Looker Studio + Claude: a Prompt Library for Marketing Dashboards (12 Chart Types and the Questions to Ask)
Contents
A few weeks ago a CMO (Chief Marketing Officer, 首席营销官) showed me her team "executive" Looker Studio dashboard. Eight pages. Forty-three charts. Most of them line charts. She asked me what I thought. I told her the truth: it was a beautiful wall of noise.
Nothing on that dashboard was wrong. Every chart loaded, every metric was technically accurate. The problem was that no chart on the page answered a question a CMO would actually ask on a Monday morning. There was a line chart of "sessions" with no benchmark, a bar chart of "top channels" with no time anchor, a stacked area chart of "all traffic ever" that nobody could read past 2024.
I rebuilt that dashboard in two afternoons using Claude as my thinking partner. It cut to five pages and eleven charts, and the CMO now opens it every Monday before her standup. The trick wasn't a fancier template or a cleverer visualization. The trick was asking one specific question before I opened Looker Studio for each chart, then handing that question to Claude to pressure-test the choice.
This article is the prompt library I built from that engagement. Twelve chart types, the question to ask before each, and a copy-paste Claude prompt that helps you design the right chart for the right decision. None of this is theory. Every prompt here was used in a real client engagement in the last six months.
Why most marketing dashboards fail
Before the library, the framework. The reason most marketing dashboards end up unread is that the people building them start with data, not decisions. They connect a Google Analytics 4 property, drag every dimension and metric into a chart, and call it done. The result is a literal database dump with axes.
Looker Studio is unusually prone to this. Unlike Tableau or Power BI (Business Intelligence, 商业智能工具), it has no enforced semantic layer (a data-modeling layer that defines what each metric means, called a "LookML model" in Looker). You can wire up any source, and Looker Studio will gladly render it. The tool's flexibility is the failure mode.
Claude's role is not to generate dashboards. Claude's role is to do the thinking that the dashboard should have done before it was built. Used right, Claude is the metric dictionary, the QA reviewer, and the chart-selection rubric that Looker Studio does not provide out of the box. Used badly, Claude is a faster way to make a worse dashboard.
The two principles I follow:
- The question beats the chart. If you cannot phrase the question a chart answers in one sentence, the chart is decoration. The question is the spec; the chart is the rendering.
- One chart, one decision. If a chart is doing two jobs (showing trend and showing composition), split it. A chart that does two things is a chart nobody can read.
Every prompt below is built around those two rules. Pass them to Claude, get back a chart spec, then build it in Looker Studio.
The 12 chart types
A few notes before the list. For each chart, I give you:
- The question to ask first — one sentence, the decision-anchoring question
- When to use it — the data shape it fits
- The Claude prompt — copy-paste, fill in the brackets, get a chart spec
- The Looker Studio quirk — a specific thing this chart does in Looker Studio that catches people
I've deliberately avoided any chart that requires code or custom visuals. Looker Studio's library is what it is — line, bar, pie, geo, scatter, pivot, table, bullet (via a community viz), heatmap (community), area, combo, funnel, treemap. That is your menu. Every prompt below maps to one of those.
1. Line chart — trend over time
The question to ask first: Is something going up, going down, or holding steady, and over what window?
Line charts are the default in Looker Studio and the most commonly misused. People stack seven metrics on one line chart and call it a "trend view." That is not a trend view. That is a spaghetti chart. A line chart answers a one-variable-over-time question. The moment you need a second variable on the same chart, you want a combo.
When to use it: One continuous metric tracked across a regular time interval (daily, weekly, monthly). Sessions, conversions, revenue, ROAS (Return on Ad Spend, 广告支出回报率) — all are fine. The metric must be summable or countable; line charts do not work for survey data or categorical data.
The Claude prompt:
I'm building a line chart in Looker Studio. The decision-maker needs
to see [TREND: e.g. "whether organic traffic is recovering after the
site migration"]. The metric is [METRIC: e.g. "Sessions"] from
[SOURCE: e.g. "GA4 property X"]. The natural time grain is [GRAIN:
"daily" or "weekly"]. The relevant window is [WINDOW: "last 90
days"].
Ask me 3 clarifying questions before specifying the chart. Then give
me: (1) the exact dimension and metric configuration, (2) the date
range type to use (fixed vs relative), (3) one trend benchmark to
overlay (e.g. previous period, target, or forecast), (4) the
biggest mistake someone would make building this chart.The Looker Studio quirk: Date range type matters. "Last 28 days" is relative — it moves with today. "Mar 1 to May 31" is fixed. If you want a year-over-year overlay, set the date range to fixed and a year back, or use the "compare to" feature. Don't trust default "auto" date ranges in a published dashboard; they will silently change the data behind your screenshots.
2. Bar chart — comparison across categories
The question to ask first: Which category is biggest, smallest, or beating the others?
A bar chart is for ranking. Looker Studio's bar chart is sortable on the dimension, which means readers will sort it. Design for that. If the answer to "which channel drives the most conversions" changes every Monday, a bar chart is correct. If the answer is always the same (paid social is always #1), you don't need a chart — you need a one-line callout.
When to use it: Categorical comparison of one metric across 2–15 categories. Past 15 categories, the chart becomes unreadable and you want a ranked table instead.
The Claude prompt:
I'm building a bar chart in Looker Studio. The decision is
[DECISION: e.g. "where to reallocate Q2 ad budget"]. The metric is
[METRIC: e.g. "Conversion value"] and the categories are
[CATEGORIES: e.g. "Google Ads campaigns"]. The grain is
[GRAIN: e.g. "last 30 days, campaign-level"]. I expect roughly
[NUMBER: e.g. "8–12"] categories.
Return: (1) the sort order that makes the decision obvious, (2)
whether to use horizontal or vertical bars given the category name
length, (3) the secondary metric to put on data labels
(e.g. share-of-total %), (4) the threshold above which a bar
should be color-flagged (e.g. >20% of total).The Looker Studio quirk: Bar chart in Looker Studio has a "limit records" trap. By default it shows the top 10. If you have 40 campaigns and 30 of them are long-tail, the chart silently hides the long tail. Set the limit explicitly and add a footnote: "Showing top N of M."
3. Stacked bar — composition over time
The question to ask first: How is the mix of a whole changing, and is the total growing or shrinking?
Stacked bars are how you answer "is the pie getting bigger and the slices reshuffling." The common mistake is using a stacked bar to show the mix at a single point in time — that is what a 100% stacked bar or a pie chart is for. Stacked bars are a time series; they need a time axis.
When to use it: A metric that is the sum of two or more subcategories, tracked across time, where the total and the share both matter. Channel mix, device mix, new vs returning, country mix over months.
The Claude prompt:
I'm building a stacked bar chart in Looker Studio. The total is
[TOTAL: e.g. "Total conversions"], broken into [SEGMENTS: e.g. "Paid
Search, Paid Social, Organic, Email, Direct"]. The time window is
[WINDOW: e.g. "last 6 months, monthly grain"]. The decision-maker
wants to see [GOAL: e.g. "whether paid mix is increasing as a share
of total and whether the total itself is growing"].
Return: (1) absolute vs 100% stacked (and which is correct for the
goal), (2) the segment ordering principle (largest on bottom, or
chronological order of appearance), (3) the right number of segments
to show without chart clutter (combine small segments into "Other"
above what threshold?), (4) the comparison period to add as a
secondary bar group.The Looker Studio quirk: Looker Studio will let you stack 12 segments and produce a chart that looks like a candy cane. There is no built-in "Other" rollup. You have to create it in the data source as a calculated field, or pre-aggregate in BigQuery / Sheets. I default to "no more than 5 segments" and roll the rest into Other.
4. Area chart — cumulative volume
The question to ask first: Is the cumulative total growing, and at what pace?
Area charts are visually softer than line charts and work well for cumulative or "running total" views. They are also the most overused chart in pitch decks because they look "big" — the filled area implies magnitude even when the metric is just a normal time series. Be honest with yourself about whether you need an area or whether a line is more honest.
When to use it: Cumulative metrics (running total of signups, cumulative ad spend, MQL — Marketing Qualified Lead, 营销合格线索 — accumulated since Jan 1), or time series where the area under the curve carries meaning (impressions, reach, total ad spend).
The Claude prompt:
I'm building an area chart in Looker Studio. The metric is
[METRIC: e.g. "Cumulative MQLs YTD"]. Source: [SOURCE]. Window:
[WINDOW]. Decision goal: [GOAL: e.g. "show pace against last year's
cumulative curve"].
Return: (1) whether this should be cumulative or non-cumulative,
(2) stacked area vs single area (do I have meaningful sub-
categories?), (3) the time grain that makes the slope readable
(daily = too noisy, monthly = too coarse?), (4) a reference line
to overlay (e.g. last year's curve, target line).The Looker Studio quirk: Looker Studio's area chart is just a line chart with area = true. It does not actually compute "cumulative" — you need a calculated field with a RUNNING_SUM window function. If you skip that, you are drawing an area under a daily value, which is meaningless.
5. Combo chart — two metrics, two scales
The question to ask first: How do two metrics relate when one is a rate and the other is a volume?
Combo is the chart you reach for when one metric is "small" (conversion rate, CTR — Click-Through Rate, 点击率, ROAS) and the other is "big" (sessions, clicks, spend). Plotting them on the same axis flattens the rate to zero. Combo lets you put the volume on bars and the rate on a line, each on its own axis.
When to use it: Volume + rate pairs. Ad spend + ROAS. Sessions + conversion rate. Impressions + CTR. Email sends + open rate. Clicks + cost per click.
The Claude prompt:
I'm building a combo chart in Looker Studio. Bar metric (left axis):
[BAR: e.g. "Ad spend, USD"]. Line metric (right axis): [LINE:
e.g. "ROAS"]. Time grain: [GRAIN]. Window: [WINDOW]. Source:
[SOURCE]. Decision: [DECISION: e.g. "is ROAS holding as we scale
spend"].
Return: (1) the correct axis ranges to use (don't let Looker Studio
auto-scale in a misleading way), (2) whether the line should be
smoothed or step-shaped, (3) the threshold to add as a horizontal
reference line (e.g. break-even ROAS = 1.0, target = 4.0), (4) the
most common misinterpretation of this combo.The Looker Studio quirk: Looker Studio's combo chart has a hidden trap: the right axis is often scaled automatically to the line metric's range, which can make a 1% change in ROAS look like a 50% swing visually. Set the right-axis min/max manually. Always.
6. Pie / Donut — share of total
The question to ask first: Out of the whole, which slice is biggest — and is that answer surprising?
Pie charts get a bad rap, mostly deserved. They are accurate for one thing: showing the share of a whole at a single point in time, when there are 2–5 categories and one slice clearly dominates. Anything more than 5 slices becomes unreadable. Anything that changes over time should be a stacked bar, not a pie.
When to use it: Share-of-total at a snapshot. Channel mix today. Revenue by product line this quarter. Traffic by country this month. Always limited to ≤5 categories and a clear leader.
The Claude prompt:
I'm building a donut chart in Looker Studio. Total: [TOTAL: e.g.
"Total revenue Q1"]. Slices: [CATEGORIES: e.g. "Product A, B, C,
D, E"]. Decision: [DECISION: e.g. "is revenue concentrated in a
small number of products"]. Window: [WINDOW].
Return: (1) whether donut or pie is more appropriate (donut = more
labels, pie = cleaner shape), (2) the slice ordering (clockwise
from 12, largest first?), (3) the threshold to label on-slice vs
in-legend, (4) the "explode" rule (do I pull out the leading
slice?), (5) one alternative chart that would answer the same
question better, and whether to use it instead.The Looker Studio quirk: Looker Studio donut charts can show percentages or raw values, not both. Pick one. If you need both, you want a horizontal bar chart with share % labels — it is honestly more readable than any pie.
7. Funnel — sequential drop-off
The question to ask first: Where in a sequence are we losing the most people (or money)?
Funnels are the one chart in Looker Studio that almost forces a useful question. The shape itself surfaces the worst stage. You have to actively ignore the funnel to misuse it.
When to use it: Any ordered sequence with a measurable entry count and exit count per stage. Conversion funnels (impression → click → lead → MQL → SQL — Sales Qualified Lead, 销售合格线索 — → opportunity → customer). Checkout funnels. Onboarding sequences. The order must be fixed and known.
The Claude prompt:
I'm building a funnel chart in Looker Studio. Stages in order:
[STAGES: e.g. "Session, Add to Cart, Begin Checkout, Purchase"].
Source: [SOURCE: e.g. "GA4 events"]. Window: [WINDOW].
Decision: [DECISION: e.g. "where is the biggest drop-off in our
checkout"].
Return: (1) the right funnel type (open funnel = show absolute
counts, closed funnel = show % of top), (2) the comparison to add
(previous period side-by-side?), (3) the drop-off threshold above
which a stage should be highlighted red, (4) whether to show
time-to-convert between stages, (5) the one stage where funnel
charts routinely lie (e.g. multi-session funnels).The Looker Studio quirk: Looker Studio's funnel chart is a community visualization, not native. The official one is in the "Google Data Studio Report Gallery" templates. Whichever you use, the bigger quirk is data: a single-session funnel is easy; a cross-session funnel in GA4 needs BigQuery export. Don't fake a funnel from page-level data — it lies.
8. Geo map — geographic distribution
The question to ask first: Where geographically is the action happening, and is that where we expected?
Geo maps are visually striking and analytically weak. They show concentration, but rarely explain it. Use them as a first screen, not a deep dive. If a region is dark blue and revenue is light blue there, you have a question to ask. The map is the question, not the answer.
When to use it: Country-, state-, or city-level distribution of any metric, with ≤200 regions. Past that, the map becomes a paint splatter and a ranked table is more useful.
The Claude prompt:
I'm building a geo map in Looker Studio. Region level: [LEVEL: e.g.
"US state"]. Metric: [METRIC: e.g. "Conversions"]. Source: [SOURCE:
e.g. "GA4 with US-state dimension enabled"]. Window: [WINDOW].
Decision: [DECISION: e.g. "where to open the next regional ad
campaign"].
Return: (1) the right map type (filled region = share, bubble =
absolute volume, heat = density), (2) the color scale (sequential
vs diverging — diverging is misleading for non-deviation metrics),
(3) the top N regions to label on the map, (4) the bottom-line
metric to put in a callout next to the map (e.g. "Top 5 states
drive 71% of conversions").The Looker Studio quirk: GA4's geo data is rough at the city level — it is derived from IP and frequently wrong for VPNs, mobile carriers, and corporate networks. For city-level decisions, use a sample of raw BigQuery export, not the GA4 UI numbers. I've seen a "top city" report that was 40% the ISP's headquarters, not the actual customer.
9. Heatmap — two-dimensional density
The question to ask first: Is there a pattern across two categorical dimensions that a line chart cannot show?
Heatmaps in Looker Studio are community visualizations, not native, but they are the best way to show a matrix. The classic use is day-of-week × hour-of-day — when is engagement highest. You can also do campaign × audience, channel × region, etc.
When to use it: Two categorical dimensions and one continuous metric, with a moderate number of cells. Anything bigger than 15×15 becomes unreadable.
The Claude prompt:
I'm building a heatmap in Looker Studio. X dimension: [X: e.g. "Day
of week"]. Y dimension: [Y: e.g. "Hour of day"]. Cell value:
[VALUE: e.g. "Conversion rate"]. Source: [SOURCE]. Window:
[WINDOW]. Decision: [DECISION: e.g. "what hours should we bid up
in Meta Ads"].
Return: (1) whether the data should be raw counts, normalized
ratios, or Z-scores (a statistical measure of how far a value is
from the mean), (2) the color scale (single-hue sequential is
honest; rainbow is not), (3) the cell threshold for labels (label
all? top quartile? outliers only?), (4) the time window over which
to compute averages to avoid one-off spikes looking like patterns.The Looker Studio quirk: Heatmaps are slow in Looker Studio when the underlying query is not aggregated. Pre-aggregate the cell values in BigQuery or Sheets; do not let Looker Studio do row-level rollups for a 168-cell heatmap. It will time out and your readers will never see the chart.
10. Scatter plot — correlation between two metrics
The question to ask first: Do two metrics move together, and is one a leading indicator of the other?
Scatter plots are the most underused chart in marketing dashboards. They are the only chart that answers "is X correlated with Y" without forcing you to make up a derived metric. The problem is that most marketing teams don't naturally think in scatter terms — they think in time series.
When to use it: Two continuous metrics on the same unit (e.g. both per-campaign), and a hypothesis that one drives the other. Spend vs ROAS. Ad rank vs CTR. Email subject length vs open rate. Time on page vs conversion rate.
The Claude prompt:
I'm building a scatter plot in Looker Studio. X metric: [X: e.g.
"Ad spend, USD"]. Y metric: [Y: e.g. "ROAS"]. Point identity:
[ID: e.g. "Campaign name"]. Source: [SOURCE]. Window: [WINDOW].
Decision: [DECISION: e.g. "is there a spend level past which ROAS
collapses"].
Return: (1) the right point-size dimension (a third metric, e.g.
volume, or fixed?), (2) the trendline to overlay (linear, log,
none?) and what the slope tells us, (3) the quadrant to highlight
(high spend + high ROAS = scale; high spend + low ROAS = cut),
(4) the data filter to apply (e.g. only campaigns with >$X spend,
to avoid $50 outliers dominating the visual).The Looker Studio quirk: Looker Studio scatter plots have a tiny maximum point size. If your point-identity is "campaign" and you have 200 campaigns, the chart becomes a black smudge. Set a max-points rule (e.g. top 50 by spend) and turn the rest into a footnote.
11. Bullet chart — actual vs target
The question to ask first: Are we hitting the goal, and how close are we to missing it badly?
Bullet charts are the right chart for KPI (Key Performance Indicator, 关键绩效指标) cards. They show the actual value as a bar, the target as a vertical mark, and a qualitative range (poor / acceptable / good) as background shading. Looker Studio doesn't have a native bullet; you fake it with a stacked horizontal bar plus a reference line. It's clunky, but it is the right idea.
When to use it: Any KPI with a known target and a known acceptable range. Monthly revenue target. CAC (Customer Acquisition Cost, 获客成本) ceiling. Email deliverability rate. NPS (Net Promoter Score, 净推荐值). The target should be a single number, not a curve.
The Claude prompt:
I'm building a bullet chart (stacked bar + reference line) in Looker
Studio. Metric: [METRIC: e.g. "Monthly recurring revenue"]. Target:
[TARGET: e.g. "$120K"]. Acceptable range: [RANGE: e.g. "$80K
(poor), $100K (acceptable), $120K+ (good)"]. Window: [WINDOW].
Decision: [DECISION: e.g. "should we double down on what's working
or course-correct"].
Return: (1) the exact stacked-bar segment values to use as the
background ranges, (2) the reference-line type (line vs marker),
(3) the data-label position (above bar, inside bar, end of bar),
(4) the color rule (red below 80% of target, yellow 80–100%,
green 100%+?), (5) the time comparison to add (last month, same
month last year?).The Looker Studio quirk: The fake-it-with-stacked-bar approach is fragile when the metric is dynamic. If the target changes every month (e.g. "this month's target is 105% of last month"), hardcoding the ranges doesn't work. You need a calculated field. Plan for that before you build.
12. Pivot table — dimensional drill-down
The question to ask first: What is the underlying number, and what dimensions do I need to slice it by?
A pivot table is not a chart. It is a tool. When a chart lies or hides, the pivot table tells the truth. The instinct in Looker Studio is to build a chart, then a summary table, then call it done. The instinct should be the reverse: build the pivot first, see what the numbers actually say, then decide if a chart earns its place on top.
When to use it: Any time the audience might ask "but what about X" within 10 seconds of seeing your summary. Channels × campaigns. Months × channels. Geos × products. Pivot tables are honest, ugly, and indispensable.
The Claude prompt:
I'm designing a pivot table for Looker Studio. Rows: [ROWS: e.g.
"Channel"]. Columns: [COLS: e.g. "Month"]. Metric: [METRIC: e.g.
"Spend"]. Source: [SOURCE]. Window: [WINDOW]. Decision:
[DECISION: e.g. "where is the spend shifting month over month"].
Return: (1) the right number of row and column dimensions (more
than 2 each = unreadable), (2) the totals row / column to add
(grand total, row totals, column totals, none?), (3) the
conditional formatting rules (top quartile green, bottom quartile
red, or compare each cell to a target?), (4) the chart to put
above the pivot to summarize it (one chart, not the pivot itself
as the only view), (5) the drill-down path (what to click
through to get to the raw data).The Looker Studio quirk: Pivot tables do not aggregate across data sources. If you blend GA4 + Google Ads + a CRM, the pivot will only sum within a single source unless you set up a data blend with a join key. Most "broken" pivots in Looker Studio are not broken — they are blending across unjoined sources.
A worked example: rebuilding the CMO's dashboard
To make this concrete, here is how the three most important charts on that rebuilt CMO dashboard actually came together.
The dashboard's job: Show the CMO, in under 60 seconds, whether the marketing engine is on track for the quarter.
Chart 1: Bullet chart for quarterly revenue target. Metric: QTD (Quarter-To-Date, 季度至今) new business revenue. Target: $1.2M. Acceptable range: $800K (poor), $1.0M (acceptable), $1.2M+ (good). I used the bullet prompt above and got back a stacked-bar spec with a reference line at the target. The chart sits at the top of the dashboard. It answers the only question the CMO needs answered first: are we hitting the number?
Chart 2: Combo chart for paid spend vs ROAS. Bar = paid spend by week. Line = ROAS by week. Time grain = weekly, last 12 weeks. The combo prompt surfaced a threshold rule I hadn't thought of: the right axis (ROAS) should be min-clamped at 1.0 to make the break-even line visually obvious. The original dashboard had ROAS floating around 3.5 on a 0-to-9 axis, which made a 1-point drop look harmless.
Chart 3: Stacked bar for channel mix over time. Total = MQLs. Segments = Paid Search, Paid Social, Organic, Email, Direct, Other. 6 months, monthly grain. The stacked-bar prompt pushed back on me: 6 segments was too many, and I should roll the bottom 2 ("Other" and "Direct") into a single "Non-paid" category. The chart now shows the shift toward paid mix over 6 months clearly. The CMO uses this chart to decide whether the team is becoming too paid-dependent.
Five pages, eleven charts total. The other eight are not interesting — they are bullet charts for sub-metrics and a couple of pivot tables for drill-down. The point is that three charts carry the entire dashboard's weight, and each one answers a single decision-anchoring question.
What Claude is actually good for in this workflow
A few honest notes on what Claude does and does not do well here.
Strong: Pressure-testing chart choice, surfacing the question you forgot to ask, generating the dimension/metric specification, and naming the chart's biggest misuse risk. These are all thinking tasks. Claude is a good thinking partner for them.
Mediocre: Generating the actual calculated field syntax. Looker Studio's calculated fields use a specific GoogleSQL-flavored syntax that Claude sometimes gets wrong, especially for window functions and date arithmetic. Always test calculated fields against a known data sample before publishing.
Weak: Generating the dashboard's narrative. Claude will happily write a "key insights" block that sounds executive-ready and is mostly filler. The insights on a real dashboard come from sitting with the data and noticing what is genuinely unusual. I write those myself.
The workflow that has worked for me, in order:
- Open Claude. Paste the chart-prompt template above. Fill in the brackets with the actual decision.
- Read Claude's response as a draft spec, not a final one. Push back if a recommendation feels off.
- Translate the spec into a Looker Studio chart, manually, one chart per session.
- After the chart is built, paste the result back into Claude and ask: "What question would a senior analyst ask about this chart that I haven't answered?"
- Iterate. Each iteration usually takes 5–10 minutes and catches one real issue.
That last step is the most underused. Claude as a critic is more valuable than Claude as a generator. Use it that way.
The dashboard does not save you
One last thing, and it is the thing I keep telling teams who build beautiful dashboards. The dashboard does not save you. The decision does. A mediocre chart that triggers a good decision is better than a perfect chart that triggers a meeting to discuss the chart.
The whole reason to ask the question first is to know whether the chart is needed at all. Some of the best dashboards I have seen have a section at the top with no chart — just three to five sentences stating the current state of the business in plain language. "Q1 is on track to miss by 8% unless paid social ROAS recovers from the May algorithm change." That sentence is the dashboard. Everything below it is the evidence.
If you take one thing from this library, let it be this: before you build a chart, write the sentence the chart is supposed to make obvious. If you cannot write that sentence, the chart is decoration. If you can, hand the sentence to Claude, get the spec, build the chart, and the dashboard will be one a CMO actually opens on Monday morning.