PromptSpark.ai Docs
Sign in
Browse documentation
Docs Prompts

Prompts

Create, import, organize, pause, and inspect the questions PromptSpark.ai sends to AI providers.

What a prompt controls

A prompt is the exact question monitored across its selected providers. It can also carry market, country, city, language, lifecycle, intent, topic, and tag context for filtering and reporting.

Prefer natural questions with a clear decision or information need. Avoid embedding the answer, listing every competitor, or mixing several unrelated questions into one prompt.

Field Purpose
Providers The AI platforms that receive this prompt.
Market and language The audience and locale represented by the question.
Topic One broad reporting folder for the prompt.
Tags Multiple flexible labels for segments such as audience funnel stage or campaign.
Lifecycle state Whether the prompt is active inactive or archived.

Add and manage prompts

  • Use suggestions for a fast starting set, then reject questions that do not match genuine buyer intent.
  • Add a prompt manually when wording, locale, market, or provider selection needs precise control.
  • Import UTF-8 CSV for larger governed sets; review validation errors before assuming every row was created.
  • Pause a prompt to stop future scheduled execution without losing its historical evidence.
  • Use an ad-hoc query for a one-off simulation that should not join the scheduled prompt set.

Read a prompt detail page

The prompt detail page narrows the evidence to one question. Response count shows completed provider answers, Visibility shows how often the tracked brand appeared, Cited Domains counts explicitly cited websites, and Avg Fanout describes the average search-result evidence per response.

Use the trend, provider, brand, search expansion, domain type, and answer panels to understand why the same question performs differently across runs or models.

Review prompt quality before scaling

{"A useful prompt represents one realistic customer question and produces answers that can be compared over time. Read the question aloud" => "it should sound like something a person could ask an AI assistant without knowing your monitoring strategy. Keep the intended audience, decision, and market clear."}

  • Use category questions to measure discovery before a customer already knows your brand.
  • Use comparison and recommendation questions to understand competitive choice and positioning.
  • Use problem-led questions to reveal the language and sources connected to a customer need.
  • Avoid near-duplicate wording that adds cost without adding a distinct intent or audience.
  • Avoid leading claims such as “Why is our brand the best?” because they distort neutral visibility measurement.
  • Review generated suggestions before tracking them; automation supplies candidates, not final research design.