Getting Reliable Numbers from Claude

Claude does not calculate anything. It calls a Superlines tool, receives a JSON payload, and writes it up. Every mismatch between “what Claude said” and “what the dashboard shows” comes from one of three things: Claude picked a different date range, Claude picked a different tool than the dashboard view you are comparing against, or Claude summarised across two payloads that were not comparable. This page shows how to remove all three.

Rule of thumb: tell Claude the brand, the exact dates, and the metric name as it appears in the dashboard. Then ask it to quote the dataFreshness.latest_response_at timestamp. If that timestamp matches the “Data updated” indicator in the dashboard, both are reading the same data.

The four things every data question needs

Include	Why	Example
Exact brand name	Without it, several tools return all your brands combined and Claude may not notice the warning.	`for the brand "Acme Insurance"`
Explicit start and end date	MCP tools default to the last 30 days including today. The dashboard defaults to the last 28 days ending yesterday. Different windows, different numbers.	`from 2026-08-15 to 2026-09-14`
End date = yesterday	Today’s collection is still running. The dashboard hides today unless you toggle “Include current date”.	`ending yesterday (2026-09-14)`
The metric name from the dashboard	”Visibility”, “mentions” and “share” each map to different fields. Naming the metric stops Claude picking a neighbour.	`Brand Visibility`, `Share of Voice`, `Citation Rate`

Copy-paste prompts

Match a dashboard KPI

Using Superlines, for the brand "Acme Insurance": call analyze_metrics with
startDate 2026-08-18, endDate 2026-09-14, metrics brand_visibility and citation_rate,
granularity total. Report the exact values and the dataFreshness.latest_response_at
timestamp from the response. Do not round.

Compare against the dashboard with the same dates and Include current date off.

Brand leaderboard (Brands by Brand Visibility)

Using Superlines, for the brand "Acme Insurance": call analyze_brand_mentions with
startDate 2026-08-18, endDate 2026-09-14, consolidate true, limit 20.
List the brands with their mention_count in a table, exactly as returned.

consolidate: true merges name variants (“OP”, “OP Ryhmä”, “OP Pohjola”) the same way the dashboard does. Without it you get raw extracted names and the list will not match.

Period-over-period change

Using Superlines, for the brand "Acme Insurance": call get_period_comparison with periodDays 28.
For each metric, show current value, previous value and the change, using only the
numbers in the response. State both date ranges the tool used.

Per-engine breakdown

Using Superlines, for the brand "Acme Insurance": call analyze_metrics with
startDate 2026-08-18, endDate 2026-09-14, metrics brand_visibility,
groupBy llm_service. Show one row per engine, values as returned.

Which engines show which competitors

Using Superlines, for the brand "Acme Insurance": call get_visibility_heatmap with
startDate 2026-08-18, endDate 2026-09-14. Show the table exactly as returned
and tell me which engine has the lowest visibility for Acme Insurance.

A verifiable summary for a client

Using Superlines, for the brand "Acme Insurance" from 2026-08-18 to 2026-09-14:
1. analyze_metrics → brand_visibility, citation_rate, mentions, share_of_voice
2. analyze_brand_mentions with consolidate true, limit 10
3. get_period_comparison with periodDays 28
Write a one-page summary. Every number must come from a tool response. After each
number, put the tool name in brackets. End with the dataFreshness.latest_response_at
timestamp so I can check it against the dashboard.

Questions to avoid, and what to ask instead

Don't: “How is Acme Insurance doing?”

No dates, no metric. Claude will pick the tool’s default window (30 days including today) and whichever metric it finds first. The result cannot be checked against anything.Ask instead: “For Acme Insurance, what was Brand Visibility from 2026-08-18 to 2026-09-14? Use analyze_metrics.”

Don't: “Compare this month to last month”

“This month” is ambiguous (calendar month vs. last 30 days) and today is partial. Claude may also call two tools with slightly different windows and compare them.Ask instead: “Call get_period_comparison with periodDays 28” — one call, two aligned windows, or give both date ranges explicitly.

Don't: “What's our visibility?” when you mean the leaderboard

analyze_brand_mentions returns a brand_visibility field that is the brand’s share of all mentions, not the dashboard’s Brand Visibility (share of responses that mention the brand). They are different metrics with the same name.Ask instead: For the KPI, use analyze_metrics → brand_visibility. For the leaderboard, use analyze_brand_mentions and read mention_count and share_of_voice.

Don't: “Add up the daily numbers”

Daily Brand Visibility values are percentages of different denominators. Summing or averaging them does not reproduce the period total.Ask instead: call the same tool with granularity: total for the period.

Don't: “Roughly how much…” / “About what percent…”

This invites rounding and estimation. The server instructs Claude to report exact values; asking for approximations works against that.Ask instead: “Report the exact values returned. Do not round.”

Don't: ask about a brand without naming it

If your API key has access to several brands and you omit the name, tools return everything combined and add a warning field. Claude should relay it, but the numbers will already be in the conversation.Ask instead: always say for the brand ”…” using the exact name from list_brands.

Don't: mix MCP output with a screenshot from a different day

Collection runs continuously. A dashboard screenshot from the morning and an MCP answer from the afternoon can legitimately differ.Ask instead: compare dataFreshness.latest_response_at in the MCP response with the dashboard’s Data updated indicator. Same timestamp, same data.

Reading `dataFreshness`

Every analytics tool response now includes:

"dataFreshness": {
  "latest_response_at": "2026-09-15T05:41:12.000Z",
  "latest_response_date": "2026-09-15",
  "hours_since_latest": 9,
  "computed_at": "2026-09-15T14:36:02.114Z",
  "domain_ids": ["abc123"]
}

latest_response_at is the newest AI response collected for the brand. The dashboard shows the same value as Data updated … ago in the filter bar and as Last collected inside the date picker.
If hours_since_latest is above 36, collection is behind. Numbers are still correct for the data that exists, but the most recent day may be incomplete. Claude is instructed to warn you.
Two answers are only comparable if their latest_response_at values match.

Setting expectations with clients

If you present Superlines data to your own clients:

Pull the numbers with one of the prompts above, with explicit dates.
Open the dashboard with the same dates and Include current date off.
Check that Data updated matches latest_response_at.
Quote the date range and the “data as of” timestamp on the slide.

If a number still differs after those steps, send us the prompt, the tool response, and a dashboard screenshot — that is a defect on our side, not a usage problem.

MCP Server

REST API

Integrations

Getting Reliable Numbers from Claude

The four things every data question needs

Copy-paste prompts

Match a dashboard KPI

Brand leaderboard (Brands by Brand Visibility)

Period-over-period change

Per-engine breakdown

Which engines show which competitors

A verifiable summary for a client

Questions to avoid, and what to ask instead

Reading `dataFreshness`

Setting expectations with clients

​The four things every data question needs

​Copy-paste prompts

​Match a dashboard KPI

​Brand leaderboard (Brands by Brand Visibility)

​Period-over-period change

​Per-engine breakdown

​Which engines show which competitors

​A verifiable summary for a client

​Questions to avoid, and what to ask instead

​Reading dataFreshness

​Setting expectations with clients

The four things every data question needs

Copy-paste prompts

Match a dashboard KPI

Brand leaderboard (Brands by Brand Visibility)

Period-over-period change

Per-engine breakdown

Which engines show which competitors

A verifiable summary for a client

Questions to avoid, and what to ask instead

Reading `dataFreshness`

Setting expectations with clients