How to Build an AI Visibility Dashboard in Looker Studio: Mentions, Citations, Traffic

Quick answer

To build an AI visibility dashboard, Canlah AI stages three source layers in one BigQuery or Google Sheets table and connects Looker Studio to that table: prompt-level mention and citation data from a monitoring platform, AI referral sessions from Google Analytics 4 and generative AI impressions from Google Search Console. Five of the seven monitoring platforms reviewed on September 23, 2026 publish a Looker Studio route, but one blend accepts at most five tables and an extracted data source stops at 100 MB, so a report that blends each connector directly runs out of room as soon as a sixth layer is added.

Canlah AI holds the measurement layer at one row per observation with 13 fields. The Canlah AI self-audit of canlah.ai run on September 7, 2026 wrote 317 raw evidence files for 233 returned samples across three engine legs, which is the grain a citation panel must read from to survive the question of where a number came from.

Treat the dashboard as a reporting surface rather than a measurement method. A chart cannot repair a prompt panel that changed between cycles, and no blend turns a single run into a frequency. Decide the date grain and freeze the prompt set first, because both become expensive to change once stakeholders have seen version one.

Disclosure: Canlah AI publishes this guide and operates the managed measurement and reporting service described below. Connector, API, export and pricing details for Peec AI, Otterly.ai, Ahrefs Brand Radar, Semrush, Scrunch AI, Profound and AthenaHQ come from public documentation reviewed on September 23, 2026 and were not verified through paid accounts. Google behaviour is quoted from Google help pages checked on the same date.

What an AI visibility dashboard has to reconcile

An AI visibility dashboard has to reconcile five measurement objects (mentions, citations, impressions, sessions and outcomes), each from a different source at a different grain. Collapsing the five into one index hides which of them moved.

• Mention: whether the engine names the brand, in which answers and at what rate across runs. It comes from a monitoring platform or your own probe archive, and it is sampled rather than counted.
• Citation: which URLs and domains the answer linked to, including your own. It comes from the same source at URL grain, which is where vendor exports differ most in shape.
• Impression: how often Google showed your pages inside AI Overviews and AI Mode. Only Search Console publishes it, and only as impressions.
• Session: how many people reached the site from an assistant. GA4 carries it through the AI Assistant channel and the referral source.
• Outcome: what those sessions did. It comes from GA4 key events, joined to the session layer on date and landing page.

Mentions are sampled, impressions are counted by Google and sessions by your tag. The three denominators are unrelated, so a chart plotting them on one axis implies a relationship the data cannot carry.

Which data sources can reach Looker Studio

Five of the seven AI visibility platforms reviewed on September 23, 2026 publish a Looker Studio route (Peec AI, Otterly.ai, Ahrefs Brand Radar, Semrush and Scrunch AI), while Profound and AthenaHQ document an API and file exports instead. GA4, Search Console, server logs staged in BigQuery and the Canlah AI evidence archive complete the 11 routes listed below.

AI visibility data sources and their documented route into Looker Studio, reviewed September 23, 2026.

Source	Route into Looker Studio	What arrives	Main point to verify
Google Analytics 4	Native GA4 connector	Sessions and key events by date, channel, source and landing page	The AI Assistant channel excludes AI Overviews and AI Mode
Google Search Console	Native Search Console connector	Site and URL impression tables by date, query, page and country	Generative AI report in the connector: not stated in the documentation reviewed
Peec AI	Community connector, project-scoped API key	12 dimensions including Prompt, Model, Domain and URL; 8 metrics including Citation, Visibility and Share of Voice	The key is created in the Peec dashboard first
Otterly.ai	Native connector on Standard, Premium and Enterprise	Coverage percentage, mentions, share of voice, average rank and URL-level citations	Whether your plan is one of the three listed
Ahrefs Brand Radar	Looker connector plus a Report Management API	History charts for AI mentions, impressions and AI Share of Voice, by brand and platform	Whether it pulls Ahrefs prompts, your custom prompts or both
Semrush	Looker Studio connector for Domain Analytics, Position Tracking and Site Audit	SEO datasets through the Semrush API without spending API units	AI Visibility Toolkit fields: not stated in the documentation reviewed
Scrunch AI	Looker Studio templates plus Query, Responses and Agent Traffic APIs	Brand presence, position score, sentiment and competitor presence by platform, persona and source URL	Whether your plan includes Responses as well as Query
Profound	REST API, Tableau connector, CSV and JSON exports on Growth; Looker Studio connector not stated in the documentation reviewed	Per-execution prompt and answer rows carrying mentions and citations	A warehouse sits between the API and Looker Studio, because the documented BI connector is Tableau
AthenaHQ	CSV export and API access; Looker Studio connector not stated in the documentation reviewed	Credit-metered responses, 1 credit per AI response, on Starter at US$295 per month	Whether API access sits in your tier or is an add-on
Server and CDN logs	BigQuery, then the native BigQuery connector	Request-level AI crawler hits by user agent, path and status	Log retention, usually shorter than the reporting window
Canlah AI evidence archive	JSON loaded into BigQuery or Google Sheets	One row per observation with 13 fields, plus the raw answer file	It ships with the engagement, not as a connector

Source: vendor and Google documentation reviewed September 23, 2026. “Not stated in the documentation reviewed” means the vendor’s or Google’s public materials did not carry that detail when they were retrieved on that date. It is not proof that the capability is absent, and current scope should be confirmed with the provider directly.

Two platform limits decide the architecture before any chart is drawn. A blend contains at most five tables, and an extracted data source is capped at 100 MB. A mention layer, a citation layer at URL grain, GA4, Search Console and a log table already fill all five slots, which is why Canlah AI stages client data in BigQuery and connects one table. Looker Studio Pro costs US$9 per user per project per month, billed per Google Cloud project, so an agency running one project per client pays per client as well as per seat.

How to build the dashboard in six steps

An AI visibility dashboard takes six steps to build in Looker Studio, and four of the six are schema decisions that cannot be made after the charts exist.

Step 1: Fix the grain before opening Looker Studio

Write down the unit of one row in each layer: one observation for mentions, one citation for URLs, one day for GA4 and Search Console. Every later join is date plus one key, and an undeclared grain gets double counted the first time someone adds a filter.

Step 2: Land every source in one table before connecting

Load the platform export and the GA4 and Search Console extracts into BigQuery or Google Sheets, reconcile the date column, then connect Looker Studio once. That avoids the five-table blend ceiling and the 100 MB extract cap, and leaves one place to fix a vendor field rename.

Step 3: Repair the GA4 channel before charting it

Confirm which assistants your property classifies as AI Assistant, then build a custom channel group for the rest. A standard property allows 2 custom channel groups of up to 50 channels each, and they apply retroactively, which the default group does not.

Step 4: Read Search Console as impressions, never as clicks

Chart the generative AI performance report as an impression trend only. Google states it shows how many times links to your site were shown in a generative AI feature, broken down by page, country, device and date, with no clicks, no click-through rate and no query data.

Step 5: Build three panels, one per question

Panel one answers whether engines name the brand, panel two which sources they cite, panel three what reached the site. Give each a date control and a footnote naming the source, the engine set and the sample size.

Step 6: Stamp the report with the panel and the freeze date

Put the prompt count, the runs per prompt, the engine list and the freeze date into the report header as text. Canlah AI prints those four values on every client report, because a dashboard without them cannot be audited.

The row your mention table needs

The mention table needs one row per observation with 13 fields, and each field exists because the dashboard breaks without it. The schema below is what Canlah AI writes for every probe, taken from the canlah.ai self-audit of September 7, 2026.

One observation row from the Canlah AI evidence archive, self-audit of September 7, 2026, and what each field supports.

Field	Example value	What the dashboard cannot do without it
engine	brightdata_aio	Split any metric by engine, the first cut stakeholders ask for
query	Verbatim buyer question	Trace a score back to the question behind it
layer_id	L2	Report funnel coverage instead of one blended mention rate
run	1	Aggregate repeated runs into a range rather than a point
gl	sg	Separate markets before comparing them
hl	en	Separate languages, which retrieve different sources
ts	1788749300172	Order observations and detect a collection gap
status	ok	Exclude failed samples from the denominator
error	null	Explain a dip caused by collection, not visibility
attempts	1	Distinguish a retried sample from a clean one
elapsed_s	4.1	Spot a throttled leg before it reads as a drop
ai_overview	Answer payload	Recount mentions and citations without re-probing
serp_keys	organic, related_questions	Show what else occupied the result page

Source: Canlah AI self-audit of canlah.ai, September 7, 2026, one returned sample shown. The run is internal and was not independently audited.

The self-audit produced 317 raw evidence files for a planned 234 samples, of which 233 returned and one failed on a read timeout. That is why status and error are not optional: a dashboard using 234 as its denominator reports a rate that is quietly wrong.

How to wire GA4 so AI sessions are not double counted

To keep AI sessions from being double counted, wire GA4 on two channel groupings and chart one grouping per panel: the default AI Assistant channel as the headline and one retroactive custom channel group as the audit trail. Google defines the AI Assistant channel as the channel by which users arrive from sources like ChatGPT, Gemini, DeepSeek, Copilot or Grok, and it explicitly excludes Google’s AI Overviews and AI Mode. The matching rule is the medium ai-assistant, and Google states that when the referrer matches its list the medium is set to ai-assistant and the campaign to (ai-assistant).

Three consequences follow. Clicks from AI Overviews and AI Mode stay inside Organic Search and cannot be separated in GA4, which leaves Search Console impressions as the only Google-published signal for those surfaces. Google names five assistants as examples without publishing the full matching list, so any assistant outside those five should be confirmed in your own source and medium report first. Assistants that send visitors without a referrer land in Direct, which sets a floor under how much AI traffic any build can attribute.

Every session is classified once in the default group and once again in each custom group, so adding the two groupings in one chart counts the same AI session twice.

What monthly AEO reporting should show

Monthly AEO reporting should show four numbers with four different denominators, each labelled with its denominator. A single blended visibility score fails that test, because it moves and nobody can say which of the four inputs moved.

1. Mention rate per funnel layer, expressed as a range across runs, with the question count in the label.
2. Citation share by domain, counting your own and the third-party sources that outrank it in the same answers.
3. Generative AI impressions from Search Console, plotted alone and never divided.
4. Sessions and key events from the AI Assistant channel and the custom channel group, with Direct called out as a known floor.

Canlah AI reports citation share as a range with confidence intervals rather than one figure, because AI answers fluctuate month to month and a single number invites a comparison the sampling cannot support.

How to interpret five common dashboard results

Five results recur often enough on an AI visibility dashboard to interpret in advance, and only one of the five, a step change on the day a connector changed, points to the pipeline rather than to visibility.

Mentions up, sessions flat: This is the normal shape, not a fault, because being named inside an answer does not generate a click. Check citation share first: a mention with no link to your domain produces exactly this chart.

Sessions up, mentions flat: Inspect the landing pages first. Assistant referrals often concentrate on one or two pages cited for a narrow question, which makes the gain page-level rather than brand-level.

Citation share concentrated on a single URL: One page is carrying the brand inside answers. Check whether it is also the only page Search Console shows in the generative AI report. A single-page dependency is a risk to write into the report.

A step change on the day a connector changed: Suspect the pipeline before the market. A renamed vendor field that silently becomes null reads as a collapse.

Search Console impressions with no matching referral: Google is showing the page inside an AI feature and users are not clicking through. Measure it against the mention layer, not the session layer.

What a Looker Studio dashboard cannot prove

A Looker Studio dashboard cannot prove four things about AI visibility, however clean the build.

• It cannot prove a stable share of voice, because the mention layer samples a distribution while the chart draws a line between the samples.
• It cannot prove that a content change caused an answer change, because nothing in the build controls for model or index updates.
• It cannot recover AI traffic that arrived without a referrer, which stays in Direct whatever the channel configuration.
• It cannot report clicks from AI Overviews or AI Mode separately, because Search Console publishes impressions only and GA4 leaves those clicks in Organic Search.

The Canlah AI evidence archive carries the same limit as every mention layer here. The September 7, 2026 self-audit was a single-day baseline for one brand in one market, run in a compressed window with no spacing between same-intent samples, so intraday variance is not represented.

When to move from a self-built dashboard to a managed report

A self-built AI visibility dashboard is sufficient while one person owns the schema, one market is tracked and each extract stays under the 100 MB cap. It stops being sufficient when the volume outgrows Google Sheets, when a second market doubles the layers or when a vendor schema change goes unnoticed for a cycle.

Canlah AI is a Singapore-based SEO + GEO agency that measures and improves brand visibility inside ChatGPT, Gemini, Google AI Overviews and Google AI Mode, reporting AI visibility as re-verifiable ranges with timestamped evidence. Engagements lock a contract query universe of 4, 8 or 16 buyer intents by tier for 90 days, keep the prompt, answer, screenshot and timestamp for every probe, and hand the archive to the client so it can be re-run.

The limitation is that Canlah AI does not sell a Looker Studio connector. Reporting arrives as a monthly report plus the archive, so a team that wants a live self-serve dashboard should buy one of the platforms above and keep the archive as its audit layer.

Do not pick a connector from the table above alone. Confirm plan scope, field names and export limits with each vendor in writing before building, and do not accept any AI visibility chart, from a vendor or from your own build, without the panel size, the engine list and the freeze date printed next to it.

Frequently asked questions

How do I build an AI visibility dashboard in Looker Studio?

To build an AI visibility dashboard in Looker Studio, export mention and citation data from your monitoring platform, extract GA4 and Search Console into the same store and reconcile the date column in BigQuery or Google Sheets. Then connect Looker Studio to that one table. A report built on five separate connectors hits the five-table blend ceiling as soon as a log layer is added.

Can Looker Studio connect directly to ChatGPT or Perplexity?

No. Looker Studio reads from connectors and databases, not from assistants, so the mention layer arrives through a monitoring platform’s connector or API, or through a probe archive such as the one Canlah AI delivers. Five of the seven platforms reviewed here publish a Looker Studio route; the other two document an API and exports that need a warehouse in between.

Which AI visibility tools have a Looker Studio connector?

Peec AI, Otterly.ai, Ahrefs Brand Radar, Semrush and Scrunch AI all document a Looker Studio route as of September 23, 2026, with Otterly.ai listing it on Standard, Premium and Enterprise and Peec AI requiring a project-scoped API key. The Semrush connector documents SEO datasets, and AI Visibility Toolkit fields were not stated in the documentation reviewed. Profound documents a Tableau connector, a REST API and exports on Growth, so confirm current scope with each vendor directly.

How do I track AI traffic in GA4?

To track AI traffic in GA4, use the AI Assistant channel in the default channel group, then add a custom channel group with explicit hostname rules for anything it misses. Google defines the channel by the medium ai-assistant and names ChatGPT, Gemini, DeepSeek, Copilot and Grok as examples, while excluding AI Overviews and AI Mode. Custom channel groups apply retroactively, which repairs the months before the AI Assistant channel existed.

Why do mentions and sessions never match in an AEO report?

Mentions and sessions never match in an AEO report because they count different things. Mentions are sampled from repeated probes and sessions are counted by your tag, while many assistant referrals arrive with no referrer and land in Direct. Canlah AI reports both with their own denominators and never divides one by the other.

Does Canlah AI offer a Looker Studio connector?

No. Canlah AI delivers a monthly report plus the evidence archive as JSON, one row per observation with 13 fields, which a team loads into BigQuery or Google Sheets and connects to Looker Studio itself. A team that wants a live self-serve dashboard should pair that archive with one of the five platforms that publish a Looker Studio route.

Method and sources

Connector, API and export details were taken from each vendor’s public documentation on September 23, 2026 and record what those pages stated on that date. No paid account was used, and no vendor reviewed this article before publication. Google behaviour was quoted from Google help pages checked on the same date. The observation schema and sample counts come from a Canlah AI self-audit of canlah.ai executed on September 7, 2026, an internal run on one brand in one market, not independently audited.

• GA4 default channel group. AI Assistant definition, the medium rule and the AI Overviews exclusion.
• GA4 custom channel groups. The 2-group and 50-channel limits, and retroactive application.
• Generative AI performance report. Impressions-only scope, the breakdowns and the 1,000-row limit.
• Looker Studio blends. Blend construction and joins.
• Peec AI Looker Studio connector. The 12 dimensions, 8 metrics and API key requirement.
• Otterly.ai Looker Studio connector. Plan availability and the citation fields.
• Ahrefs Looker Studio documentation. Brand Radar history charts and prompt selection.
• Semrush Looker Studio integration. Datasets the connector carries.
• Scrunch developer documentation. The endpoints and the Looker Studio templates.
• Profound REST API. Per-execution prompt and answer rows.
• AthenaHQ plans. Starter price, credit allowance and export scope.
• Canlah AI self-audit of canlah.ai, September 7, 2026, internal and not independently audited. The 13-field schema, 317 evidence files, 234 planned and 233 returned samples.
• Canlah AI AI-info page. Contract query universe, archive contents and citation share reported as a range.

See which answers name your brand before you build the chart. Run the free AI visibility audit →

Related articles

• How to choose AI search prompts to track
• How to fix ChatGPT 404 referrals
• AI visibility tool pricing in 2026
• What are the best AI search monitoring tools?