Skip to content
Depra AI

Docs

Prompts and topics: what Depra AI asks the AI engines

What a prompt, a topic, a tag and an intent label are, how suggestions are written, how slots are counted, and how to choose what to track.

Updated

A prompt in Depra AI is one question your buyers ask AI, such as "best sunscreen for oily skin". Depra AI asks each tracked prompt to the four AI engines on schedule and scores the answers. Read this page before you add, edit or archive prompts.

What are prompts, topics, tags and intent?

Prompts, topics, tags and intent in Depra AI
TermWhat it isExample
PromptOne question a buyer asks AI. Depra AI asks it to the tracked engines on schedule.best sunscreen for oily skin
TopicHow a buyer names what they are shopping for, in 2 to 4 words. A topic holds a set of prompts.Sunscreen For Oily Skin
TagA label you put on prompts to group them.A product line
IntentA label Depra AI gives each prompt automatically.Commercial, Informational or Transactional

Topics cover your product category and the job the buyer wants done. They also cover other ways to get that job done, and what sets your products apart. A single feature becomes a topic only when buyers shop for it on its own.

You can filter the reports by topic and by tag. Tags also let you act on several prompts at once.

Where do suggested prompts come from?

Depra AI suggests up to 5 topics with 8 buying questions each when you set up a project. You tick the questions to keep. The Prompts page lists suggestions on its Suggested tab.

  • No brand name: none of the suggested questions name your brand.
  • Plain first: each topic starts with up to 2 short, plain questions, such as "best sunscreen for oily skin".
  • One detail at most: the other questions add at most one detail each: a market, a budget, an audience or a feature.
  • Your markets: with one market the question names the place. With several markets the place is left out.
  • Hinglish: when India is a market, each question also gets a Hinglish version. Hinglish is Hindi and English typed in English letters.

The Prompts page sorts questions into tabs.

Tabs on the Prompts page
TabWhat it lists
ActiveTracked and paused questions
SuggestedSuggested questions
ArchivedArchived questions
On the Prompts page, each row is one question. For each question Depra AI keeps the English text, the Hinglish version, the topic, the intent, the tags, the status and the last successful run.

What can I change on a prompt?

You can change prompts one at a time or in bulk.

  • Add: write your own prompts alongside the suggested ones.
  • Edit: change the wording. Earlier versions are kept.
  • Pause: stop checking a prompt without removing it.
  • Tag: group prompts, for example by product line.
  • Archive or remove: take a prompt out of tracking. Archived questions are listed on the Archived tab.

Why are prompts that name my brand left out of scores?

A prompt that already contains your brand name almost always returns your brand. Counting it would raise your visibility without showing whether AI recommends you to new buyers. So Depra AI leaves these prompts out of your scores by default.

A prompt is branded when it contains your brand name or one of its name variants. Depra AI works this out for each prompt when you add it.

Branded and non-branded prompts for a brand called Brand X
PromptTypeCounts in scores
is Brand X sunscreen good for oily skinBrandedNo, by default
Brand X vs Brand Y sunscreenBrandedNo, by default
Brand Y vs Brand Z sunscreenNon-brandedYes
best sunscreen for oily skin under ₹500Non-brandedYes
Brand X stands for your own brand. Brand Y and Brand Z stand for other brands.

You can still add a branded prompt and read every answer it gets. A prompt that names only a competitor is not branded for you, so it counts. It shows whether AI brings you up when a buyer asks about a rival. The scoring rules are in metrics.

Note: The Depra AI MCP server is the read-only connection that lets AI apps read your data. Reports read through it follow the same default. They count branded prompts only when you ask for that. See the MCP tools reference.

How many prompt slots does my plan have?

Your plan sets the number of prompt slots. Each language version of a question uses one slot in each market. The markets are listed in engines and markets.

Prompt slots on each Depra AI plan
PlanPrompt slotsProjects
Starter161
Growth702
Scale120, pooled5
Enterprise300, pooled10
Your prompt count is the total across all your projects. On Scale and Enterprise the slots are pooled, so you can move them between brands or clients. See plans and limits.
How markets and languages multiply the slots one question uses
Question tracked inSlots used
English, in 1 market1
English and Hinglish, in 1 market2
English, in 2 markets2
English and Hinglish, in 2 markets4

Starter's 16 slots hold 16 English questions in one market, or 8 questions tracked in English and Hinglish.

Warning: If you exceed your prompt allowance, archive prompts or upgrade to resume tracking. Existing projects and saved results are kept.

Can I track a prompt for one product?

Yes. Add a buying question for one product, such as "best body balm under ₹3,000". Depra AI records whether each answer names that product, a rival's product, or neither.

This works on all four engines, because Depra AI reads the answer text. It is separate from the product cards ChatGPT can show, which are covered in product visibility.

How should I choose prompts?

  • Ask what buyers ask: track questions a buyer asks before choosing, with no brand name of yours in them. Examples are "best X for Y", "X vs Y" and "which X under ₹N".
  • Stay close to what you sell: pick the suggestions closest to your products, up to your plan's slots.
  • Prune: pause any prompt that stops matching what you sell.
  • Count slots first: a question tracked in English and Hinglish uses 2 slots in each market.
  • Read Volume as a rough guide: where data is available, the Volume column shows measured monthly AI keyword demand from DataForSEO. It measures a short English term or a broader category term, not how often people ask the full question.

Keep reading