Prompts is the foundation of Meridian. Everything you measure in Analytics starts with the prompts you track. If you track the right prompts, Meridian can tell you where you are being recommended, where competitors are winning, and what to do next. This page covers the Prompts tab end-to-end, including how to add and organize prompts, and how to use Analyze to inspect platform breakdowns, run-level results, and the full response (including citations and shopping output when available).

​

What prompts are (and why they matter)

Prompts are the questions Meridian tracks across AI platforms. They define what you are measuring. A helpful way to think about prompts is: “What are the queries we want to show up more for in AI search?” Meridian reports on (and help you improve visibility) for the prompts you track. If you track the right prompts, you get a clear view of:

• where you are being recommended today,
• where competitors are winning,
• and which actions (content, off-page, website fixes) are most likely to increase visibility over time.

If you track the wrong prompts, the dashboard can look “good” without actually measuring discovery.

​

How often prompts run (daily)

Meridian runs each tracked prompt once per day across the AI platforms/models enabled in your account. This daily cadence is important because AI answers can vary day to day, so you want enough repeated runs to see real trends instead of one-off noise.

​

Choosing the right prompts matters (a lot)

It is completely normal, and often more valuable, to add prompts where you currently have no or low visibility, especially higher-funnel “best X”, “top X”, and comparison prompts. Try to avoid building a prompt set that is mostly:

• too branded (e.g., “Is [Your Brand] good?”, “What is [Your Brand]?”), or
• too company-specific (queries only someone who already knows you would ask).

Those prompts often produce artificially high visibility because the assistant already expects your brand to appear. That can make performance look strong while not actually measuring discovery. A healthy prompt set usually includes:

• discovery prompts (“best __”, “top __”, “___ alternatives”),
• comparison prompts (“__ vs __”),
• use-case prompts (“__ for beginners”, “__ for [specific need]”),
• and a smaller set of branded prompts for monitoring reputation.

​

What you can do in Prompts

Prompts is designed for four core workflows:

1. Build a prompt set that matches how customers ask AI for recommendations.
2. Organize prompts into Topics so you can evaluate performance by category/product line.
3. Tag prompts with Attributes so you can filter and segment.
4. Diagnose performance using Analyze and run-level evidence.

​

How the Prompts table is structured

Each row in the table is a prompt you are tracking. For each prompt, Meridian shows:

• Topic: the category the prompt belongs to.
• Visibility score: how often your brand is mentioned for this prompt.
• Share of voice: your share of mentions compared to tracked brands.
• Prominence: your average rank when mentioned (lower is better).

These prompt-level metrics roll up into Home and into topic-level summaries when you group by Topic.

​

Adding prompts (three ways)

Click Add prompts to open the menu. You can add prompts in three ways:

1. Add Prompt (one at a time)
2. Add Topic (create a topic and optionally auto-generate prompts)
3. Upload CSV (bulk import)

​

1) Add Prompt

Use Add Prompt when you are adding a handful of high-value prompts manually. You can set:

• the prompt text,
• the topic,
• optional attributes.

​

2) Add Topic (and auto-generate prompts)

Use Add Topic when you want to scale quickly. A topic is usually a category or product line. You can optionally enable Generate 10 relevant prompts for this topic, which is a fast way to create a baseline prompt set for that category.

​

3) Upload CSV

Use Upload CSV when you already have a prompt library and want to import at scale. The upload modal lists required and optional columns. In the UI shown:

• Required columns: prompt, topic
• Optional columns: attributeId, attributeValue

​

Time range and previous period comparison

At the top of the Prompts tab you can set:

• a date range (for example: Last 7 days),
• and compare against Prev. Period.

This helps you answer “are we improving week-over-week?” without manually tracking numbers.

​

Filters (scoping your analysis)

Filters narrow the dataset you are looking at. Use filters when you want to answer questions like:

• “How are we doing on ChatGPT only?”
• “Which prompts in Topic X are consistently missing us?”
• “How are we doing on a specific segment of prompts (attributes)?”

​

Group by: Topic (and why it matters)

Prompts supports grouping so you can evaluate performance at a higher level than a single prompt. Group by: Topic is the most common view because it answers:

• “Which categories are we winning?”
• “Where are we consistently missing or ranked low?”
• “Which topic should we invest in next?”

When grouped, each topic section shows a rollup of:

• visibility score,
• share of voice,
• prominence,

plus the prompts inside the topic.

​

How to use topic rollups

• If a topic has strong visibility but weak prominence, you are being included but not preferred. That usually calls for better comparisons, proof, and citations for that topic.
• If a topic has low visibility, you likely have a coverage or authority gap for that category.

​

View settings

View settings lets you change how the table is displayed (for example, which columns you see). Use it to keep the table focused on the metrics your team cares about.

​

Prompt usage and limits

At the top right you may see a usage indicator like “97 / 300 prompts.” This shows how many prompts you are tracking relative to your plan limit.

​

Exports

Use Export when you want to share performance with teammates or archive a reporting snapshot.

​

Attributes (segmentation)

Attributes allow you to tag prompts so you can filter and compare different slices of your prompt set. Examples:

• Intent: Research vs Decision
• Audience: Beginner vs Expert
• Internal: Campaign tags

Click Attributes to manage and create attributes.

​

Creating a new attribute

You can create custom attributes with types like:

• Text
• Number
• Single select
• Multi select

​

Analyze (run-level evidence)

Analyze is where you stop looking at aggregate scores and inspect what actually happened in the underlying AI responses. Use Analyze when:

• a prompt is important and you want to understand why you are winning or losing,
• you see a sudden change in visibility or prominence,
• you want to compare platform performance (ChatGPT vs Gemini vs Perplexity),
• or you want to see which sources are being cited in the answers.

​

How to open Analyze

From Analytics → Prompts, click Analyze on a specific prompt (or open the prompt row if your UI opens analysis directly). Once you open Analyze, you’ll see three layers of detail:

1. platform-level performance,
2. trend over time,
3. and the run table (the evidence).

​

Platform Breakdown

The Platform Breakdown chart shows how performance differs by platform. This matters because different platforms rely on different sources and can return different ranked lists.

​

How to interpret Platform Breakdown

• If one platform is much lower, you often need a platform-specific plan (usually citations/off-page).
• If one platform is high and others are low, your content and source ecosystem likely match the “high” platform better than the others.

A practical next step is to open runs from the weaker platform and inspect which sources are cited.

​

Brand Visibility trend

The Brand Visibility chart shows visibility over time for the selected prompt. It helps you spot:

• when a drop began,
• whether changes are consistent or noisy,
• and whether there are clear competitor-driven shifts.

​

How to use the trend

• A one-day dip is often noise.
• A sustained dip across multiple days is usually a real change worth investigating.

​

Response table (the evidence)

The response table shows each run for the prompt, including:

• Date
• Mentioned (Yes/No)
• Position (if mentioned)
• All mentions (icons showing which brands/products appeared)
• Response snippet
• Platform
• Region

​

How to interpret the run table

• If you have many Not mentioned rows, you have a coverage/authority gap for this prompt.
• If you are mentioned but positions are #4+, you are included but not preferred. That often points to positioning, proof, or citation gaps.
• If performance varies widely by platform, focus your investigation on the weaker platform’s runs first.

​

Opening a response (Mention details)

Click a response row to open the detailed response modal. This is where you can see the full answer, mentions, shopping output (if available), and citations.

​

What you see at the top

• Prompt name
• Properties (platform, date, region)
• Mentions list (ranked entities)

This tells you immediately:

• whether you were included,
• who outranked you,
• and which competitors appear in the same answers.

​

Response tab

The Response tab shows the full answer text. This is useful for understanding:

• what criteria the answer uses,
• how the answer frames the category,
• and what “proof” it relies on (often reflected in citations).

​

Shopping tab (when available)

Some prompts produce a shopping-style output, such as products with images, prices, and retailers. The Shopping tab shows that view when it is present. If Shopping output exists for your category, it can change the strategy:

• product page structure matters more,
• and citations/sources can influence which products appear.

​

Citations tab

The Citations tab shows the sources referenced in the answer (domains/URLs). This is often the most actionable part of a run because it tells you what the model trusted when forming the answer.

​

How to use citations in a losing run

If a competitor outranks you and the answer cites a specific editorial page or domain, that source is often a high-leverage target for:

• Off-page → Outreach (earn inclusion), or
• improving your own source pages so you can be cited instead (Website → Pages), or
• publishing content aligned to what the source covers (Opportunities / Content).