Prompt & Question Research

Summary#

Prompt & Question Research sends a bank of real buyer-intent prompts to the live AI engines, records which answers mention your brand and which do not, and returns a coverage score, the prompts you win, the prompts a competitor won instead, and the questions your category is asking.

Purpose#

Ranking on Google no longer guarantees you are named in an AI answer. A buyer who asks an assistant "what is the best running shoe for flat feet" sees one paragraph, naming two or three brands. This tool answers the question that follows: which prompts already name us, which name someone else, and what should we publish to change that.

Overview#

Enter a brand or domain. Metric Vault builds twelve category prompts — grounded on your real top-ranking keywords and recent news headlines, and never containing your brand name — then sends every one of them to each AI engine it has a live connection to: ChatGPT, Gemini, Perplexity, Claude and Google's AI Overview. It reads each answer, checks whether your brand appears, records how early in the answer it appears, and keeps the verbatim text.

If you have saved your own tracked prompts for this brand, those are used instead of the generated set. See AI prompt tracking.

Nothing on this screen is written by an AI describing your brand. Every number is a count of measured appearances across real answers.

The screen carries a tab strip with two tools: Prompt Research and Questions. The sidebar entry for both reads Prompt & Question Research.

Benefits#

• One credit buys twelve live prompts across up to five AI engines.
• Separates prompts you win from prompts you are missing, so the work list writes itself.
• Every missing prompt comes with an excerpt of the answer a buyer actually saw.
• Per-engine coverage shows whether the gap is universal or specific to one assistant.
• The Questions tab pairs the AI view with real search demand and difficulty.

Use Cases#

• Content briefing. You need the next five articles to be things buyers ask AI, so you brief from the Opportunity Prompts list instead of a keyword export.
• Competitive reality check. A rival claims category leadership; you check how many of twelve category prompts actually name them and not you.
• Engine triage. Coverage is strong on ChatGPT and zero on Perplexity, telling you the problem is citation-worthy sources, not brand awareness.
• Launch validation. After publishing a comparison page, you re-run to see whether the prompt it targets now names you.
• Question sourcing. The Questions tab gives you real questions with volume and difficulty for an FAQ or a hub page.

Requirements#

• A signed-in account and a paid plan; Free covers the zero-credit technical tools only.
• One credit available. See How credits work.
• A recognizable brand or domain. Very new or private brands often return no appearances at all — a real result, not an error.

Permissions#

Plan decides access; team role does not.

Cost#

1 credit per run, on each tab, as printed on the buttons Analyze Prompts · 1 report and Find Questions · 1 report.

Both are light tools: recorded in your usage but never blocked when the monthly premium allowance is spent, and subject to the 100 light-tool calls per hour fair-use limit. A shared-cache hit still charges the credit — Prompt Research is cached for two days, Questions for seven. Reopening a saved result is free. See How credits work and Result caching and freshness.

Navigation Path#

Dashboard → AI Visibility → Prompt & Question Research

Inputs#

Step-by-Step Guide#

1. Open Dashboard → AI Visibility → Prompt & Question Research.
2. Optional: open the Tracked Prompts button at the bottom right of the app, enter your brand, and add the exact prompts you care about. Runs then use your list instead of the generated one.
3. Stay on the Prompt Research tab.
4. Type your brand or domain into the input marked e.g. nike.com.
5. Click Analyze Prompts · 1 report, or press Enter in the input field.
6. If you have analyzed this brand before, choose Open saved result (free) or Run fresh (1 credits).
7. Wait while twelve prompts run against every connected engine. This is the slowest class of tool in the app because it waits on live model responses.
8. Read the coverage gauge, then work down to Opportunity Prompts.
9. Switch to the Questions tab and run the same brand or a category topic to pair the AI view with real search demand.

Reading the Results#

Prompt Research#

The header names the domain, states how many prompts were tested and across how many engines, and shows a large COVERAGE SCORE dial. Coverage is the share of all brand-by-engine checks in which you appeared — twelve prompts across four engines is 48 checks, and appearing in 20 of them is 42%. The band under the dial reads Dominant at 70 and above, Strong from 50, Building from 30 and Behind below that. Three buttons sit alongside: Export Report, Schedule Daily Tracking and Compare Competitors.

The five KPI tiles.

• Total Prompts — how many prompts ran. Twelve on a generated set; your own tracked list can be shorter or longer.
• Brand Appearances — the raw count of answers naming you, across all engines. This is the numerator behind the coverage score.
• Coverage Score — the percentage, with its band repeated. Under 30 means most buyers asking these questions never hear your name.
• Avg Position — how early in the answer you are named, on a 1-to-10 scale where 1 is the opening words. Lower is better. A good coverage score with a position of 8 means you are the afterthought at the end of a list.
• Top Engine — the engine that names you most often, with its coverage percentage.

Per-Engine Coverage gives one dial per engine: the percentage of prompts on which that engine named you, the raw N of M prompts beneath, and your average position there when one was measured. Read the spread, not just the average. Even coverage across engines is a content problem you can fix everywhere at once; one engine at zero usually means that engine's sources do not mention you, which is a citation problem — check AI Citation Tracker.

Winning Prompts lists up to eight prompts where you surfaced, each tagged Brand surfaces and chipped with the engines that named you. These are assets. Note them and re-test after any site change so you notice when one is lost.

Opportunity Prompts lists up to eight prompts where no engine named you, each tagged Brand absent. Every card carries a 180-character excerpt of the answer a buyer actually received — usually naming your competitors — and a Strategy line. This is the most actionable block on the page: each card is a specific question, a specific answer that beat you, and a content asset to build.

Insights and Action Plan close the report with the measured summary and an ordered list of next steps. The recommendation panel below can expand any of this into a detailed plan; see Get Recommendations.

Questions#

The Questions tab reports real search demand rather than model answers. Its KPI row reads Total Questions, Topic Volume, Live PAA, Quick Wins and Top Intent, followed by an Intent Mix chart (How-To, Comparison, Price, Definition, Problem, Review), a Volume Distribution (<100 through >100K), a Difficulty Distribution (bands 0-10 to 71-100), a Question Cluster Mind-Map and a Top Questions table. Quick Wins are the questions with usable volume and low difficulty — start there.

Examples#

Screenshots#

Tips#

• Add your own tracked prompts before the first run. Generated prompts are good; the questions your sales team actually hears are better.
• Read Avg Position next to coverage. Being named last in eight answers is worth less than being named first in four.
• Copy an Opportunity Prompt verbatim into your article's H1 or FAQ. AI answers reward pages that answer the question in the asked form.
• Re-run after publishing, not before. That is the only way to tell whether the asset worked.

Best Practices#

• Treat Winning Prompts as a defence list and Opportunity Prompts as an attack list, and work one item from each per sprint.
• Pair with Prompt Tracking once you know which prompts matter — this tool finds them, that one watches them over time.
• Use the Questions tab to choose between two Opportunity Prompts: the one with real search volume earns organic traffic as well as AI coverage.
• Keep the country selection consistent between runs so the saved history compares like with like.

Common Mistakes#

• Clicking a TRY chip to explore. The chips fill the field and run immediately, spending a credit.
• Expecting your brand name inside the prompts. Prompts are deliberately brand-free — a prompt naming you would guarantee a mention and measure nothing.
• Reading zero coverage as a bug. For a small or new brand it is the honest answer, and the Opportunity Prompts are the whole value of that run.
• Comparing coverage between different prompt sets. Change your tracked prompts and the score is not comparable with earlier runs.
• Confusing this with Visibility Overview. That gives the overall visibility score; this gives the prompt-level detail behind it.

Limitations#

• Twelve prompts per generated run, ten winning prompts and eight opportunity prompts retained, with eight of each shown.
• Only engines Metric Vault has a live connection to are queried; the report names which engines ran.
• Brand detection is a text match on your brand name, so a brand whose name is a common word can over-count.
• Position is a coarse 1-to-10 estimate of where in the answer you appear, not a ranking.
• Answers vary between runs by the nature of the models; treat single-run movements of a few points as noise.
• The Questions tab covers the United States, English-language index.

Troubleshooting#

FAQs#

Where do the prompts come from? From your tracked prompt list if you have one. Otherwise they are generated from your real top-ranking keywords and recent news coverage, as a mix of category recommendation, comparison, problem-solving and pricing questions — none containing your brand name.

Which AI engines are queried? ChatGPT, Gemini, Perplexity, Claude and Google's AI Overview, limited to those with a live connection at the time of the run. The report lists the engines that answered.

Why did the score move without my changing anything? Language models are not deterministic, and their underlying indexes change. Read the trend across several runs rather than a single delta. For tracked movement over time use Prompt Tracking.

Does a cached result still cost a credit? Yes. A shared-cache hit charges 1 credit. Only reopening your own saved result from the library is free.

Is this available on the Free plan? No. It needs a paid plan, then keeps working after your monthly premium allowance is spent because it costs fewer than 3 credits. See Quotas and rate limits.