# Why your prompts matter

> Your prompts are the questions EOMA puts to the AI engines. What you write in your brand profile becomes those questions — and the wrong question measures the wrong thing.

Source: https://eoma.ai/docs/concepts/why-prompts-matter
Product: EOMA — AI search visibility for brands (https://eoma.ai)
Last updated: 2026-08-14

---
## In 10 seconds

- A prompt is one whole question a buyer might ask an AI assistant — not a keyword.
- What you write in your brand profile is what the questions are built from, so a vague profile produces questions that measure somebody else's market.
- You can add your own questions, in any language. EOMA classifies and locates them for you rather than making you learn the categories.

## The short version

Everything on your AI Visibility screen is downstream of one thing: the list of
questions being asked. Change the questions and every number changes, because
you are measuring a different market.

This is the part of EOMA most worth your attention, and the part founders usually
skip. A question like *"best coffee"* measures a global commodity market you will
never appear in. *"Where can I buy Turkish filter coffee online in Germany?"*
measures the market you actually sell into. Both are valid questions. Only one of
them tells you anything.

EOMA writes your first set for you, from your website, your category and your
markets. Your plan sets two numbers, and they are different on purpose: how many
questions you can **keep** (50 on Starter, 200 on Growth, 300 on Scale, unlimited
on Enterprise) and how many get **asked each month** (30 · 100 · 200 · 400). Keep
more than you ask and the extra questions are not wasted — the rotation reaches
them on a later cycle, oldest first. Then you can edit, deactivate, and add your
own.

A prompt is a **question**, not a keyword. It is asked as a person would type it,
and the engine answers in prose. That is why the questions run up to 500
characters and why they are full sentences.

## How it works

1. **Your brand profile becomes the source material.** Your website, the
   competitors you named, your markets, your industry and your footprint are all
   read when the question set is written. This is why the setup questions during
   onboarding are not paperwork — a brand that leaves "SaaS & Software" sitting in
   the industry field because it was pre-filled gets a question set about SaaS.
   Edit it in [Brand profile](/docs/settings/brand-profile).

2. **Every question gets a category, and the category picks the instrument.**
   There are eight: recognition, market position, sentiment, share of voice,
   share of voice in a niche, product recommendation, buying guide, and category
   discovery. The category decides how the answer is scored and whether the
   question counts towards your share of voice. A miscategorised question is a
   wrong number arrived at silently, which is why nothing is ever defaulted.

3. **Questions you add are classified, not guessed at.** When you submit your own
   questions, EOMA reads them against the same specification its generator writes
   against, and assigns the category the generator would have used. It then checks
   that answer mechanically: three categories *require* your brand name to appear
   in the question, five *forbid* it, and sentiment allows either. A classification
   that breaks its own rule is sent back for a second attempt, and a question that
   still cannot be placed is rejected with a note rather than stored wearing a
   made-up category.

4. **Geography is looked up, never inferred from your words.** The country a
   question is measured in comes from your brand's own market settings. The city
   field, where it is used, describes **where the person asking is sitting** — not
   what they are asking about. Somebody asking about Antalya villas is usually in
   Istanbul. An earlier version of this read the city out of the question text and
   tagged that Antalya question as an Antalya searcher, which measured a narrower
   question than the customer had.

5. **Any language works.** Turkish, German, Arabic — the question is asked as
   written and never translated. Language changes neither the category nor the
   market it is measured in. A Turkish question is classified exactly as its
   English equivalent would be.

6. **Adding questions tops up; it never replaces.** Regenerating your set inserts
   only questions that are genuinely new. Existing questions, their on/off state
   and their whole answer history are untouched. Editing a question's wording
   keeps its identity, so its history continues on the same line with an
   "edited here" marker rather than starting over.

## What this means for you

**Your plan limits how many questions can be *active at once*, not how many you
may own.** Starter tracks 100 active, Growth 200, Enterprise 400. If you are at
the cap, activating another question asks you to switch one off first. Questions
you switch off keep every answer they have already earned.

**Your own questions are capped at 100 per brand, on every plan**, and at 25 per
upload. That pool is the same on every tier because it is a size limit, not a
feature you are being sold.

**A question that names your brand and a question that does not are measuring
different things.** "What do you know about my brand?" tells you whether the
engine has correct information. "What are the best X?" tells you whether it
recommends you. Both belong in your set. Only the second one moves the number
buyers care about.

**If your brand has no market set, your questions cannot be stored.** The country
comes from your market settings, so a brand with none gets a plain refusal rather
than a silent default. Set your market in Brand profile first.

**Watch what your first set actually asks.** It is written from your website, and
a website that undersells one of your lines will produce a question set that
never measures it. Half an hour here changes every number on every other screen.

## Common misunderstandings

| People assume | Actually |
|---|---|
| "Prompts are keywords." | They are whole questions in natural language, up to 500 characters, answered in prose. Keyword thinking produces questions nobody would ever type into an assistant. |
| "I should write my questions in English so the engines understand." | Write them the way your buyers ask them. The engines answer in the language of the question, and that answer is what your buyers would see. |
| "I added a question, so it will be asked in the next run." | Active means eligible, not scheduled. Each run takes a slice sized by your plan, and a new question waits its turn in the rotation. |
| "I should include my city in the question to measure my city." | The location is a setting, not text. Where the *asker* sits is a separate thing from what they are asking about, and confusing the two narrows what you measure. |
| "Deleting a question deletes its data." | Switching one off keeps every answer it earned. Its history stays readable and stays in the pool the score is computed from until the window moves past it. |

## Related

- [Brand profile](/docs/settings/brand-profile) — the facts your question set is written from
- [What a run is](/docs/concepts/what-a-run-is) — how your questions actually go out
- [How your visibility is scored](/docs/concepts/how-visibility-is-scored) — what the eight categories do to the numbers