Prompting POM AI
POM AI prompts are shorter than people expect. The tools already know your business, your language and the task; the prompt supplies only what they cannot infer.
Start here
Principles
- Anatomy of an effective prompt — what a prompt is made of.
- Describe context, goal and constraints — the three parts worth naming.
- Request structured output — asking for a shape, not just a subject.
- Use examples and reference copy — showing instead of describing.
- Iterate on a prompt — how to improve one without wasting credits.
- Keep sensitive data out of prompts — what must never go in a prompt field.
Reusable prompts
- Use saved prompt templates — what templates are and how they behave.
- Create a prompt template
- Edit and delete prompt templates
Recipes
What you do not need to write
This is the most useful thing to know before writing your first prompt. POM AI already sends, automatically:
- Your business context — the description fields from Settings → POM AI → AI Setup;
- The site language, taken from the site locale;
- The task, implied by the tool you are using;
- The current content, in the editor assistant;
- The structure, in the multi-stage generators;
- The style and length, from the form controls.
Restating any of this in the prompt wastes input, and in the worst case contradicts the setting you have configured. If you find yourself writing "we are a friendly, approachable company" in every prompt, that belongs in the business description instead. See Provide useful business context.
Length limits
| Field | Limit |
|---|---|
| Long-form post prompt | 100,000 characters |
| Image generator and enhancer prompt | 3,100 characters |
| Prompt template | 4,000 characters |
| Other tool prompts | See the counter below the field |
Every field shows a live counter and blocks submission when the limit is exceeded.
These are ceilings, not targets. The long-form builder's own help text warns that extending a prompt beyond the length of its supplied example is unnecessary and tends to produce worse results.
Prompt length affects cost
The prompt is part of the input, and cost is driven by input plus requested output. A long prompt costs slightly more on every run that uses it. POM AI therefore selects only the relevant structured business-profile sections for each tool and caps the compiled context sent with one request.
That is a small effect per request and a real one at scale. Write what is needed and stop.
The controls that beat prompting
Several things are set with a form control rather than words, and the control always wins:
| Instead of writing | Use |
|---|---|
| "about 800 words" | The word count field |
| "in a formal tone" | The copywriting style selector |
| "as a studio photograph" | The style preset |
| "widescreen" | The aspect ratio selector |
| "high quality" | The model profile |
A prompt that fights a control produces inconsistent results. Set the control and say nothing about it.
A working method
- Set the controls first.
- Write two or three sentences of prompt covering only what the controls cannot express.
- Estimate — it is free — and read the number.
- Generate, read the result, and change one thing.
- When a prompt works, save it as a template so nobody has to rediscover it.
Where prompts do not exist
Two POM AI workflows have no prompt field at all:
- Translation takes no prompt. Its context comes from the business short description and the language pair. See Provide context and terminology.
- Image metadata takes no prompt. It describes what it sees.
For both, the way to influence output is configuration and review, not prompting.