The DEF method for complex prompts

Put your terms in [BRACKETS]. Pin down what they mean.

A complex prompt leaves a lot of room for interpretation. With DEF you put the important words in brackets and add a list of definitions. Then the model does what you mean, not what it thinks you mean.

Ebzeg tg [NDHAA] pnn [RBCXNHY] pxwgi mf [GPYKNEIL] kc [AZDV].

Gijsurugicj:
[EQYMK]    = 9 lwsdmstwx, umu. 55 nujte, bhmq zkjz g tbquqtbh
[APEZRWE]  = HjsgIsvho vyuctrkr kmmwmwku
[DMLJPQQQ] = acdqhq olpbs fjbv 52–318 hhgsqz ozx uxj
[EURV]     = mlywgsbyqiev, kqvn, fh qagkxl
New · DEF prompt generator

Answer a few questions, get a DEF prompt

Choose what you want to make, fill in your goal and click suggestions. You see your prompt take shape right away, with brackets, definitions and a list of what is still missing. You can also add images and code. Free and without an account. The prompt is built in your browser; if you like, you run it directly on the site with Google Gemini. To improve the generator we only store statistics when you run a prompt, and when you send it to WhatsApp also the prompt text (never your full number). We only measure visits if you consent. See the privacy statement.

Open the generator →
01 — Why it works

Words are vague. Definitions are not.

Words like "short", "professional" or "the customer" mean something different to everyone, AI models included. Brackets show which words matter. The list of definitions says exactly what they mean.

[ ]

No guesswork

The model does not have to guess what "short" or "informal" means. You have pinned it down, so the model follows your standard.

↻

Reusable

The instruction stays the same. Want something different? Change just one definition, for example a different [AUDIENCE], and the rest stays as it is.

=

Consistent

A term that appears in five places in your prompt means the same thing everywhere. That gives you fewer contradictory or changing answers.

◎

Clear

The instruction itself stays short and readable. All the details sit neatly at the bottom, in one list.

✓

Easy to check

Is the result off? Then you immediately see which definition to sharpen, instead of rewriting the whole prompt.

⇄

Easy to share

A colleague sees straight away what is fixed and what they can fill in themselves. The prompt explains itself.

02 — How to apply it

A DEF prompt in five steps

Start with your normal prompt. You don't have to throw it away, you just make it more precise.

First write your instruction in one or two sentences

What should the model do? For example: "Write a newsletter for our customers about the new opening hours."

Mark the words that can be read in more than one way

Think of the form (what kind of text), the topic, who it is for, the tone and the length. In the example: newsletter, customers, opening hours.

Put those words in brackets, in capitals

It becomes: "Write a [NEWSLETTER] for [CUSTOMERS] about [TOPIC]." Always use exactly the same spelling.

Write a list of definitions below the instruction

For each term, state in one line exactly what it is. Make it measurable where you can: word counts, dos and don'ts, a concrete example.

Test, and sharpen the definitions

Not happy with the result? Then change the definition that is off, not the whole instruction. After a few rounds you have a prompt you can use again and again.

03 — Before and after

Two worked examples

These are made-up examples to show the method. On the left a normal prompt, on the right the same request as a DEF prompt.

The prompts themselves are blurred. Read the explanation below them, or make your own DEF prompt with the generator.

Example 1

A product description for an online shop

Normal prompt
Cfnzp q saeud, tuvlmd akeqifh fuoflzbijrv fpf emt efe mffzc ymxoxj. Dadw kf duicdknzxgqf zgk jyd, opu ofon cqxd ec sucpz hpbf.
What is "short"? What is "professional but fun"? Who is the text for? And what does "rank well" mean? The model has to fill all of that in itself.
DEF prompt
Fnfnl z [RKXHBTGQLUF] sai [BMEPQZJ], lqjnx ov [NZSTFXRP], ox [TWDG]. Aofbfah [OSMEGAEL] ldv loy azoa [CBJAWC].

Mykbiawjrmf:
[LOWMRXGUMRW] = 64–739 wqvxb, 9 yoqizjzhe qlil 6 shgzgw dqmkhd fozb dwanwvxt
[MWGTBXE]     = lcqgztbv hkwqzynsb ncddo yoqis rmotgy, 214 fk, rgsno nybtoi pphr nge 91 irybk
[MNYHQVWW]    = yrahdiiw azd xylzqgvqw rjum 19–68
[OUWN]        = lwmphfyxdkcf ado vbkaqs, iaqjtgl gco zvihov fn "fxw", em flzxhcednbp ewmyc
[FMHZUOHL]    = "lenpckdmx selrj pafbt jatnpv" oeg "203 gj mhqulf", askh uyydrsf sctj
[DMQBUG]      = wnp yrkafjtq jacrqzrh wvyyna kr ehear, ayawnjp muxxhezydj c wthzqtmv
Length, structure, audience, tone and keywords are fixed. A different bottle? Then you only change [PRODUCT].
Example 2

An email summarising a meeting

Normal prompt
Fvuqyvhgm xjvpm faurnln tllqh zv bn xnojm pf cal wvru. Igctrce mjt qpvb mncjqq gef kqz jbtzp pv ub jfdc. Nhg tga qgfe.
"The team", "main points" and "not too long" leave a lot open. Nor does it say anywhere what the action items should look like.
DEF prompt
Rrkhgllkm [XQJNK] ga vj [DRHCF] ww [CEUPRKQGUY]. Obyjgjh [CRMOWBWKN] jnu [UZDIBDPTPQAZ], fl [ZHPT].

Oomctjrqteh:
[ZFQMM]        = gwg bmgh S wrjou rmopx, eodhbrv ---
[ILGQK]        = rtrysbc jthk + ujd. 205 jfjkt, mempzvlo "Wxqvtdyes" sxi "Cvwnkge"
[OBWWDRBZKN]   = ioieujf nupj pj 0 mbjekn dwh vquzkywu zzh bikbvqp
[EEECJFGDU]    = reot aook dac tsmuquyg fxbayd, aj ciwtvzatvg ouqosp
[WSMBDFUBBUNB] = ncpq gr gmn ehry: kcta – ehkd – pmwvdhjm; hs hrcnbcrxt fm livomjm, haqqs "ej ah mhjggnl"
[ICKS]         = pnafylmcxnfq lmi djkbo, akhxwer cmt utrjqi fb "okx", fk hqunzqp ueucpusrifag

---
(cdaef yurr cwxjo katc)
---
The model now knows what does and does not belong in the email, how long it may be, and what to do when information is missing.
04 — Common mistakes

Tips to get it right

The method is simple, but there are a few pitfalls.

✕ A term without a definition

Every word in brackets must be in the list. A [TONE] without explanation is just as vague as the word tone without brackets.

✕ Vague definitions

[LENGTH] = short solves nothing. Write [LENGTH] = max. 100 words instead. Make it measurable wherever possible.

✕ Inconsistent spelling

[AUDIENCE] in one place and [TARGET AUDIENCE] or [audience] in another is confusing. Pick one form and stick to it.

✕ Definitions that contradict each other

"Max. 50 words" for [INTRO] and "detailed" for [TONE] don't go together. Check your list for contradictions.

✓ Don't put everything in brackets

Ordinary words like "write" or "for" need no brackets. Choose the words that can be read in more than one way or that you want to be able to change later.

✓ Give an example in the definition

Sometimes an example says more than a description: [TITLE] = question form, e.g. "When were you last offline?"

05 — Template

Copy and fill in

A starting point for your own DEF prompt. Replace the text after each = with your definition. Remove the terms you don't need, and add your own.

DEF template
Oaq qdw [RYPW]. Kwrrkg [EEOFWU] inoes [WYMWB] dgy [GETNZLXC], zb [SNFP] gar trmg [RYJAQG]. Bxr [EKFPOLNWS], kwyjex [ZCDKT] ygh kmdsmtr su fk [CKFIXN].

Jkhpwsmneqr:
[LYPG]      = mux sqo ckrpa sxzded yx, p.o. "xepaflxrqnc hzswqvhfbl ett fzofmb jhdis"
[SPXKEW]    = yxoe nmeo sa upue kc ubwlyz, f.a. "d simbdnu gdyfiztbzie"
[XSDVC]     = rzzjgfu kpzn xn mv csuij, lfun xzw bqs rqzat
[AFRJOICF]  = ivg usaw, zx gwoqhudv lv qupmqglo (pof, ahxnlcqtv, oapad ypawvjwik)
[FECE]      = yij fr ghtbsa dxnpq, r.g. "zuwhletfjige, ehzm, xckopuhw, wf mmljhm"
[CUFWNV]    = gnjthxpnos, j.c. "cey. 693 iicpm" de "4 ruvirhfvtb"
[QQGJEUDII] = sci cbhfq hm hsynb, l.u. "ehidxtr, 4 tjpgoceyl, 4 zzzwiy ossqik"
[GTKLN]     = kkga dz ouh eo iqh ktlsaqy, f.l. "ie jyixbllgpqo mszjt, hk cyrvzf"
[VRAHOT]    = phl pii zcm fz ezoq, b.l. "wsein zpom" yd "a jvndp"

The template and the examples are blurred. Make your own DEF prompt with the generator.