocre result afterward isn't a reason for a new prompt, it's a reason for a follow-up instruction.

Three follow-ups cover almost every case: “Condense this to half the length without losing information.” “Take the opposing view: what's wrong with this draft, what's missing?” “Rewrite the same content for [different audience].” Name specifically what's bothering you, like “too formal” or “point X is missing,” instead of starting from scratch. The AI already has the context, don't throw it away.

A real-world example: a new work-from-home policy needs to be announced.

Search-engine prompt

“Write a staff memo about the new work-from-home policy.” Result: interchangeable corporate-speak that could sit on any company's intranet.

Prompt-kit prompt

Role frame with audience (all staff, including those without a company laptop), the policy itself in the context attachment, boundary sentence, a 150-word format spec, and the ban list. Result after one round of “condense to half”: a text that actually sounds like your company.

The three failure patterns

The same three prompting mistakes keep showing up in HR teams, regardless of which tool they use.

1
Adjective prompts
“Professional, friendly, modern” is not information. Sample texts are information.
2
The mega-prompt
Cramming everything into one giant instruction and hoping for a perfect result, instead of starting solid and sharpening it with two or three follow-ups.
3
Adopting output unreviewed
Ignoring the [check] markers means the honesty anchor was set for nothing. Personal data stays out of open tools regardless.

Your setup checklist for today

The full setup takes about 30 minutes: create the “Prompt Kit” document and write out the six building blocks, copy the boundary sentence and honesty anchor word for word, start the ban list with your first five buzzwords, and pick one real task to test the kit on today.

Is an interim manager an employee or self-employed?

An interim HR manager is almost always engaged as a self-employed contractor on a project mandate, not as an employee of your company. That's precisely why there's no notice period or severance if the mandate ends, and no employer contributions on top of the day rate, the flip side of the higher daily cost.

For companies engaging interim managers in Germany, this makes proper contract structuring important: the mandate needs to be genuinely project-based and independent in how the work is carried out, so it doesn't drift into what German law treats as disguised employment (Scheinselbstständigkeit). A clear scope, a defined end date, and the manager working for other clients in parallel are the usual markers of a clean setup, and worth clarifying before the mandate starts, not after.

AI Consulting for HR: What Actually Matters When You Adopt ItHow an AI adoption in an HR team actually unfolds, with pitfalls from hands-on practice.

More from practice: AI Consulting for HR

How an AI adoption in an HR team actually unfolds, which pitfalls are typical, and where AI genuinely fits.