Skip to main content
AI Search Prompts page with top cards and prompt-level performance table
Use this page to decide whether your tracked prompt set matches real buyer demand.

Important: Better prompt coverage beats higher prompt count. Focus on business-critical prompts first.

Questions this page should answer

  1. Are we tracking the prompts that matter most for pipeline?
  2. Which prompts are underperforming right now?
  3. Which prompt gaps should become this sprint’s content work?

Before you analyze

  • Review the active prompt list before judging performance.
  • Keep the same reporting window used in other AI Search pages.
  • Compare by intent category, not only by total rows.

What this page gives you

  • Top prompt health indicators like average visibility and average position.
  • Prompt-level table with position, mention frequency, top results, and category.
  • An AI search engines control for choosing which plan-supported engines run each active prompt.
  • Manage prompts flow with one add workspace that combines manual entry and AI-generated suggestions.
  • A decision layer to connect prompt gaps to page updates.

How prompt data is collected

Atomic runs each active prompt on the engines selected in AI search engines. Available engines and automation frequency depend on your plan. We then compare patterns over time, because AI responses naturally vary between runs. These tracked prompts should be written as conversational questions (for example What's the best CRM for marketing agencies under 50 people?), not short keyword fragments. To mirror real buyer behavior, Atomic collects prompt results through AI product interfaces, not only via API outputs. This helps you measure what users are likely to see in real interactions. Why this matters:
  • Authentic experience: data reflects the same interface-level outputs users interact with.
  • Real-world accuracy: results are less abstract than API-only snapshots.
  • Broader coverage: you can track platforms and model flows with limited public API access.

Choose tracking engines

Open Manage prompts. When engine setup is required, choose the available engines in AI search engines and select Continue before adding prompts. To change an existing selection, open AI search engines in the prompt workspace and select Save engines. The choices and maximum engine count depend on your plan; check the capacity calculation before continuing. Frontier Analyst and Frontier Growth include DeepSeek with web search. The Manage prompts picker currently omits DeepSeek. To check one answer on either plan, use the MCP prompt test with deepseek as the engine. A test uses one run and leaves scheduled tracking unchanged.
Manage prompts on Frontier Analyst with ChatGPT and Gemini selected and capacity for 400 active prompts
Your tracking configuration has a per-execution capacity shared across selected engines:
For example, the Starter plan’s capacity of 75 prompt-engine pairs can cover 75 prompts on one engine, 37 prompts on two engines, or 25 prompts on three engines. Selecting fewer engines increases distinct-prompt capacity. Historical results remain available after an engine is deselected; only future runs use the new selection. If a plan change makes the current selection exceed capacity, tracking pauses without deleting prompts or engine choices. Open AI search engines and reduce the engine count, or deactivate prompts, until the configuration fits.

Billing-cycle run allowance

In the plan selector, this allowance is labeled AI answers per month. It is separate from the number of prompts you can track and from AI credits. Tracking capacity is different from the number of runs consumed during a billing cycle. Each execution of one prompt on one engine uses one run. Running 25 prompts on three engines uses 75 runs each time that tracking job executes. Scheduled tracking and one-off MCP prompt tests draw from the same allowance. A test does not save the prompt or turn on ongoing tracking. Failed attempts also use a run. Requesting a prompt and engine that already has a pending run returns that run without using another. When the allowance is exhausted, new runs are rejected until allowance is available. Reducing the engine count lowers future consumption but does not restore runs already used. A successful billing renewal resets the used-run count; changing plans does not reset usage. See Billing for the plan allowances. If results stop refreshing, check both whether the tracking configuration fits the plan and whether its billing-cycle run allowance is exhausted.

How to read the top cards

  • Avg. visibility percentage: position-weighted visibility across evaluated answers.
  • Your avg position: average placement quality.
Use both:
  • Better position with low visibility means narrow coverage.
  • Visibility up with weak position means presence without competitiveness.

Key signal: Prioritize prompts where mention frequency is high but your average position is weak. Those are usually your fastest visibility gains.

How these metrics are calculated (simple)

Avg. visibility percentage

Visibility averages position-weighted scores across evaluated answers. First place scores 100, second 85, and missing observations 0. It is distinct from mention percentage. See AI search metrics for the complete scale.

Your avg position

The comparison average includes ranked positions and a position-10 penalty for each missing observation. This penalty does not imply an actual tenth-place result. See AI search metrics. Lower value is better (closer to top placement).

Mention frequency

Failed, pending, and unclassified answers are excluded. Frequency labels are High (75% or more), Medium (25% to below 75%), and Low (below 25%). See AI search metrics.

How to read the prompt table

  • Your position: current placement for that prompt.
  • Mention frequency: how often your brand appears.
  • Top results: recurring sources shown in answers.
  • Category: funnel context.
Prioritize prompts that are high intent, high frequency, and weak in position.

How to add new prompts

Select Manage prompts to open the prompt workspace. Complete engine setup first if shown, then enter prompts in the main panel and use the groups sidebar on the right to organize the tracked set. The sidebar shows how many prompt slots are used for the selected engine count.

Suggested prompts

Manage prompts workspace with suggested prompts, Accept and Deny controls, and tracked prompts below
Expand Suggested prompts to review ideas and their explanations. Select Accept to add an idea to tracking, or Deny to dismiss it. Check the wording against your campaign goals before accepting. The screenshot uses illustrative local prompts. How suggestions are generated:
  • When AI traffic is available, Atomic analyzes top AI search pages from the last 30 days.
  • Otherwise, Atomic uses your domain, project business context, and project Knowledge, including synced website content. GA4 is not required.
  • If website content has not been synced to Knowledge, Atomic loads the domain homepage and uses its text alongside project Knowledge. If the page cannot be loaded, retry or sync website content before generating suggestions.
  • The system generates prompt ideas across funnel categories (awareness, consideration, decision, branded).
  • Each suggestion includes a confidence score to help you prioritize.

Manual entry

Manage prompts workspace with one prompt per line, a funnel stage selector, group controls, and Add prompts
Use manual entry when you already know the exact prompt to track (for example from sales calls, campaign briefs, or competitor monitoring). How to add manually:
  1. Select Manage prompts and confirm the AI search engines if prompted.
  2. Enter one prompt per line in the text area.
  3. Choose the funnel stage and a group, or keep Auto group.
  4. Select Add prompts. Check the tracked list and the remaining capacity in the groups sidebar.

Quick weekly checklist

  1. Review decision-intent prompts first.
  2. Remove low-signal prompts that do not map to business goals.
  3. Add missing high-intent prompts from sales conversations.
  4. Assign one action per priority prompt.

How to use filters

  • Use category filters to separate awareness and decision queries.
  • Use date range to confirm trend stability.
  • Change one filter at a time when diagnosing.

What to fix first

Team routine

  1. Weekly: refresh priority prompt backlog.
  2. Bi-weekly: audit category balance.
  3. Monthly: align prompt coverage with pipeline goals.

Keep in mind

  • Prompt quality is more important than prompt count.
  • One viral query can distort short windows.
  • Repeated gaps are more important than one-off misses.

Where to go next

See Query fan-out to inspect searches captured across saved prompt runs and compare their evidence.