AI Guides › Prompting
The Difference Between A Prompt That Informs And One That Persuades
By Nigel Guy · 2 min read
Ask for a "summary" or an "explanation" of something and you can still get
back a document that reads like a pitch — reasons stacked toward a
conclusion, weaknesses left out, confidence higher than the underlying facts
support. This isn't usually because the model decided to persuade you. It's
because the request never actually settled whether you wanted an account of
something or a case for something, and those two documents can be worded
almost identically while doing very different jobs.
The rule: decide, before you write the prompt, whether the reader is
meant to be informed or persuaded — the same words in a different order
produce one or the other, and you get to choose which.
The mechanism: the purpose check
| Informs |
Persuades |
| "Explain," "summarise," "describe" |
"Make the case for," "write copy that," "convince" |
| Includes drawbacks and counterarguments by default |
Omits or downplays weaknesses by default |
| Reader is still deciding |
Reader has already decided, or isn't the one deciding |
| Confidence matches the actual state of the evidence |
Confidence is pitched to move the reader |
- Ask yourself who's reading this and what they're doing with it. Are
they deciding, or being sold to after a decision's already made
elsewhere?
- If you want to inform, say so explicitly — and ask for the drawbacks
too. A neutral-sounding request doesn't automatically produce balance;
tell it directly to include the weaknesses and counterarguments, because
a persuasive default won't volunteer them unprompted.
- If you want to persuade, say that too, rather than leaving the purpose
implicit. Don't make the model guess whether you want advocacy or
balance — state which one you're asking for.
- Reread the result against the purpose you actually intended. Does
the balance of what's included match "informs" or "persuades" as you
defined it, not just as it happened to come out?
What to skip
Skip trying to get one document that's genuinely both at once — a balanced
account and a persuasive pitch pull in different directions by design.
Write the one you actually need, and adapt it afterwards for the other
purpose if you need both.
Guardrails
- Asking to "inform" doesn't guarantee neutrality on its own — the model
can still default to a more confident, more polished register than the
underlying facts actually support.
- If the output is going somewhere it'll be relied on as a factual account
— a briefing, a report, a decision memo — verify that it's genuinely
balanced rather than trusting that the instruction alone produced that.
- This distinction is about framing, not accuracy. A persuasive document
and an informative one can both contain errors; neither purpose is a
substitute for checking the facts.
All 751 AI guides · JulieMango plans from £17/mo